term_project.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Kaidavid/data_mining/blob/main/term_project.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lTvx5gsbV7kn"
      },
      "source": [
        "# **GPU and RAM**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "NUED00mlWAX5",
        "outputId": "75e37810-d5da-4777-88dc-c2ad565b1d13"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Wed Jun 15 13:53:01 2022       \n",
            "+-----------------------------------------------------------------------------+\n",
            "| NVIDIA-SMI 460.32.03    Driver Version: 460.32.03    CUDA Version: 11.2     |\n",
            "|-------------------------------+----------------------+----------------------+\n",
            "| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\n",
            "| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\n",
            "|                               |                      |               MIG M. |\n",
            "|===============================+======================+======================|\n",
            "|   0  Tesla T4            Off  | 00000000:00:04.0 Off |                    0 |\n",
            "| N/A   52C    P8    10W /  70W |      0MiB / 15109MiB |      0%      Default |\n",
            "|                               |                      |                  N/A |\n",
            "+-------------------------------+----------------------+----------------------+\n",
            "                                                                               \n",
            "+-----------------------------------------------------------------------------+\n",
            "| Processes:                                                                  |\n",
            "|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |\n",
            "|        ID   ID                                                   Usage      |\n",
            "|=============================================================================|\n",
            "|  No running processes found                                                 |\n",
            "+-----------------------------------------------------------------------------+\n",
            "Your runtime has 27.3 gigabytes of available RAM\n",
            "\n",
            "You are using a high-RAM runtime!\n"
          ]
        }
      ],
      "source": [
        "from psutil import virtual_memory\n",
        "\n",
        "gpu_info = !nvidia-smi\n",
        "gpu_info = '\\n'.join(gpu_info)\n",
        "if gpu_info.find('failed') >= 0:\n",
        "  print('Not connected to a GPU')\n",
        "else:\n",
        "  print(gpu_info)\n",
        "\n",
        "\n",
        "ram_gb = virtual_memory().total / 1e9\n",
        "print('Your runtime has {:.1f} gigabytes of available RAM\\n'.format(ram_gb))\n",
        "\n",
        "if ram_gb < 20:\n",
        "  print('Not using a high-RAM runtime')\n",
        "else:\n",
        "  print('You are using a high-RAM runtime!')"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RoGvPxGPsJ1Y"
      },
      "source": [
        "# **Libraries**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "_EvqD_kTSnaa"
      },
      "outputs": [],
      "source": [
        "import os\n",
        "import random\n",
        "import time\n",
        "import copy\n",
        "import numpy as np\n",
        "import pandas as pd\n",
        "import networkx as nx\n",
        "import matplotlib.pyplot as plt\n",
        "from datetime import timedelta\n",
        "\n",
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "from torch.autograd import Variable\n",
        "\n",
        "from tensorflow import keras\n",
        "from sklearn.manifold import TSNE\n",
        "from sklearn.decomposition import PCA\n",
        "from sklearn.preprocessing import StandardScaler\n",
        "from sklearn.ensemble import RandomForestClassifier\n",
        "from sklearn.metrics import roc_auc_score, f1_score, accuracy_score\n",
        "from gensim.models import Word2Vec"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Mount G-Drive**"
      ],
      "metadata": {
        "id": "KTECLf2mHy4U"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "xp7XdgdZBQkg",
        "outputId": "3549b481-7de7-4847-e316-8573d2072606"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Mounted at /gdrive\n",
            "'8m 21s node_embedding1_connected.csv'\t term_project.ipynb\n",
            " description.pdf\t\t\t test_classification_question.csv\n",
            " for_graph.csv\t\t\t\t test_completion_question.csv\n",
            " models\t\t\t\t\t train.csv\n",
            " node_ingredient.csv\t\t\t validation_classification_answer.csv\n",
            "'Presentation Video.mp4'\t\t validation_classification_question.csv\n",
            " report_template\t\t\t validation_completion_answer.csv\n",
            " report_template.zip\t\t\t validation_completion_question.csv\n",
            " results\n"
          ]
        }
      ],
      "source": [
        "from google.colab import drive\n",
        "\n",
        "drive.mount('/gdrive')\n",
        "# Specify the directory path where `term_project.ipynb` exists.\n",
        "# then set root = '/gdrive/My Drive/Data Mining/project'\n",
        "\n",
        "root = '/gdrive/My Drive/Data Mining/project'\n",
        "### Change the working directory to root ###\n",
        "os.chdir(root)\n",
        "!ls"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7SKlRDxPsZRq"
      },
      "source": [
        "# **Installation**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "FrxnBWlLo7-r"
      },
      "outputs": [],
      "source": [
        "# !pip install torchtext==0.11.0\n",
        "# !pip install torch==1.10.0+cu111 torchvision==0.11.0+cu111 torchaudio==0.10.0 -f https://download.pytorch.org/whl/torch_stable.html"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "ab6NmYXSnXow"
      },
      "outputs": [],
      "source": [
        "# def format_pytorch_version(version):\n",
        "#     return version.split('+')[0]\n",
        "\n",
        "# def format_cuda_version(version):\n",
        "#     return 'cu' + version.replace('.', '')\n",
        "\n",
        "# TORCH_version = torch.__version__\n",
        "# TORCH = format_pytorch_version(TORCH_version)\n",
        "# CUDA_version = torch.version.cuda\n",
        "# CUDA = format_cuda_version(CUDA_version)\n",
        "\n",
        "# !pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-{TORCH}+{CUDA}.html\n",
        "# !pip install torch-sparse -f https://pytorch-geometric.com/whl/torch-{TORCH}+{CUDA}.html\n",
        "# !pip install torch-cluster -f https://pytorch-geometric.com/whl/torch-{TORCH}+{CUDA}.html\n",
        "# !pip install torch-spline-conv -f https://pytorch-geometric.com/whl/torch-{TORCH}+{CUDA}.html\n",
        "# !pip install torch-geometric\n",
        "# !pip install torch_geometric"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "hVK7X8WYsdrd"
      },
      "outputs": [],
      "source": [
        "# import torch_geometric\n",
        "# import torch_geometric.transforms as T\n",
        "# from torch_geometric.utils import to_networkx\n",
        "# from torch_geometric.data import Data"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "-_7I9HO9lUeb",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a293b3e0-bdbb-4801-e93b-5eb2b7d97943"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\u001b[K     |████████████████████████████████| 433 kB 39.2 MB/s \n",
            "\u001b[K     |████████████████████████████████| 482 kB 88.8 MB/s \n",
            "\u001b[K     |████████████████████████████████| 41 kB 589 kB/s \n",
            "\u001b[?25h  Building wheel for mplleaflet (setup.py) ... \u001b[?25l\u001b[?25hdone\n"
          ]
        }
      ],
      "source": [
        "!pip install -q stellargraph[demos]==1.2.0"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "6OlMsUI9lY_7"
      },
      "outputs": [],
      "source": [
        "import stellargraph as sg\n",
        "from stellargraph.data import BiasedRandomWalk\n",
        "from stellargraph.data import UnsupervisedSampler"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Helper Functions**"
      ],
      "metadata": {
        "id": "a-gftenuH2Xe"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Mb4cp8g9RA-1"
      },
      "outputs": [],
      "source": [
        "def get_ingredient_name(ing_ids):\n",
        "  if ing_ids == None:\n",
        "    return None\n",
        "  \n",
        "  with open (\"node_ingredient.csv\", \"r\") as f:\n",
        "    lines = f.readlines()\n",
        "    if type(ing_ids) == list:\n",
        "      # make sure they are numeric values\n",
        "      for id in ing_ids:\n",
        "        if not id.isnumeric():\n",
        "          return\n",
        "      return [lines[int(x)].strip() for x in ing_ids]\n",
        "    else:\n",
        "      if not ing_ids.isnumeric():\n",
        "          return\n",
        "      return lines[int(ing_ids)].strip()\n",
        "\n",
        "\n",
        "def get_unused_ingredient(arr):\n",
        "  unused = np.nonzero(np.sum(arr, axis=1)== 0)[0].tolist()\n",
        "  unused = [str(idx) for idx in unused]\n",
        "  return get_ingredient_name(unused)\n",
        "\n",
        "\n",
        "def data_info(path, index):\n",
        "  with open (path, \"r\") as f:\n",
        "    data = f.readlines()\n",
        "    if index >= len(data):\n",
        "      return \"Index out of bound\"\n",
        "    value = data[index].strip().split(\",\")\n",
        "    label = None\n",
        "    missing = None\n",
        "    \n",
        "    if path == \"train.csv\":\n",
        "      label = value.pop()\n",
        "\n",
        "    elif path == \"validation_classification_question.csv\":\n",
        "      label = \"Found in validation_classification_answer.csv\"\n",
        "    elif path == \"validation_classification_answer.csv\":\n",
        "      label = value[0]\n",
        "      value = \"Found in validation_classification_question.csv\"\n",
        "    elif path == \"test_classification_question.csv\":\n",
        "      label = \"?????\"\n",
        "\n",
        "    elif path == \"validation_completion_question.csv\":\n",
        "      missing = \"Found in validation_completion_answer.csv\"\n",
        "    elif path == \"validation_completion_answer.csv\":\n",
        "      missing = value[0]\n",
        "      value = \"Found in validation_completion_question.csv\"\n",
        "    elif path == \"test_completion_question.csv\":\n",
        "      missing = \"?????\"\n",
        "    else:\n",
        "      return \"Wrong input path\"\n",
        "    \n",
        "    if type(value) == list and type(value[0]) == int:\n",
        "      value = [int(x) for x in value]\n",
        "    print(f\"Recipe Number: {index}/{len(data)} from {path}\")\n",
        "    print(f\"Recipe Origin: {label}\")\n",
        "    print(f\"Missing ingredient: ID = {missing}, Name = {get_ingredient_name(missing)}\")\n",
        "    print(f\"Ingredients ID: {value}\")\n",
        "    print(f\"Ingredients Name: {get_ingredient_name(value)}\")\n",
        "\n",
        "\n",
        "def node_feature_pair(node_features):\n",
        "  pair = {}\n",
        "  for i in range(node_features.shape[0]):\n",
        "    pair[i] = node_features[i].tolist()\n",
        "  return pair\n",
        "\n",
        "\n",
        "def connected_graph(arr):\n",
        "  edges = []\n",
        "  for i in range(len(arr) - 1):\n",
        "    edges.append((arr[i], arr[i+1]))\n",
        "  edges.append((arr[-1], arr[0]))\n",
        "  return edges"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "data_info(\"validation_completion_question.csv\", 100)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "JdvqSOQcClUI",
        "outputId": "0f6a230f-f78d-4eb4-aab5-9c301bb51664"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Recipe Number: 100/7848 from validation_completion_question.csv\n",
            "Recipe Origin: None\n",
            "Missing ingredient: ID = Found in validation_completion_answer.csv, Name = None\n",
            "Ingredients ID: ['167', '281', '791', '909', '1308', '1953', '2651', '2945', '5379', '5418', '6187']\n",
            "Ingredients Name: ['vegetable oil', 'flour', 'flank steak', 'baking soda', 'sugar', 'corn starch', 'brown sugar', 'water', 'broccoli', 'low sodium soy sauce', 'garlic']\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "D7VcttYAaNAD"
      },
      "source": [
        "# **Node Featurizing**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "2yPsEYKSXE3R"
      },
      "outputs": [],
      "source": [
        "from easydict import EasyDict\n",
        "from pathlib import Path\n",
        "\n",
        "args = EasyDict()\n",
        "args.dim = 128\n",
        "args.walk_number = 10   # number of random walks per root node\n",
        "args.walk_length = 30    # maximum length of a random walk 50\n",
        "\n",
        "args.batch_size = 64\n",
        "args.lr = 1e-3\n",
        "args.momentum = 0.9\n",
        "args.weight_decay = 5e-4\n",
        "args.epoch = 4\n",
        "args.epoch_ = 100\n",
        "args.random = 30\n",
        "args.graph_method = \"complete\"\n",
        "args.tensorboard = True\n",
        "args.gpu = True\n",
        "args.b1 = 0.5\n",
        "args.b2 = 0.999\n",
        "args.pred_dim = 128\n",
        "args.pred_epochs = 50\n",
        "args.pred_lr = 1e-5\n",
        "args.latent_dim = 100\n",
        "\n",
        "device = 'cuda' if torch.cuda.is_available() and args.gpu else 'cpu'\n",
        "\n",
        "results = Path(root) / 'results'\n",
        "results.mkdir(parents=True, exist_ok=True)\n",
        "\n",
        "models_path = Path(root) / 'models'\n",
        "models_path.mkdir(parents=True, exist_ok=True)\n",
        "\n",
        "labels = {\n",
        "    'brazilian': 0, 'british': 1, 'cajun_creole': 2, 'chinese': 3, 'filipino': 4, 'french': 5, 'greek': 6, 'indian': 7, 'irish': 8, 'italian': 9, 'jamaican': 10,\n",
        "    'japanese': 11, 'korean': 12, 'mexican': 13, 'moroccan': 14, 'russian': 15, 'southern_us': 16, 'spanish': 17, 'thai': 18, 'vietnamese': 19\n",
        "}"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def create_graph(labels, graph_method, randomize=False):\n",
        "  G = nx.Graph()\n",
        "  \n",
        "  #   Create node feature array\n",
        "  with open (\"node_ingredient.csv\", \"r\") as f:\n",
        "    lines = f.readlines()\n",
        "    node_feature = np.zeros((len(lines), len(labels)))\n",
        "  \n",
        "  #   Create graph from training data\n",
        "  with open (\"train.csv\", \"r\") as f:\n",
        "    lines = f.readlines()\n",
        "    random_recipes = sorted(random.sample(range(0, len(lines)), args.random)) if randomize else [i for i in range(len(lines))]\n",
        "    if graph_method == \"connected\":\n",
        "      for num in random_recipes:\n",
        "        recipe = lines[num].strip().split(\",\")\n",
        "        label = recipe.pop()\n",
        "        recipe = [int(x) for x in recipe]\n",
        "        \n",
        "        #   Create a complete graph between the ingredients in one cuisine and add them to the main graph G\n",
        "        G.add_edges_from(connected_graph(recipe))\n",
        "\n",
        "        #   Each node (ingredient) will be featurized by the countries (labels) its used at\n",
        "        for ingredient in recipe:\n",
        "          node_feature[ingredient, labels[label]] += 1\n",
        "    elif graph_method == \"complete\":\n",
        "      for num in random_recipes:\n",
        "        recipe = lines[num].strip().split(\",\")\n",
        "        label = recipe.pop()\n",
        "        recipe = [int(x) for x in recipe]\n",
        "        \n",
        "        #   Create a complete graph between the ingredients in one cuisine and add them to the main graph G\n",
        "        G.add_edges_from(list(nx.complete_graph(recipe).edges))\n",
        "\n",
        "        #   Each node (ingredient) will be featurized by the countries (labels) its used at\n",
        "        for ingredient in recipe:\n",
        "          node_feature[ingredient, labels[label]] += 1\n",
        "  \n",
        "  #   Include validation data in the graph\n",
        "  with open (\"validation_classification_answer.csv\", \"r\") as f:\n",
        "    lines = f.readlines()\n",
        "    val_clf_answer = []\n",
        "    for line in lines:\n",
        "      val_clf_answer.append(line.strip())\n",
        "\n",
        "  with open (\"validation_classification_question.csv\", \"r\") as f:\n",
        "    lines = f.readlines()\n",
        "    random_recipes = sorted(random.sample(range(0, len(lines)), args.random//2)) if randomize else [i for i in range(len(lines))]\n",
        "    if graph_method == \"connected\":\n",
        "      for num in random_recipes:\n",
        "        recipe = lines[num].strip().split(\",\")\n",
        "        recipe = [int(x) for x in recipe]\n",
        "        label = val_clf_answer[num]\n",
        "        \n",
        "        #   Create a complete graph between the ingredients in one cuisine and add them to the main graph G\n",
        "        G.add_edges_from(connected_graph(recipe))\n",
        "\n",
        "        #   Each node (ingredient) will be featurized by the countries (labels) its used at\n",
        "        for ingredient in recipe:\n",
        "          node_feature[ingredient, labels[label]] += 1\n",
        "\n",
        "    elif graph_method == \"complete\":\n",
        "      for num in random_recipes:\n",
        "        recipe = lines[num].strip().split(\",\")\n",
        "        recipe = [int(x) for x in recipe]\n",
        "        label = val_clf_answer[num]\n",
        "        \n",
        "        #   Create a complete graph between the ingredients in one cuisine and add them to the main graph G\n",
        "        G.add_edges_from(list(nx.complete_graph(recipe).edges))\n",
        "\n",
        "        #   Each node (ingredient) will be featurized by the countries (labels) its used at\n",
        "        for ingredient in recipe:\n",
        "          node_feature[ingredient, labels[label]] += 1\n",
        "  \n",
        "  #   Include unused ingredients as nodes with 0 degrees\n",
        "  unused = np.nonzero(np.sum(node_feature, axis=1) == 0)[0].tolist()\n",
        "  G.add_nodes_from(unused)\n",
        "\n",
        "  return G, node_feature"
      ],
      "metadata": {
        "id": "7nk7hjjYGqAH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "G, node_features = create_graph(labels, args.graph_method)"
      ],
      "metadata": {
        "id": "P5Q3xeTvCmRL"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "1fODnPzU9ylP"
      },
      "outputs": [],
      "source": [
        "scaler = StandardScaler().fit(node_features)\n",
        "scaled_node_features = scaler.transform(node_features)\n",
        "df = pd.DataFrame(scaled_node_features, columns = labels.keys())"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(f\"Number of Node: \", G.number_of_nodes())\n",
        "print(f\"Number of Edges: \", G.number_of_edges())\n",
        "print(f\"Top10 Nodes with highest degree (Ingredient, Dishes): \", sorted(G.degree, key=lambda x: x[1], reverse=True)[:10])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "hxFcNQObECv2",
        "outputId": "11127748-ae98-427a-efb3-a63bacea2c42"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Number of Node:  6714\n",
            "Number of Edges:  421566\n",
            "Top10 Nodes with highest degree (Ingredient, Dishes):  [(937, 4537), (2945, 3593), (5536, 3386), (6187, 3285), (1308, 3186), (5377, 3182), (2122, 2984), (2518, 2809), (167, 2784), (5648, 2716)]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XBxioxaEO_dp",
        "outputId": "ece89516-ef25-4f52-9272-135ff31aac65"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Most used ingredient of each country (Dishes, Ingredient, Country):  [(152.0, 'salt', 'brazilian'), (329.0, 'salt', 'british'), (580.0, 'salt', 'cajun_creole'), (1083.0, 'soy sauce', 'chinese'), (338.0, 'salt', 'filipino'), (960.0, 'salt', 'french'), (446.0, 'salt', 'greek'), (1514.0, 'salt', 'indian'), (297.0, 'salt', 'irish'), (2741.0, 'salt', 'italian'), (252.0, 'salt', 'jamaican'), (438.0, 'soy sauce', 'japanese'), (317.0, 'soy sauce', 'korean'), (2155.0, 'salt', 'mexican'), (331.0, 'olive oil', 'moroccan'), (233.0, 'salt', 'russian'), (1773.0, 'salt', 'southern_us'), (360.0, 'salt', 'spanish'), (480.0, 'fish sauce', 'thai'), (321.0, 'fish sauce', 'vietnamese')]\n"
          ]
        }
      ],
      "source": [
        "x = list(zip(np.max(node_features, axis=0), get_ingredient_name([str(idx) for idx in list(np.argmax(node_features, axis=0))]), labels.keys()))\n",
        "print(f\"Most used ingredient of each country (Dishes, Ingredient, Country): \", x)"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "RdDJ9JT6-s3Q",
        "outputId": "c112b5d2-b565-4493-afca-cc26f44c35c3"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Out of 6714 ingredients 410 ingredients are not used in any of the given recipes\n"
          ]
        }
      ],
      "source": [
        "unused = get_unused_ingredient(node_features)\n",
        "print(f\"Out of {node_features.shape[0]} ingredients {len(unused)} ingredients are not used in any of the given recipes\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "6B-KCIrfpnS_",
        "outputId": "ff0c4597-995b-438e-c4dd-57e3ab71d50b"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "StellarGraph: Undirected multigraph\n",
            " Nodes: 6714, Edges: 421566\n",
            "\n",
            " Node types:\n",
            "  default: [6714]\n",
            "    Features: float32 vector, length 20\n",
            "    Edge types: default-default->default\n",
            "\n",
            " Edge types:\n",
            "    default-default->default: [421566]\n",
            "        Weights: all 1 (default)\n",
            "        Features: none\n"
          ]
        }
      ],
      "source": [
        "pairs = node_feature_pair(node_features)\n",
        "G_w_node_feature = sg.StellarGraph.from_networkx(G, node_features=zip(pairs.keys(), pairs.values()))\n",
        "print(G_w_node_feature.info())\n",
        "\n",
        "# G_w_node_feature = sg.StellarGraph.from_networkx(G, node_features=zip([i for i in range(0, node_features.shape[0])], node_features.tolist())) \n",
        "# print(G_w_node_feature.info())\n",
        "\n",
        "\n",
        "# G_wo_node_feature = sg.StellarGraph.from_networkx(G)\n",
        "# print(G_wo_node_feature.info())\n",
        "\n",
        "# Make sure every ingredient is represented as a node in the graph\n",
        "assert(list(pairs.keys()) == [x for x in range(node_features.shape[0])])"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "G_degrees = sorted(G.degree, key=lambda x: x[1], reverse=True)[:20]"
      ],
      "metadata": {
        "id": "VSmxOXiUi4w5"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "P, node_features_ = create_graph(labels, \"connected\")"
      ],
      "metadata": {
        "id": "wEeWki-Qi8hg"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "P_degrees = sorted(P.degree, key=lambda x: x[1], reverse=True)[:20]"
      ],
      "metadata": {
        "id": "6KpQPpZ2jUAf"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "nodes_G = [str(i) for i,j in G_degrees]\n",
        "nodes_P = [str(i) for i,j in P_degrees]\n",
        "degree_G = [j for i,j in G_degrees]\n",
        "degree_P = [j for i,j in P_degrees]\n",
        "\n",
        "nodes_G = get_ingredient_name(nodes_G)\n",
        "nodes_P = get_ingredient_name(nodes_P)\n",
        "all_nodes = nodes_G + nodes_P\n",
        "all_degree = degree_G + degree_P\n",
        "all_ctg = [\"complete\"]*20 + [\"connected\"]*20"
      ],
      "metadata": {
        "id": "O5gbptlGjbu4"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "fig1 = px.bar(x=all_nodes, y=all_degree, color=all_ctg, barmode=\"group\")\n",
        "fig1.update_layout(\n",
        "    xaxis={'categoryorder':'total descending'},\n",
        "    title=\"Degree Disribution in Complete and Connnected Graph\",\n",
        "    xaxis_title=\"Top 20 Ingredients\",\n",
        "    yaxis_title=\"Degree\",\n",
        "    template=\"plotly_dark\")\n",
        "fig1.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 542
        },
        "id": "i_K6AfZRjbri",
        "outputId": "3a1d7d2e-66ed-4381-bb62-7c1b1782cf9f"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<html>\n",
              "<head><meta charset=\"utf-8\" /></head>\n",
              "<body>\n",
              "    <div>            <script src=\"https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/MathJax.js?config=TeX-AMS-MML_SVG\"></script><script type=\"text/javascript\">if (window.MathJax) {MathJax.Hub.Config({SVG: {font: \"STIX-Web\"}});}</script>                <script type=\"text/javascript\">window.PlotlyConfig = {MathJaxConfig: 'local'};</script>\n",
              "        <script src=\"https://cdn.plot.ly/plotly-2.8.3.min.js\"></script>                <div id=\"250866f9-094f-45d8-a5df-c1befc989f84\" class=\"plotly-graph-div\" style=\"height:525px; width:100%;\"></div>            <script type=\"text/javascript\">                                    window.PLOTLYENV=window.PLOTLYENV || {};                                    if (document.getElementById(\"250866f9-094f-45d8-a5df-c1befc989f84\")) {                    Plotly.newPlot(                        \"250866f9-094f-45d8-a5df-c1befc989f84\",                        [{\"alignmentgroup\":\"True\",\"hovertemplate\":\"color=complete<br>x=%{x}<br>y=%{y}<extra></extra>\",\"legendgroup\":\"complete\",\"marker\":{\"color\":\"#636efa\",\"pattern\":{\"shape\":\"\"}},\"name\":\"complete\",\"offsetgroup\":\"complete\",\"orientation\":\"v\",\"showlegend\":true,\"textposition\":\"auto\",\"x\":[\"salt\",\"water\",\"onions\",\"garlic\",\"sugar\",\"olive oil\",\"garlic cloves\",\"ground black pepper\",\"vegetable oil\",\"butter\",\"pepper\",\"all-purpose flour\",\"kosher salt\",\"carrots\",\"eggs\",\"tomatoes\",\"green onions\",\"black pepper\",\"large eggs\",\"unsalted butter\"],\"xaxis\":\"x\",\"y\":[4537,3593,3386,3285,3186,3182,2984,2809,2784,2716,2693,2521,2501,2453,2431,2257,2223,2219,2202,2085],\"yaxis\":\"y\",\"type\":\"bar\"},{\"alignmentgroup\":\"True\",\"hovertemplate\":\"color=connected<br>x=%{x}<br>y=%{y}<extra></extra>\",\"legendgroup\":\"connected\",\"marker\":{\"color\":\"#EF553B\",\"pattern\":{\"shape\":\"\"}},\"name\":\"connected\",\"offsetgroup\":\"connected\",\"orientation\":\"v\",\"showlegend\":true,\"textposition\":\"auto\",\"x\":[\"salt\",\"water\",\"olive oil\",\"sugar\",\"onions\",\"garlic\",\"garlic cloves\",\"butter\",\"vegetable oil\",\"ground black pepper\",\"pepper\",\"all-purpose flour\",\"kosher salt\",\"eggs\",\"large eggs\",\"extra-virgin olive oil\",\"green onions\",\"carrots\",\"unsalted butter\",\"tomatoes\"],\"xaxis\":\"x\",\"y\":[1955,1431,1326,1236,1197,1158,1087,1069,1044,1016,979,920,853,819,771,768,767,742,731,700],\"yaxis\":\"y\",\"type\":\"bar\"}],                        {\"template\":{\"data\":{\"barpolar\":[{\"marker\":{\"line\":{\"color\":\"rgb(17,17,17)\",\"width\":0.5},\"pattern\":{\"fillmode\":\"overlay\",\"size\":10,\"solidity\":0.2}},\"type\":\"barpolar\"}],\"bar\":[{\"error_x\":{\"color\":\"#f2f5fa\"},\"error_y\":{\"color\":\"#f2f5fa\"},\"marker\":{\"line\":{\"color\":\"rgb(17,17,17)\",\"width\":0.5},\"pattern\":{\"fillmode\":\"overlay\",\"size\":10,\"solidity\":0.2}},\"type\":\"bar\"}],\"carpet\":[{\"aaxis\":{\"endlinecolor\":\"#A2B1C6\",\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"minorgridcolor\":\"#506784\",\"startlinecolor\":\"#A2B1C6\"},\"baxis\":{\"endlinecolor\":\"#A2B1C6\",\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"minorgridcolor\":\"#506784\",\"startlinecolor\":\"#A2B1C6\"},\"type\":\"carpet\"}],\"choropleth\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"type\":\"choropleth\"}],\"contourcarpet\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"type\":\"contourcarpet\"}],\"contour\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"contour\"}],\"heatmapgl\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"heatmapgl\"}],\"heatmap\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"heatmap\"}],\"histogram2dcontour\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"histogram2dcontour\"}],\"histogram2d\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"histogram2d\"}],\"histogram\":[{\"marker\":{\"pattern\":{\"fillmode\":\"overlay\",\"size\":10,\"solidity\":0.2}},\"type\":\"histogram\"}],\"mesh3d\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"type\":\"mesh3d\"}],\"parcoords\":[{\"line\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"parcoords\"}],\"pie\":[{\"automargin\":true,\"type\":\"pie\"}],\"scatter3d\":[{\"line\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatter3d\"}],\"scattercarpet\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scattercarpet\"}],\"scattergeo\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scattergeo\"}],\"scattergl\":[{\"marker\":{\"line\":{\"color\":\"#283442\"}},\"type\":\"scattergl\"}],\"scattermapbox\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scattermapbox\"}],\"scatterpolargl\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatterpolargl\"}],\"scatterpolar\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatterpolar\"}],\"scatter\":[{\"marker\":{\"line\":{\"color\":\"#283442\"}},\"type\":\"scatter\"}],\"scatterternary\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatterternary\"}],\"surface\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"surface\"}],\"table\":[{\"cells\":{\"fill\":{\"color\":\"#506784\"},\"line\":{\"color\":\"rgb(17,17,17)\"}},\"header\":{\"fill\":{\"color\":\"#2a3f5f\"},\"line\":{\"color\":\"rgb(17,17,17)\"}},\"type\":\"table\"}]},\"layout\":{\"annotationdefaults\":{\"arrowcolor\":\"#f2f5fa\",\"arrowhead\":0,\"arrowwidth\":1},\"autotypenumbers\":\"strict\",\"coloraxis\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"colorscale\":{\"diverging\":[[0,\"#8e0152\"],[0.1,\"#c51b7d\"],[0.2,\"#de77ae\"],[0.3,\"#f1b6da\"],[0.4,\"#fde0ef\"],[0.5,\"#f7f7f7\"],[0.6,\"#e6f5d0\"],[0.7,\"#b8e186\"],[0.8,\"#7fbc41\"],[0.9,\"#4d9221\"],[1,\"#276419\"]],\"sequential\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"sequentialminus\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]]},\"colorway\":[\"#636efa\",\"#EF553B\",\"#00cc96\",\"#ab63fa\",\"#FFA15A\",\"#19d3f3\",\"#FF6692\",\"#B6E880\",\"#FF97FF\",\"#FECB52\"],\"font\":{\"color\":\"#f2f5fa\"},\"geo\":{\"bgcolor\":\"rgb(17,17,17)\",\"lakecolor\":\"rgb(17,17,17)\",\"landcolor\":\"rgb(17,17,17)\",\"showlakes\":true,\"showland\":true,\"subunitcolor\":\"#506784\"},\"hoverlabel\":{\"align\":\"left\"},\"hovermode\":\"closest\",\"mapbox\":{\"style\":\"dark\"},\"paper_bgcolor\":\"rgb(17,17,17)\",\"plot_bgcolor\":\"rgb(17,17,17)\",\"polar\":{\"angularaxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"},\"bgcolor\":\"rgb(17,17,17)\",\"radialaxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"}},\"scene\":{\"xaxis\":{\"backgroundcolor\":\"rgb(17,17,17)\",\"gridcolor\":\"#506784\",\"gridwidth\":2,\"linecolor\":\"#506784\",\"showbackground\":true,\"ticks\":\"\",\"zerolinecolor\":\"#C8D4E3\"},\"yaxis\":{\"backgroundcolor\":\"rgb(17,17,17)\",\"gridcolor\":\"#506784\",\"gridwidth\":2,\"linecolor\":\"#506784\",\"showbackground\":true,\"ticks\":\"\",\"zerolinecolor\":\"#C8D4E3\"},\"zaxis\":{\"backgroundcolor\":\"rgb(17,17,17)\",\"gridcolor\":\"#506784\",\"gridwidth\":2,\"linecolor\":\"#506784\",\"showbackground\":true,\"ticks\":\"\",\"zerolinecolor\":\"#C8D4E3\"}},\"shapedefaults\":{\"line\":{\"color\":\"#f2f5fa\"}},\"sliderdefaults\":{\"bgcolor\":\"#C8D4E3\",\"bordercolor\":\"rgb(17,17,17)\",\"borderwidth\":1,\"tickwidth\":0},\"ternary\":{\"aaxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"},\"baxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"},\"bgcolor\":\"rgb(17,17,17)\",\"caxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"}},\"title\":{\"x\":0.05},\"updatemenudefaults\":{\"bgcolor\":\"#506784\",\"borderwidth\":0},\"xaxis\":{\"automargin\":true,\"gridcolor\":\"#283442\",\"linecolor\":\"#506784\",\"ticks\":\"\",\"title\":{\"standoff\":15},\"zerolinecolor\":\"#283442\",\"zerolinewidth\":2},\"yaxis\":{\"automargin\":true,\"gridcolor\":\"#283442\",\"linecolor\":\"#506784\",\"ticks\":\"\",\"title\":{\"standoff\":15},\"zerolinecolor\":\"#283442\",\"zerolinewidth\":2}}},\"xaxis\":{\"anchor\":\"y\",\"domain\":[0.0,1.0],\"title\":{\"text\":\"Top 20 Ingredients\"},\"categoryorder\":\"total descending\"},\"yaxis\":{\"anchor\":\"x\",\"domain\":[0.0,1.0],\"title\":{\"text\":\"Degree\"}},\"legend\":{\"title\":{\"text\":\"color\"},\"tracegroupgap\":0},\"margin\":{\"t\":60},\"barmode\":\"group\",\"title\":{\"text\":\"Degree Disribution in Complete and Connnected Graph\"}},                        {\"responsive\": true}                    ).then(function(){\n",
              "                            \n",
              "var gd = document.getElementById('250866f9-094f-45d8-a5df-c1befc989f84');\n",
              "var x = new MutationObserver(function (mutations, observer) {{\n",
              "        var display = window.getComputedStyle(gd).display;\n",
              "        if (!display || display === 'none') {{\n",
              "            console.log([gd, 'removed!']);\n",
              "            Plotly.purge(gd);\n",
              "            observer.disconnect();\n",
              "        }}\n",
              "}});\n",
              "\n",
              "// Listen for the removal of the full notebook cells\n",
              "var notebookContainer = gd.closest('#notebook-container');\n",
              "if (notebookContainer) {{\n",
              "    x.observe(notebookContainer, {childList: true});\n",
              "}}\n",
              "\n",
              "// Listen for the clearing of the current output cell\n",
              "var outputEl = gd.closest('.output');\n",
              "if (outputEl) {{\n",
              "    x.observe(outputEl, {childList: true});\n",
              "}}\n",
              "\n",
              "                        })                };                            </script>        </div>\n",
              "</body>\n",
              "</html>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "0k9-o5Gq3zqo"
      },
      "source": [
        "# **Building Embedding Graph Models**"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "ZkPn5Ylw5DHl"
      },
      "outputs": [],
      "source": [
        "def create_biased_random_walker(graph, p, q):\n",
        "  #   p: defines (unormalised) probability, 1/p, of returning to source node\n",
        "  #   q: defines (unormalised) probability, 1/q, for moving away from source node\n",
        "  return BiasedRandomWalk(graph, n=args.walk_number, length=args.walk_length, p=p, q=q)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "### **Word2Vec**"
      ],
      "metadata": {
        "id": "UoRalqQYLnVM"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def word2vec_embedding(G):\n",
        "  random_walk = BiasedRandomWalk(G)\n",
        "  walker = random_walk.run(nodes=list(G.nodes()), n=args.walk_number, length=args.walk_length, p=0.5, q=0.2)\n",
        "  str_walks = [[str(n) for n in walk] for walk in walker]\n",
        "  model = Word2Vec(str_walks, size=args.dim, window=5, min_count=0, sg=1, workers=2, iter=1)\n",
        "  return model.wv.vectors"
      ],
      "metadata": {
        "id": "3EQDIamZLlqO"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "### **Node2Vec**"
      ],
      "metadata": {
        "id": "oKUvwMOfLpVh"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "tkIuk743Wgbx"
      },
      "outputs": [],
      "source": [
        "from stellargraph.mapper import Node2VecLinkGenerator, Node2VecNodeGenerator\n",
        "from stellargraph.layer import Node2Vec, link_classification\n",
        "\n",
        "def node2vec_embedding(G): \n",
        "  #   Create the biased random walker to generate random walks\n",
        "  walker = create_biased_random_walker(G, 0.5, 0.2)\n",
        "  #   Create the unsupervised sampler to sample (target, context) pairs from random walks\n",
        "  unsupervised_samples = UnsupervisedSampler(G, nodes=list(G.nodes()), walker=walker)\n",
        "\n",
        "  #   Define a Node2Vec training generator, which generates batches of training pairs\n",
        "  generator = Node2VecLinkGenerator(G, args.batch_size)\n",
        "\n",
        "  #   Create the Node2Vec model\n",
        "  node2vec = Node2Vec(args.dim, generator=generator)\n",
        "\n",
        "  #   Build the model and expose input and output sockets of Node2Vec, for node pair inputs\n",
        "  x_inp, x_out = node2vec.in_out_tensors()\n",
        "\n",
        "  #   Use the link_classification function to generate the output of the Node2Vec model\n",
        "  prediction = link_classification(output_dim=1, output_act=\"sigmoid\", edge_embedding_method=\"dot\")(x_out)\n",
        "\n",
        "  #   Stack the Node2Vec encoder and prediction layer into a Keras model, and specify the loss\n",
        "  model = keras.Model(inputs=x_inp, outputs=prediction)\n",
        "  \n",
        "  model.compile(\n",
        "      optimizer=keras.optimizers.Adam(learning_rate=args.lr),\n",
        "      loss=keras.losses.binary_crossentropy,\n",
        "      metrics=[keras.metrics.binary_accuracy]\n",
        "  )\n",
        "\n",
        "  es_callback = keras.callbacks.EarlyStopping(monitor=\"binary_accuracy\", patience=3, verbose=1, mode=\"max\")\n",
        "\n",
        "  #   Train the model\n",
        "  model.fit(\n",
        "      generator.flow(unsupervised_samples),\n",
        "      epochs=args.epoch,\n",
        "      verbose=2,\n",
        "      use_multiprocessing=False,\n",
        "      workers=4,\n",
        "      shuffle=True,\n",
        "      callbacks=[es_callback]\n",
        "  )\n",
        "\n",
        "  #   Build the model to predict node representations from node ids with the learned Node2Vec model parameters\n",
        "  x_inp_src = x_inp[0]\n",
        "  x_out_src = x_out[0]\n",
        "  embedding_model = keras.Model(inputs=x_inp_src, outputs=x_out_src)\n",
        "\n",
        "  # Get representations for all nodes in ``graph``\n",
        "  node_gen = Node2VecNodeGenerator(G, args.batch_size).flow(list(G.nodes()))\n",
        "  return embedding_model.predict(node_gen, workers=1, verbose=0)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "### **GraphSAGE**"
      ],
      "metadata": {
        "id": "nLbGEXGALdNd"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "M5fP-j2fmOyo"
      },
      "outputs": [],
      "source": [
        "from stellargraph.mapper import GraphSAGELinkGenerator, GraphSAGENodeGenerator\n",
        "from stellargraph.layer import GraphSAGE\n",
        "\n",
        "def graphsage_embedding(G):\n",
        "  #   Defining a 2-layer GraphSAGE model\n",
        "  num_samples = [10, 5]  \n",
        "  layer_sizes = [args.dim, args.dim]\n",
        "  assert len(layer_sizes) == len(num_samples)\n",
        "\n",
        "  \n",
        "  #   Create the biased random walker to generate random walks\n",
        "  walker = create_biased_random_walker(G, 0.5, 0.2)\n",
        "  \n",
        "  #   Create the unsupervised sampler to sample (target, context) pairs from random walks\n",
        "  unsupervised_samples = UnsupervisedSampler(G, nodes=list(G.nodes()), walker=walker)\n",
        "\n",
        "  #   Define a GraphSAGE training generator, which generates batches of training pairs\n",
        "  generator = GraphSAGELinkGenerator(G, args.batch_size, num_samples=num_samples)\n",
        "\n",
        "  #   Create the GraphSAGE model\n",
        "  graphsage = GraphSAGE(\n",
        "        layer_sizes=layer_sizes,\n",
        "        generator=generator,\n",
        "        bias=True,\n",
        "        dropout=0.2,\n",
        "        normalize=\"l2\"\n",
        "  )\n",
        "\n",
        "  #   Build the model and expose input and output sockets of GraphSAGE, for node pair inputs\n",
        "  x_inp, x_out = graphsage.in_out_tensors()\n",
        "\n",
        "  #   Use the link_classification function to generate the output of the GraphSAGE model\n",
        "  prediction = link_classification(output_dim=1, output_act=\"relu\", edge_embedding_method=\"ip\")(x_out)\n",
        "\n",
        "  #   Stack the GraphSAGE encoder and prediction layer into a Keras model, and specify the loss\n",
        "  model = keras.Model(inputs=x_inp, outputs=prediction)\n",
        "  \n",
        "  model.compile(\n",
        "      optimizer=keras.optimizers.Adam(learning_rate=args.lr),\n",
        "      loss=keras.losses.binary_crossentropy,\n",
        "      metrics=[keras.metrics.binary_accuracy]\n",
        "  )\n",
        "\n",
        "  es_callback = keras.callbacks.EarlyStopping(monitor=\"binary_accuracy\", patience=3, verbose=1, mode=\"max\")\n",
        "\n",
        "  #   Train the model\n",
        "  model.fit(\n",
        "      generator.flow(unsupervised_samples),\n",
        "      epochs=args.epoch,\n",
        "      verbose=2,\n",
        "      use_multiprocessing=False,\n",
        "      workers=4,\n",
        "      shuffle=True,\n",
        "      callbacks=[es_callback]\n",
        "  )\n",
        "\n",
        "  #   Build the model to predict node representations from node features with the learned GraphSAGE model parameters\n",
        "  x_inp_src = x_inp[0::2]\n",
        "  x_out_src = x_out[0]\n",
        "  embedding_model = keras.Model(inputs=x_inp_src, outputs=x_out_src)\n",
        "\n",
        "  # Get representations for all nodes in ``graph``\n",
        "  node_gen = GraphSAGENodeGenerator(G, args.batch_size, num_samples).flow(list(G.nodes()))\n",
        "  return embedding_model.predict(node_gen, workers=1, verbose=0)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "### **GCN**"
      ],
      "metadata": {
        "id": "vkKl21XxLghB"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "Qzrpa7IYNuEy"
      },
      "outputs": [],
      "source": [
        "from stellargraph.mapper import FullBatchLinkGenerator, FullBatchNodeGenerator\n",
        "from stellargraph.layer import GCN, LinkEmbedding\n",
        "\n",
        "def gcn_embedding(G):\n",
        "  #   Defining a 2-layer GCN model\n",
        "  layer_sizes = [args.dim, args.dim]\n",
        "  \n",
        "  #   Create the biased random walker to generate random walks\n",
        "  walker = create_biased_random_walker(G, 1.0, 1.0)\n",
        "  \n",
        "  #   Create the unsupervised sampler to sample (target, context) pairs from random walks\n",
        "  unsupervised_samples = UnsupervisedSampler(G, nodes=list(G.nodes()), walker=walker)\n",
        "\n",
        "  #   Define a GCN training generator, which generates the full batch of training pairs\n",
        "  generator = FullBatchLinkGenerator(G, method=\"gcn\")\n",
        "\n",
        "  #   Create the GCN model\n",
        "  gcn = GCN(\n",
        "      layer_sizes=layer_sizes,\n",
        "      activations=[\"relu\", \"relu\"],\n",
        "      generator=generator,\n",
        "      dropout=0.3\n",
        "  )\n",
        "\n",
        "  #   Build the model and expose input and output sockets of GCN, for node pair inputs\n",
        "  x_inp, x_out = gcn.in_out_tensors()\n",
        "\n",
        "  #   Use the dot product of node embeddings to make node pairs co-occurring in short random walks represented closely\n",
        "  prediction = LinkEmbedding(activation=\"sigmoid\", method=\"ip\")(x_out)\n",
        "  prediction = keras.layers.Reshape((-1,))(prediction)\n",
        "\n",
        "  #   Stack the GCN encoder and prediction layer into a Keras model, and specify the loss\n",
        "  model = keras.Model(inputs=x_inp, outputs=prediction)\n",
        "  \n",
        "  model.compile(\n",
        "      optimizer=keras.optimizers.Adam(learning_rate=1e-3),\n",
        "      loss=keras.losses.binary_crossentropy,\n",
        "      metrics=[keras.metrics.binary_accuracy]\n",
        "  )\n",
        "\n",
        "  # Train the model\n",
        "  batches = unsupervised_samples.run(args.batch_size)\n",
        "  for epoch in range(args.epoch):\n",
        "    for batch in batches:\n",
        "      samples = generator.flow(batch[0], targets=batch[1], use_ilocs=True)[0]\n",
        "      [loss, accuracy] = model.train_on_batch(x=samples[0], y=samples[1])\n",
        "    print(\n",
        "      f\"Epoch: {epoch+1}/{args.epoch} - \"\n",
        "      + \" loss: {:6.4f}\".format(loss)\n",
        "      + \" - binary_accuracy: {:6.4f}\".format(accuracy)\n",
        "    )\n",
        "\n",
        "  # Get representations for all nodes in ``graph``\n",
        "  embedding_model = keras.Model(inputs=x_inp, outputs=x_out)\n",
        "  node_embeddings = embedding_model.predict(generator.flow(list(zip(list(G.nodes()), list(G.nodes())))))\n",
        "  return node_embeddings[0][:, 0, :]"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Training Embedding Graph Models**"
      ],
      "metadata": {
        "id": "THgu6SvpLRE2"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def train_embedding(method, G):\n",
        "  start_time = time.time()\n",
        "\n",
        "  if method == \"Word2Vec\":\n",
        "    node_embedding = word2vec_embedding(G)\n",
        "  elif method == \"Node2Vec\":\n",
        "    node_embedding = node2vec_embedding(G)\n",
        "  elif method == \"GraphSAGE\":\n",
        "    node_embedding = graphsage_embedding(G)\n",
        "  elif method == \"GCN\":\n",
        "    node_embedding = gcn_embedding(G)\n",
        "  else:\n",
        "    print(\"Pick appropriate embedding method\")\n",
        "    return\n",
        "  \n",
        "  pd.DataFrame(node_embedding).to_csv(os.path.join(results,\n",
        "    f\"{method}_{args.graph_method}\" + \n",
        "    f\"graph_{str(timedelta(seconds=time.time() - start_time))}\" +\n",
        "    f\"dim_{args.dim}\" +\n",
        "    f\"epochs_{args.epoch}\" +\n",
        "    f\"batchsize_{args.batch_size}\" +\n",
        "    f\"walknumber_{args.walk_number}\" +\n",
        "    f\"walklength_{args.walk_length}.csv\"))\n"
      ],
      "metadata": {
        "id": "p1Dy5rjYYW75"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "methods = [\"Word2Vec\", \"Node2Vec\", \"GraphSAGE\", \"GCN\"]\n",
        "method = methods[1]\n",
        "print(f\"Selected embedding method {method}\")\n",
        "train_embedding(method, G_w_node_feature)"
      ],
      "metadata": {
        "id": "T2o8SeqtZYFj"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Pick Embedding**"
      ],
      "metadata": {
        "id": "Ww63IIxnH7gL"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "paths = os.listdir(results)\n",
        "for i in range(len(paths)):\n",
        "  print(f\"{i}: {paths[i]}\")"
      ],
      "metadata": {
        "id": "fQemBxtPHPgD",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "7e35431e-725e-4991-b0cd-5ef8ad504c7d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "0: Word2Vec_connectedgraph_0:12:08.268310dim_128epochs_10batchsize_64walknumber_10walklength_100.csv\n",
            "1: Node2Vec_connectedgraph_2:11:51.515414dim_128epochs_10batchsize_64walknumber_10walklength_100.csv\n",
            "2: Word2Vec_completegraph_0:20:48.676683dim_128epochs_10batchsize_64walknumber_10walklength_100.csv\n",
            "3: Node2Vec_completegraph_3:28:50.135091dim_128epochs_10batchsize_64walknumber_10walklength_100.csv\n",
            "4: GCN_connectedgraph_1:52:23.785886dim_64epochs_10batchsize_64walknumber_10walklength_20.csv\n",
            "5: GraphSAGE_connectedgraph_7:03:52.822946dim_128epochs_4batchsize_64walknumber_10walklength_30.csv\n",
            "6: GCN_completegraph_3:06:44.202400dim_128epochs_4batchsize_64walknumber_10walklength_30.csv\n",
            "7: test_classification_answer.csv\n",
            "8: test_completion_answer.csv\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "path = paths[1]\n",
        "print(f\"Chosen embedding: {path}\")\n",
        "embeddings = pd.read_csv(os.path.join(results, path))\n",
        "embeddings = embeddings.drop(embeddings.columns[0], axis=1)"
      ],
      "metadata": {
        "id": "aR1YR-dfbVVq",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "0273499b-2c7c-4ef2-b2ed-ec4eb9f89222"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Chosen embedding: Node2Vec_connectedgraph_2:11:51.515414dim_128epochs_10batchsize_64walknumber_10walklength_100.csv\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Classification**"
      ],
      "metadata": {
        "id": "iFdTk74nSlls"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Dataset Constructor**"
      ],
      "metadata": {
        "id": "nmef-LWGEqhY"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "class MyDataset(torch.utils.data.Dataset):\n",
        "  def __init__(self, recipes, targets):\n",
        "    self.recipes = torch.from_numpy(recipes).float()\n",
        "    self.targets = torch.from_numpy(targets).float()\n",
        "\n",
        "  def __len__(self):\n",
        "    return len(self.targets)\n",
        "\n",
        "  def __getitem__(self, idx):\n",
        "    target = self.targets[idx]\n",
        "    recipe = self.recipes[idx]\n",
        "    sample = {\"recipe\": recipe, \"label\": target}\n",
        "    return sample"
      ],
      "metadata": {
        "id": "yY2Ky-9lVtxD"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Data Loader**"
      ],
      "metadata": {
        "id": "mWJPf8_vHmpR"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def load_data_classification(embedding, labels, stage):\n",
        "  if stage == \"train\":\n",
        "    with open (\"train.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      emb = np.zeros((len(data), args.dim))\n",
        "      target = np.zeros((len(data), len(labels)))\n",
        "      \n",
        "      for i in range(len(data)):\n",
        "        ingredients = data[i].strip().split(\",\")\n",
        "        origin = ingredients.pop()\n",
        "        ingredients = [int(x) for x in ingredients]\n",
        "        average = np.zeros((len(ingredients), args.dim))\n",
        "        for j in range(len(ingredients)):\n",
        "          average[j, :] = embedding.iloc[ingredients[j], :]\n",
        "        \n",
        "        emb[i, :] = np.mean(average, axis=0)\n",
        "        target[i, labels[origin]] = 1\n",
        "    \n",
        "  elif stage == \"val\":\n",
        "    with open (\"validation_classification_answer.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      target = np.zeros((len(data), len(labels)))\n",
        "      for i in range(len(data)):\n",
        "        origin = data[i].strip()\n",
        "        target[i, labels[origin]] = 1\n",
        "    \n",
        "    with open (\"validation_classification_question.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      emb = np.zeros((len(data), args.dim))\n",
        "      for i in range(len(data)):\n",
        "        ingredients = data[i].strip().split(\",\")\n",
        "        ingredients = [int(x) for x in ingredients]\n",
        "        average = np.zeros((len(ingredients), args.dim))\n",
        "        for j in range(len(ingredients)):\n",
        "          average[j, :] = embedding.iloc[ingredients[j], :]\n",
        "        \n",
        "        emb[i, :] = np.mean(average, axis=0)\n",
        "    \n",
        "  elif stage == \"test\":\n",
        "    with open (\"test_classification_question.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      emb = np.zeros((len(data), args.dim))\n",
        "      target = np.zeros((len(data), 1))\n",
        "      for i in range(len(data)):\n",
        "        ingredients = data[i].strip().split(\",\")\n",
        "        ingredients = [int(x) for x in ingredients]\n",
        "        average = np.zeros((len(ingredients), args.dim))\n",
        "        \n",
        "        for j in range(len(ingredients)):\n",
        "          average[j, :] = embedding.iloc[ingredients[j], :]\n",
        "        \n",
        "        emb[i, :] = np.mean(average, axis=0)\n",
        "\n",
        "  return emb, target"
      ],
      "metadata": {
        "id": "D__noB6kMOyU"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model**"
      ],
      "metadata": {
        "id": "p_nCEIGX2Uzc"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "class Classifier(nn.Module):\n",
        "  def __init__(self, num_feature, num_class):\n",
        "    super().__init__()\n",
        "    \n",
        "    self.fc1 = nn.Linear(num_feature, 512)\n",
        "    self.bn1 = nn.BatchNorm1d(512)\n",
        "    \n",
        "    self.fc2 = nn.Linear(512, 64)\n",
        "    self.bn2 = nn.BatchNorm1d(64)\n",
        "\n",
        "    self.fc3 = nn.Linear(64, num_class)\n",
        "    self.bn3 = nn.BatchNorm1d(num_class)\n",
        "\n",
        "    self.dp = nn.Dropout(0.3)\n",
        "    self.act = nn.ReLU()\n",
        "\n",
        "  def forward(self, x):\n",
        "    output = self.dp(self.act(self.bn1(self.fc1(x))))\n",
        "    output = self.dp(self.act(self.bn2(self.fc2(output))))\n",
        "    output = self.bn3(self.fc3(output))\n",
        "\n",
        "    return output"
      ],
      "metadata": {
        "id": "vc7VFTDSh0wd"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **RF**"
      ],
      "metadata": {
        "id": "MOTvWOtMIZdW"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def load_data_classification_RF(embedding, labels, stage):\n",
        "  if stage == \"train\":\n",
        "    with open (\"train.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      emb = np.zeros((len(data), args.dim))\n",
        "      target = []\n",
        "      \n",
        "      for i in range(len(data)):\n",
        "        ingredients = data[i].strip().split(\",\")\n",
        "        origin = ingredients.pop()\n",
        "        ingredients = [int(x) for x in ingredients]\n",
        "        average = np.zeros((len(ingredients), args.dim))\n",
        "        for j in range(len(ingredients)):\n",
        "          average[j, :] = embedding.iloc[ingredients[j], :]\n",
        "        \n",
        "        emb[i, :] = np.mean(average, axis=0)\n",
        "        target.append(labels[origin])\n",
        "    \n",
        "  elif stage == \"val\":\n",
        "    with open (\"validation_classification_answer.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      target = []\n",
        "      for i in range(len(data)):\n",
        "        origin = data[i].strip()\n",
        "        target.append(labels[origin])\n",
        "    \n",
        "    with open (\"validation_classification_question.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      emb = np.zeros((len(data), args.dim))\n",
        "      for i in range(len(data)):\n",
        "        ingredients = data[i].strip().split(\",\")\n",
        "        ingredients = [int(x) for x in ingredients]\n",
        "        average = np.zeros((len(ingredients), args.dim))\n",
        "        for j in range(len(ingredients)):\n",
        "          average[j, :] = embedding.iloc[ingredients[j], :]\n",
        "        \n",
        "        emb[i, :] = np.mean(average, axis=0)\n",
        "  return emb, target"
      ],
      "metadata": {
        "id": "RMbMdIfgMLWb"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x_train_clf_dataset_RF, y_train_clf_dataset_RF = load_data_classification_RF(embeddings, labels, \"train\")\n",
        "x_val_clf_dataset_RF, y_val_clf_dataset_RF = load_data_classification_RF(embeddings, labels, \"val\")"
      ],
      "metadata": {
        "id": "YyiLwnGdPyNK"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "estimators = [15, 20, 30]\n",
        "for i in estimators:\n",
        "  clf = RandomForestClassifier( bootstrap=True, class_weight=None, criterion='gini',\n",
        "                                max_depth = 20, max_features = 'auto', max_leaf_nodes = None,\n",
        "                                min_impurity_decrease = 0.0,\n",
        "                                min_samples_leaf = 3, min_samples_split = 3,\n",
        "                                min_weight_fraction_leaf = 0.0, n_estimators = i, n_jobs =-1,\n",
        "                                random_state = 25,warm_start = False)\n",
        "\n",
        "  clf.fit(x_train_clf_dataset_RF, y_train_clf_dataset_RF) # fitting the classifier\n",
        "  \n",
        "  train_sc = accuracy_score(y_train_clf_dataset_RF, clf.predict(x_train_clf_dataset_RF))\n",
        "  val_sc = accuracy_score(y_val_clf_dataset_RF, clf.predict(x_val_clf_dataset_RF))\n",
        "  \n",
        "  print('Estimators = ',i,'Train Score',train_sc,'test Score',val_sc)"
      ],
      "metadata": {
        "id": "7v2m943rMfHL"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Training**"
      ],
      "metadata": {
        "id": "AoMCjJG7sjti"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "x_train_clf_dataset, y_train_clf_dataset = load_data_classification(embeddings, labels, \"train\")\n",
        "x_val_clf_dataset, y_val_clf_dataset = load_data_classification(embeddings, labels, \"val\")\n",
        "x_test_clf_dataset, y_test_clf_dataset = load_data_classification(embeddings, labels, \"test\")\n",
        "\n",
        "train_clf_dataset = MyDataset(x_train_clf_dataset, y_train_clf_dataset)\n",
        "val_clf_dataset = MyDataset(x_val_clf_dataset, y_val_clf_dataset)\n",
        "test_clf_dataset = MyDataset(x_test_clf_dataset, y_test_clf_dataset)\n",
        "\n",
        "train_clf_dataloader = torch.utils.data.DataLoader(train_clf_dataset, batch_size=args.batch_size, shuffle=True, num_workers=2)\n",
        "val_clf_dataloader = torch.utils.data.DataLoader(val_clf_dataset, batch_size=args.batch_size, shuffle=False, num_workers=2)\n",
        "test_clf_dataloader = torch.utils.data.DataLoader(test_clf_dataset, batch_size=args.batch_size, shuffle=False, num_workers=2)"
      ],
      "metadata": {
        "id": "eI6Sq_SySoKP"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def train_classifier(model, optimizer, scheduler):\n",
        "  best_accuracy, best_f1, best_f2 = 0, 0, 0\n",
        "  t_loss, t_accuracy = [], []\n",
        "  v_loss, v_accuracy = [], []\n",
        "  v_f1, v_f2 = [], []\n",
        "  for epoch in range(args.epoch_):\n",
        "    # Here starts the train loop.\n",
        "    train_loss = []\n",
        "    train_accuracy = []\n",
        "    model.train()\n",
        "    for batch_idx, data in enumerate(train_clf_dataloader):\n",
        "      x = data['recipe'].to(device)\n",
        "      y = data['label'].to(device)\n",
        "      y = y.type(torch.float)\n",
        "\n",
        "      #   Feed `x` into the network, get an output, and keep it in a variable called `logit`. \n",
        "      pred = model(x)\n",
        "      pred = pred.type(torch.float).to(device)\n",
        "                                 \n",
        "      #   Accuracy\n",
        "      accuracy = (torch.argmax(torch.log_softmax(pred, 1), 1) == torch.argmax(y, 1)).float().mean()\n",
        "\n",
        "      #   Compute loss using `logit` and `y`, and keep it in a variable called `loss`.\n",
        "      loss = nn.CrossEntropyLoss()(pred, y)\n",
        "\n",
        "      #   Flush out the previously computed gradient.\n",
        "      optimizer.zero_grad()\n",
        "\n",
        "      #   Backward the computed loss. \n",
        "      loss.backward()\n",
        "\n",
        "      #   Update the network weights. \n",
        "      optimizer.step()\n",
        "      \n",
        "      train_loss.append(loss.item())\n",
        "      train_accuracy.append(accuracy.item())\n",
        "      \n",
        "      if batch_idx % 50 == 0:\n",
        "        print(f'Epoch : {epoch} || {batch_idx}/{len(train_clf_dataloader)} || \\\n",
        "                loss : {loss.item():.3f}, acc : {accuracy:.3f}')\n",
        "\n",
        "    mean_train_loss = sum(train_loss) / len(train_loss)\n",
        "    mean_train_accuracy = sum(train_accuracy) / len(train_accuracy)\n",
        "\n",
        "    t_loss.append(mean_train_loss)          # t_loss += train_loss\n",
        "    t_accuracy.append(mean_train_accuracy)  # t_accuracy += train_accuracy\n",
        "\n",
        "    model.eval()\n",
        "    with torch.no_grad():\n",
        "      val_loss = []\n",
        "      val_accuracy = []\n",
        "      val_f1, val_f2 = [], []\n",
        "      for batch_idx, data in enumerate(val_clf_dataloader):\n",
        "        x = data['recipe'].to(device=device)\n",
        "        y = data['label'].to(device=device)\n",
        "        y = y.type(torch.float)\n",
        "  \n",
        "        #   Feed `x` into the network, get an output, and keep it in a variable called `logit`. \n",
        "        pred = model(x)\n",
        "        pred = pred.type(torch.float).to(device)\n",
        "        \n",
        "        #   Accuracy\n",
        "        accuracy = (torch.argmax(torch.log_softmax(pred, 1), 1) == torch.argmax(y, 1)).float().mean()\n",
        "        y_true = torch.argmax(y, 1).cpu().detach().numpy()\n",
        "        y_pred = torch.argmax(torch.log_softmax(pred, 1), 1).cpu().detach().numpy()\n",
        "\n",
        "        acc1 = f1_score(y_true, y_pred, average='micro')\n",
        "        acc2 = f1_score(y_true, y_pred, average='macro')\n",
        "        #   Compute loss using `logit` and `y`, and keep it in a variable called `loss`.\n",
        "        loss = nn.CrossEntropyLoss()(pred, y)\n",
        "\n",
        "        val_loss.append(loss.item())\n",
        "        val_accuracy.append(accuracy.item())\n",
        "        val_f1.append(acc1)\n",
        "        val_f2.append(acc2)\n",
        "\n",
        "      mean_val_loss = sum(val_loss) / len(val_loss)\n",
        "      mean_val_accuracy = sum(val_accuracy) / len(val_accuracy)\n",
        "      mean_val_f1 = sum(val_f1) / len(val_f1)\n",
        "      mean_val_f2 = sum(val_f2) / len(val_f2)\n",
        "      v_loss.append(mean_val_loss)\n",
        "      v_accuracy.append(mean_val_accuracy)\n",
        "      v_f1.append(mean_val_f1)\n",
        "      v_f2.append(mean_val_f2)\n",
        "      print(f'Validation result of epoch {epoch}/{args.epoch_} || val_loss : {mean_val_loss:.3f} val_acc : {mean_val_accuracy:.3f} ')\n",
        "\n",
        "      #   Whenever `test_accuracy` is greater than `best_accuracy`, save network weights with the filename 'best.pt' in the directory specified by `ckpt_dir`.\n",
        "      if mean_val_accuracy > best_accuracy:\n",
        "          best_accuracy = mean_val_accuracy\n",
        "          best_val_model = copy.deepcopy(model.state_dict())\n",
        "      if mean_val_f1 > best_f1:\n",
        "          best_f1 = mean_val_f1\n",
        "      if mean_val_f2 > best_f2:\n",
        "          best_f2 = mean_val_f2\n",
        "    \n",
        "    scheduler.step()\n",
        "\n",
        "  torch.save(best_val_model, os.path.join(models_path, \"classifier\"))\n",
        "  return best_accuracy, best_f1, best_f2, t_loss, t_accuracy, v_loss, v_accuracy, best_val_model"
      ],
      "metadata": {
        "id": "2AhfsXSjJtvh"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model = Classifier(args.dim, len(labels))\n",
        "model = model.to(device)\n",
        "optimizer = torch.optim.Adam(model.parameters(), lr=args.lr, weight_decay=0.0001)\n",
        "scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[50,80], gamma=0.5)\n",
        "start_time = time.time()\n",
        "accuracy, best_f1, best_f2, train_loss_clf, train_accuracy_clf, val_loss_clf, val_accuracy_clf, best_val_model = train_classifier(model, optimizer, scheduler)\n",
        "duration = time.time() - start_time\n",
        "print(f'Best validation accuracy CrossE of classifier network : {accuracy:.3f} took {duration:.3f} secs')\n",
        "print(f'Best validation accuracy Micro of classifier network : {best_f1:.3f} took {duration:.3f} secs')\n",
        "print(f'Best validation accuracy Macro of classifier network : {best_f2:.3f} took {duration:.3f} secs')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4vSDyOT45oI8",
        "outputId": "fa71718e-c087-44f6-eb76-ab9de5c3582d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch : 0 || 0/368 ||                 loss : 3.264, acc : 0.125\n",
            "Epoch : 0 || 50/368 ||                 loss : 2.591, acc : 0.156\n",
            "Epoch : 0 || 100/368 ||                 loss : 2.270, acc : 0.422\n",
            "Epoch : 0 || 150/368 ||                 loss : 2.395, acc : 0.328\n",
            "Epoch : 0 || 200/368 ||                 loss : 1.999, acc : 0.484\n",
            "Epoch : 0 || 250/368 ||                 loss : 2.119, acc : 0.391\n",
            "Epoch : 0 || 300/368 ||                 loss : 1.927, acc : 0.484\n",
            "Epoch : 0 || 350/368 ||                 loss : 1.928, acc : 0.500\n",
            "Validation result of epoch 0/100 || val_loss : 1.808 val_acc : 0.535 \n",
            "Epoch : 1 || 0/368 ||                 loss : 1.867, acc : 0.516\n",
            "Epoch : 1 || 50/368 ||                 loss : 1.847, acc : 0.500\n",
            "Epoch : 1 || 100/368 ||                 loss : 1.985, acc : 0.469\n",
            "Epoch : 1 || 150/368 ||                 loss : 1.904, acc : 0.484\n",
            "Epoch : 1 || 200/368 ||                 loss : 1.824, acc : 0.500\n",
            "Epoch : 1 || 250/368 ||                 loss : 1.655, acc : 0.516\n",
            "Epoch : 1 || 300/368 ||                 loss : 1.591, acc : 0.578\n",
            "Epoch : 1 || 350/368 ||                 loss : 1.672, acc : 0.578\n",
            "Validation result of epoch 1/100 || val_loss : 1.601 val_acc : 0.559 \n",
            "Epoch : 2 || 0/368 ||                 loss : 1.481, acc : 0.562\n",
            "Epoch : 2 || 50/368 ||                 loss : 1.620, acc : 0.562\n",
            "Epoch : 2 || 100/368 ||                 loss : 1.759, acc : 0.469\n",
            "Epoch : 2 || 150/368 ||                 loss : 1.801, acc : 0.438\n",
            "Epoch : 2 || 200/368 ||                 loss : 1.633, acc : 0.531\n",
            "Epoch : 2 || 250/368 ||                 loss : 1.553, acc : 0.516\n",
            "Epoch : 2 || 300/368 ||                 loss : 1.949, acc : 0.516\n",
            "Epoch : 2 || 350/368 ||                 loss : 1.736, acc : 0.516\n",
            "Validation result of epoch 2/100 || val_loss : 1.493 val_acc : 0.575 \n",
            "Epoch : 3 || 0/368 ||                 loss : 1.400, acc : 0.578\n",
            "Epoch : 3 || 50/368 ||                 loss : 1.366, acc : 0.672\n",
            "Epoch : 3 || 100/368 ||                 loss : 1.286, acc : 0.656\n",
            "Epoch : 3 || 150/368 ||                 loss : 1.387, acc : 0.594\n",
            "Epoch : 3 || 200/368 ||                 loss : 1.564, acc : 0.484\n",
            "Epoch : 3 || 250/368 ||                 loss : 1.648, acc : 0.531\n",
            "Epoch : 3 || 300/368 ||                 loss : 1.822, acc : 0.422\n",
            "Epoch : 3 || 350/368 ||                 loss : 1.406, acc : 0.625\n",
            "Validation result of epoch 3/100 || val_loss : 1.447 val_acc : 0.586 \n",
            "Epoch : 4 || 0/368 ||                 loss : 1.486, acc : 0.594\n",
            "Epoch : 4 || 50/368 ||                 loss : 1.266, acc : 0.656\n",
            "Epoch : 4 || 100/368 ||                 loss : 1.250, acc : 0.703\n",
            "Epoch : 4 || 150/368 ||                 loss : 1.662, acc : 0.547\n",
            "Epoch : 4 || 200/368 ||                 loss : 1.635, acc : 0.500\n",
            "Epoch : 4 || 250/368 ||                 loss : 1.443, acc : 0.594\n",
            "Epoch : 4 || 300/368 ||                 loss : 1.327, acc : 0.625\n",
            "Epoch : 4 || 350/368 ||                 loss : 1.260, acc : 0.656\n",
            "Validation result of epoch 4/100 || val_loss : 1.399 val_acc : 0.596 \n",
            "Epoch : 5 || 0/368 ||                 loss : 1.473, acc : 0.500\n",
            "Epoch : 5 || 50/368 ||                 loss : 1.333, acc : 0.594\n",
            "Epoch : 5 || 100/368 ||                 loss : 1.184, acc : 0.656\n",
            "Epoch : 5 || 150/368 ||                 loss : 1.112, acc : 0.703\n",
            "Epoch : 5 || 200/368 ||                 loss : 1.257, acc : 0.688\n",
            "Epoch : 5 || 250/368 ||                 loss : 1.462, acc : 0.531\n",
            "Epoch : 5 || 300/368 ||                 loss : 1.404, acc : 0.594\n",
            "Epoch : 5 || 350/368 ||                 loss : 1.397, acc : 0.625\n",
            "Validation result of epoch 5/100 || val_loss : 1.368 val_acc : 0.603 \n",
            "Epoch : 6 || 0/368 ||                 loss : 1.483, acc : 0.578\n",
            "Epoch : 6 || 50/368 ||                 loss : 1.211, acc : 0.625\n",
            "Epoch : 6 || 100/368 ||                 loss : 1.225, acc : 0.641\n",
            "Epoch : 6 || 150/368 ||                 loss : 1.226, acc : 0.688\n",
            "Epoch : 6 || 200/368 ||                 loss : 1.286, acc : 0.641\n",
            "Epoch : 6 || 250/368 ||                 loss : 1.313, acc : 0.656\n",
            "Epoch : 6 || 300/368 ||                 loss : 1.244, acc : 0.609\n",
            "Epoch : 6 || 350/368 ||                 loss : 1.219, acc : 0.672\n",
            "Validation result of epoch 6/100 || val_loss : 1.330 val_acc : 0.617 \n",
            "Epoch : 7 || 0/368 ||                 loss : 1.234, acc : 0.672\n",
            "Epoch : 7 || 50/368 ||                 loss : 1.119, acc : 0.703\n",
            "Epoch : 7 || 100/368 ||                 loss : 1.140, acc : 0.641\n",
            "Epoch : 7 || 150/368 ||                 loss : 1.572, acc : 0.500\n",
            "Epoch : 7 || 200/368 ||                 loss : 1.142, acc : 0.625\n",
            "Epoch : 7 || 250/368 ||                 loss : 1.248, acc : 0.641\n",
            "Epoch : 7 || 300/368 ||                 loss : 1.256, acc : 0.625\n",
            "Epoch : 7 || 350/368 ||                 loss : 1.155, acc : 0.703\n",
            "Validation result of epoch 7/100 || val_loss : 1.318 val_acc : 0.614 \n",
            "Epoch : 8 || 0/368 ||                 loss : 1.007, acc : 0.734\n",
            "Epoch : 8 || 50/368 ||                 loss : 1.253, acc : 0.609\n",
            "Epoch : 8 || 100/368 ||                 loss : 1.126, acc : 0.672\n",
            "Epoch : 8 || 150/368 ||                 loss : 1.388, acc : 0.547\n",
            "Epoch : 8 || 200/368 ||                 loss : 1.291, acc : 0.609\n",
            "Epoch : 8 || 250/368 ||                 loss : 1.234, acc : 0.625\n",
            "Epoch : 8 || 300/368 ||                 loss : 1.313, acc : 0.656\n",
            "Epoch : 8 || 350/368 ||                 loss : 1.351, acc : 0.578\n",
            "Validation result of epoch 8/100 || val_loss : 1.301 val_acc : 0.626 \n",
            "Epoch : 9 || 0/368 ||                 loss : 1.360, acc : 0.562\n",
            "Epoch : 9 || 50/368 ||                 loss : 1.094, acc : 0.672\n",
            "Epoch : 9 || 100/368 ||                 loss : 1.475, acc : 0.578\n",
            "Epoch : 9 || 150/368 ||                 loss : 1.157, acc : 0.688\n",
            "Epoch : 9 || 200/368 ||                 loss : 1.224, acc : 0.594\n",
            "Epoch : 9 || 250/368 ||                 loss : 1.171, acc : 0.672\n",
            "Epoch : 9 || 300/368 ||                 loss : 1.155, acc : 0.656\n",
            "Epoch : 9 || 350/368 ||                 loss : 1.014, acc : 0.719\n",
            "Validation result of epoch 9/100 || val_loss : 1.292 val_acc : 0.623 \n",
            "Epoch : 10 || 0/368 ||                 loss : 1.035, acc : 0.734\n",
            "Epoch : 10 || 50/368 ||                 loss : 1.051, acc : 0.734\n",
            "Epoch : 10 || 100/368 ||                 loss : 1.031, acc : 0.672\n",
            "Epoch : 10 || 150/368 ||                 loss : 1.046, acc : 0.734\n",
            "Epoch : 10 || 200/368 ||                 loss : 1.676, acc : 0.500\n",
            "Epoch : 10 || 250/368 ||                 loss : 0.955, acc : 0.781\n",
            "Epoch : 10 || 300/368 ||                 loss : 1.163, acc : 0.641\n",
            "Epoch : 10 || 350/368 ||                 loss : 1.144, acc : 0.641\n",
            "Validation result of epoch 10/100 || val_loss : 1.280 val_acc : 0.629 \n",
            "Epoch : 11 || 0/368 ||                 loss : 1.124, acc : 0.656\n",
            "Epoch : 11 || 50/368 ||                 loss : 1.050, acc : 0.703\n",
            "Epoch : 11 || 100/368 ||                 loss : 1.075, acc : 0.672\n",
            "Epoch : 11 || 150/368 ||                 loss : 1.267, acc : 0.641\n",
            "Epoch : 11 || 200/368 ||                 loss : 1.498, acc : 0.578\n",
            "Epoch : 11 || 250/368 ||                 loss : 1.077, acc : 0.719\n",
            "Epoch : 11 || 300/368 ||                 loss : 1.433, acc : 0.594\n",
            "Epoch : 11 || 350/368 ||                 loss : 1.492, acc : 0.594\n",
            "Validation result of epoch 11/100 || val_loss : 1.271 val_acc : 0.632 \n",
            "Epoch : 12 || 0/368 ||                 loss : 1.094, acc : 0.688\n",
            "Epoch : 12 || 50/368 ||                 loss : 1.397, acc : 0.609\n",
            "Epoch : 12 || 100/368 ||                 loss : 1.220, acc : 0.656\n",
            "Epoch : 12 || 150/368 ||                 loss : 1.238, acc : 0.625\n",
            "Epoch : 12 || 200/368 ||                 loss : 1.479, acc : 0.625\n",
            "Epoch : 12 || 250/368 ||                 loss : 1.292, acc : 0.625\n",
            "Epoch : 12 || 300/368 ||                 loss : 1.358, acc : 0.516\n",
            "Epoch : 12 || 350/368 ||                 loss : 1.296, acc : 0.672\n",
            "Validation result of epoch 12/100 || val_loss : 1.259 val_acc : 0.637 \n",
            "Epoch : 13 || 0/368 ||                 loss : 1.029, acc : 0.734\n",
            "Epoch : 13 || 50/368 ||                 loss : 1.390, acc : 0.625\n",
            "Epoch : 13 || 100/368 ||                 loss : 1.421, acc : 0.609\n",
            "Epoch : 13 || 150/368 ||                 loss : 1.440, acc : 0.531\n",
            "Epoch : 13 || 200/368 ||                 loss : 1.219, acc : 0.641\n",
            "Epoch : 13 || 250/368 ||                 loss : 0.863, acc : 0.766\n",
            "Epoch : 13 || 300/368 ||                 loss : 1.221, acc : 0.672\n",
            "Epoch : 13 || 350/368 ||                 loss : 1.345, acc : 0.656\n",
            "Validation result of epoch 13/100 || val_loss : 1.258 val_acc : 0.636 \n",
            "Epoch : 14 || 0/368 ||                 loss : 1.057, acc : 0.625\n",
            "Epoch : 14 || 50/368 ||                 loss : 0.959, acc : 0.734\n",
            "Epoch : 14 || 100/368 ||                 loss : 0.834, acc : 0.688\n",
            "Epoch : 14 || 150/368 ||                 loss : 0.994, acc : 0.703\n",
            "Epoch : 14 || 200/368 ||                 loss : 1.181, acc : 0.688\n",
            "Epoch : 14 || 250/368 ||                 loss : 1.320, acc : 0.562\n",
            "Epoch : 14 || 300/368 ||                 loss : 1.108, acc : 0.609\n",
            "Epoch : 14 || 350/368 ||                 loss : 1.367, acc : 0.625\n",
            "Validation result of epoch 14/100 || val_loss : 1.253 val_acc : 0.640 \n",
            "Epoch : 15 || 0/368 ||                 loss : 1.180, acc : 0.578\n",
            "Epoch : 15 || 50/368 ||                 loss : 0.902, acc : 0.797\n",
            "Epoch : 15 || 100/368 ||                 loss : 1.098, acc : 0.641\n",
            "Epoch : 15 || 150/368 ||                 loss : 1.074, acc : 0.688\n",
            "Epoch : 15 || 200/368 ||                 loss : 1.068, acc : 0.656\n",
            "Epoch : 15 || 250/368 ||                 loss : 1.252, acc : 0.641\n",
            "Epoch : 15 || 300/368 ||                 loss : 0.998, acc : 0.688\n",
            "Epoch : 15 || 350/368 ||                 loss : 0.925, acc : 0.656\n",
            "Validation result of epoch 15/100 || val_loss : 1.253 val_acc : 0.637 \n",
            "Epoch : 16 || 0/368 ||                 loss : 1.046, acc : 0.734\n",
            "Epoch : 16 || 50/368 ||                 loss : 0.971, acc : 0.750\n",
            "Epoch : 16 || 100/368 ||                 loss : 0.997, acc : 0.719\n",
            "Epoch : 16 || 150/368 ||                 loss : 0.926, acc : 0.766\n",
            "Epoch : 16 || 200/368 ||                 loss : 0.951, acc : 0.703\n",
            "Epoch : 16 || 250/368 ||                 loss : 0.818, acc : 0.812\n",
            "Epoch : 16 || 300/368 ||                 loss : 0.889, acc : 0.688\n",
            "Epoch : 16 || 350/368 ||                 loss : 0.895, acc : 0.750\n",
            "Validation result of epoch 16/100 || val_loss : 1.249 val_acc : 0.639 \n",
            "Epoch : 17 || 0/368 ||                 loss : 0.885, acc : 0.734\n",
            "Epoch : 17 || 50/368 ||                 loss : 0.868, acc : 0.703\n",
            "Epoch : 17 || 100/368 ||                 loss : 1.071, acc : 0.688\n",
            "Epoch : 17 || 150/368 ||                 loss : 0.889, acc : 0.734\n",
            "Epoch : 17 || 200/368 ||                 loss : 0.827, acc : 0.734\n",
            "Epoch : 17 || 250/368 ||                 loss : 1.044, acc : 0.703\n",
            "Epoch : 17 || 300/368 ||                 loss : 1.153, acc : 0.641\n",
            "Epoch : 17 || 350/368 ||                 loss : 1.247, acc : 0.609\n",
            "Validation result of epoch 17/100 || val_loss : 1.244 val_acc : 0.641 \n",
            "Epoch : 18 || 0/368 ||                 loss : 1.032, acc : 0.672\n",
            "Epoch : 18 || 50/368 ||                 loss : 0.935, acc : 0.734\n",
            "Epoch : 18 || 100/368 ||                 loss : 0.920, acc : 0.734\n",
            "Epoch : 18 || 150/368 ||                 loss : 1.094, acc : 0.625\n",
            "Epoch : 18 || 200/368 ||                 loss : 1.228, acc : 0.547\n",
            "Epoch : 18 || 250/368 ||                 loss : 0.903, acc : 0.719\n",
            "Epoch : 18 || 300/368 ||                 loss : 1.124, acc : 0.656\n",
            "Epoch : 18 || 350/368 ||                 loss : 1.069, acc : 0.703\n",
            "Validation result of epoch 18/100 || val_loss : 1.243 val_acc : 0.638 \n",
            "Epoch : 19 || 0/368 ||                 loss : 1.077, acc : 0.609\n",
            "Epoch : 19 || 50/368 ||                 loss : 0.882, acc : 0.703\n",
            "Epoch : 19 || 100/368 ||                 loss : 1.224, acc : 0.625\n",
            "Epoch : 19 || 150/368 ||                 loss : 1.097, acc : 0.672\n",
            "Epoch : 19 || 200/368 ||                 loss : 1.236, acc : 0.672\n",
            "Epoch : 19 || 250/368 ||                 loss : 1.172, acc : 0.641\n",
            "Epoch : 19 || 300/368 ||                 loss : 1.027, acc : 0.641\n",
            "Epoch : 19 || 350/368 ||                 loss : 0.792, acc : 0.734\n",
            "Validation result of epoch 19/100 || val_loss : 1.235 val_acc : 0.639 \n",
            "Epoch : 20 || 0/368 ||                 loss : 1.031, acc : 0.703\n",
            "Epoch : 20 || 50/368 ||                 loss : 0.943, acc : 0.750\n",
            "Epoch : 20 || 100/368 ||                 loss : 1.078, acc : 0.656\n",
            "Epoch : 20 || 150/368 ||                 loss : 1.082, acc : 0.703\n",
            "Epoch : 20 || 200/368 ||                 loss : 0.938, acc : 0.672\n",
            "Epoch : 20 || 250/368 ||                 loss : 1.013, acc : 0.734\n",
            "Epoch : 20 || 300/368 ||                 loss : 0.917, acc : 0.719\n",
            "Epoch : 20 || 350/368 ||                 loss : 0.994, acc : 0.688\n",
            "Validation result of epoch 20/100 || val_loss : 1.240 val_acc : 0.645 \n",
            "Epoch : 21 || 0/368 ||                 loss : 0.564, acc : 0.812\n",
            "Epoch : 21 || 50/368 ||                 loss : 0.829, acc : 0.766\n",
            "Epoch : 21 || 100/368 ||                 loss : 1.109, acc : 0.688\n",
            "Epoch : 21 || 150/368 ||                 loss : 0.905, acc : 0.688\n",
            "Epoch : 21 || 200/368 ||                 loss : 1.232, acc : 0.641\n",
            "Epoch : 21 || 250/368 ||                 loss : 0.868, acc : 0.797\n",
            "Epoch : 21 || 300/368 ||                 loss : 1.107, acc : 0.656\n",
            "Epoch : 21 || 350/368 ||                 loss : 1.046, acc : 0.609\n",
            "Validation result of epoch 21/100 || val_loss : 1.235 val_acc : 0.644 \n",
            "Epoch : 22 || 0/368 ||                 loss : 1.067, acc : 0.641\n",
            "Epoch : 22 || 50/368 ||                 loss : 0.995, acc : 0.719\n",
            "Epoch : 22 || 100/368 ||                 loss : 0.868, acc : 0.734\n",
            "Epoch : 22 || 150/368 ||                 loss : 0.741, acc : 0.750\n",
            "Epoch : 22 || 200/368 ||                 loss : 0.715, acc : 0.766\n",
            "Epoch : 22 || 250/368 ||                 loss : 0.784, acc : 0.797\n",
            "Epoch : 22 || 300/368 ||                 loss : 0.994, acc : 0.672\n",
            "Epoch : 22 || 350/368 ||                 loss : 1.068, acc : 0.719\n",
            "Validation result of epoch 22/100 || val_loss : 1.241 val_acc : 0.643 \n",
            "Epoch : 23 || 0/368 ||                 loss : 0.735, acc : 0.766\n",
            "Epoch : 23 || 50/368 ||                 loss : 0.641, acc : 0.844\n",
            "Epoch : 23 || 100/368 ||                 loss : 0.941, acc : 0.688\n",
            "Epoch : 23 || 150/368 ||                 loss : 1.151, acc : 0.625\n",
            "Epoch : 23 || 200/368 ||                 loss : 0.934, acc : 0.797\n",
            "Epoch : 23 || 250/368 ||                 loss : 0.900, acc : 0.703\n",
            "Epoch : 23 || 300/368 ||                 loss : 0.812, acc : 0.750\n",
            "Epoch : 23 || 350/368 ||                 loss : 0.877, acc : 0.797\n",
            "Validation result of epoch 23/100 || val_loss : 1.241 val_acc : 0.640 \n",
            "Epoch : 24 || 0/368 ||                 loss : 0.787, acc : 0.781\n",
            "Epoch : 24 || 50/368 ||                 loss : 1.009, acc : 0.688\n",
            "Epoch : 24 || 100/368 ||                 loss : 0.961, acc : 0.688\n",
            "Epoch : 24 || 150/368 ||                 loss : 0.989, acc : 0.703\n",
            "Epoch : 24 || 200/368 ||                 loss : 1.185, acc : 0.656\n",
            "Epoch : 24 || 250/368 ||                 loss : 0.804, acc : 0.734\n",
            "Epoch : 24 || 300/368 ||                 loss : 1.110, acc : 0.703\n",
            "Epoch : 24 || 350/368 ||                 loss : 0.679, acc : 0.812\n",
            "Validation result of epoch 24/100 || val_loss : 1.245 val_acc : 0.642 \n",
            "Epoch : 25 || 0/368 ||                 loss : 0.815, acc : 0.750\n",
            "Epoch : 25 || 50/368 ||                 loss : 1.202, acc : 0.656\n",
            "Epoch : 25 || 100/368 ||                 loss : 0.898, acc : 0.734\n",
            "Epoch : 25 || 150/368 ||                 loss : 0.806, acc : 0.781\n",
            "Epoch : 25 || 200/368 ||                 loss : 0.988, acc : 0.719\n",
            "Epoch : 25 || 250/368 ||                 loss : 0.934, acc : 0.703\n",
            "Epoch : 25 || 300/368 ||                 loss : 0.919, acc : 0.750\n",
            "Epoch : 25 || 350/368 ||                 loss : 1.028, acc : 0.688\n",
            "Validation result of epoch 25/100 || val_loss : 1.253 val_acc : 0.639 \n",
            "Epoch : 26 || 0/368 ||                 loss : 0.738, acc : 0.734\n",
            "Epoch : 26 || 50/368 ||                 loss : 0.935, acc : 0.734\n",
            "Epoch : 26 || 100/368 ||                 loss : 1.280, acc : 0.594\n",
            "Epoch : 26 || 150/368 ||                 loss : 1.148, acc : 0.641\n",
            "Epoch : 26 || 200/368 ||                 loss : 0.930, acc : 0.719\n",
            "Epoch : 26 || 250/368 ||                 loss : 1.114, acc : 0.625\n",
            "Epoch : 26 || 300/368 ||                 loss : 0.753, acc : 0.781\n",
            "Epoch : 26 || 350/368 ||                 loss : 0.976, acc : 0.688\n",
            "Validation result of epoch 26/100 || val_loss : 1.231 val_acc : 0.644 \n",
            "Epoch : 27 || 0/368 ||                 loss : 1.216, acc : 0.625\n",
            "Epoch : 27 || 50/368 ||                 loss : 1.001, acc : 0.609\n",
            "Epoch : 27 || 100/368 ||                 loss : 0.760, acc : 0.734\n",
            "Epoch : 27 || 150/368 ||                 loss : 1.090, acc : 0.719\n",
            "Epoch : 27 || 200/368 ||                 loss : 0.929, acc : 0.703\n",
            "Epoch : 27 || 250/368 ||                 loss : 0.879, acc : 0.719\n",
            "Epoch : 27 || 300/368 ||                 loss : 0.905, acc : 0.703\n",
            "Epoch : 27 || 350/368 ||                 loss : 0.880, acc : 0.688\n",
            "Validation result of epoch 27/100 || val_loss : 1.248 val_acc : 0.641 \n",
            "Epoch : 28 || 0/368 ||                 loss : 0.705, acc : 0.797\n",
            "Epoch : 28 || 50/368 ||                 loss : 0.727, acc : 0.703\n",
            "Epoch : 28 || 100/368 ||                 loss : 0.827, acc : 0.703\n",
            "Epoch : 28 || 150/368 ||                 loss : 1.013, acc : 0.672\n",
            "Epoch : 28 || 200/368 ||                 loss : 0.696, acc : 0.812\n",
            "Epoch : 28 || 250/368 ||                 loss : 0.941, acc : 0.672\n",
            "Epoch : 28 || 300/368 ||                 loss : 0.838, acc : 0.641\n",
            "Epoch : 28 || 350/368 ||                 loss : 0.723, acc : 0.828\n",
            "Validation result of epoch 28/100 || val_loss : 1.244 val_acc : 0.646 \n",
            "Epoch : 29 || 0/368 ||                 loss : 0.913, acc : 0.641\n",
            "Epoch : 29 || 50/368 ||                 loss : 0.851, acc : 0.750\n",
            "Epoch : 29 || 100/368 ||                 loss : 0.969, acc : 0.734\n",
            "Epoch : 29 || 150/368 ||                 loss : 0.837, acc : 0.812\n",
            "Epoch : 29 || 200/368 ||                 loss : 0.940, acc : 0.672\n",
            "Epoch : 29 || 250/368 ||                 loss : 1.064, acc : 0.641\n",
            "Epoch : 29 || 300/368 ||                 loss : 0.615, acc : 0.812\n",
            "Epoch : 29 || 350/368 ||                 loss : 1.027, acc : 0.734\n",
            "Validation result of epoch 29/100 || val_loss : 1.253 val_acc : 0.643 \n",
            "Epoch : 30 || 0/368 ||                 loss : 0.778, acc : 0.750\n",
            "Epoch : 30 || 50/368 ||                 loss : 0.696, acc : 0.844\n",
            "Epoch : 30 || 100/368 ||                 loss : 1.125, acc : 0.609\n",
            "Epoch : 30 || 150/368 ||                 loss : 0.807, acc : 0.750\n",
            "Epoch : 30 || 200/368 ||                 loss : 0.995, acc : 0.766\n",
            "Epoch : 30 || 250/368 ||                 loss : 0.901, acc : 0.750\n",
            "Epoch : 30 || 300/368 ||                 loss : 0.670, acc : 0.766\n",
            "Epoch : 30 || 350/368 ||                 loss : 1.123, acc : 0.672\n",
            "Validation result of epoch 30/100 || val_loss : 1.242 val_acc : 0.644 \n",
            "Epoch : 31 || 0/368 ||                 loss : 0.578, acc : 0.844\n",
            "Epoch : 31 || 50/368 ||                 loss : 0.840, acc : 0.797\n",
            "Epoch : 31 || 100/368 ||                 loss : 1.053, acc : 0.703\n",
            "Epoch : 31 || 150/368 ||                 loss : 0.748, acc : 0.734\n",
            "Epoch : 31 || 200/368 ||                 loss : 0.940, acc : 0.703\n",
            "Epoch : 31 || 250/368 ||                 loss : 1.051, acc : 0.688\n",
            "Epoch : 31 || 300/368 ||                 loss : 0.785, acc : 0.797\n",
            "Epoch : 31 || 350/368 ||                 loss : 0.818, acc : 0.688\n",
            "Validation result of epoch 31/100 || val_loss : 1.252 val_acc : 0.644 \n",
            "Epoch : 32 || 0/368 ||                 loss : 0.900, acc : 0.797\n",
            "Epoch : 32 || 50/368 ||                 loss : 0.911, acc : 0.750\n",
            "Epoch : 32 || 100/368 ||                 loss : 0.856, acc : 0.719\n",
            "Epoch : 32 || 150/368 ||                 loss : 0.663, acc : 0.781\n",
            "Epoch : 32 || 200/368 ||                 loss : 0.772, acc : 0.797\n",
            "Epoch : 32 || 250/368 ||                 loss : 0.792, acc : 0.750\n",
            "Epoch : 32 || 300/368 ||                 loss : 0.851, acc : 0.672\n",
            "Epoch : 32 || 350/368 ||                 loss : 1.030, acc : 0.703\n",
            "Validation result of epoch 32/100 || val_loss : 1.254 val_acc : 0.644 \n",
            "Epoch : 33 || 0/368 ||                 loss : 1.173, acc : 0.672\n",
            "Epoch : 33 || 50/368 ||                 loss : 0.924, acc : 0.781\n",
            "Epoch : 33 || 100/368 ||                 loss : 0.824, acc : 0.766\n",
            "Epoch : 33 || 150/368 ||                 loss : 1.134, acc : 0.703\n",
            "Epoch : 33 || 200/368 ||                 loss : 0.896, acc : 0.703\n",
            "Epoch : 33 || 250/368 ||                 loss : 1.087, acc : 0.641\n",
            "Epoch : 33 || 300/368 ||                 loss : 0.834, acc : 0.672\n",
            "Epoch : 33 || 350/368 ||                 loss : 0.724, acc : 0.750\n",
            "Validation result of epoch 33/100 || val_loss : 1.251 val_acc : 0.644 \n",
            "Epoch : 34 || 0/368 ||                 loss : 0.791, acc : 0.672\n",
            "Epoch : 34 || 50/368 ||                 loss : 0.762, acc : 0.734\n",
            "Epoch : 34 || 100/368 ||                 loss : 0.703, acc : 0.844\n",
            "Epoch : 34 || 150/368 ||                 loss : 0.771, acc : 0.766\n",
            "Epoch : 34 || 200/368 ||                 loss : 0.837, acc : 0.703\n",
            "Epoch : 34 || 250/368 ||                 loss : 1.052, acc : 0.641\n",
            "Epoch : 34 || 300/368 ||                 loss : 0.871, acc : 0.750\n",
            "Epoch : 34 || 350/368 ||                 loss : 0.791, acc : 0.781\n",
            "Validation result of epoch 34/100 || val_loss : 1.248 val_acc : 0.643 \n",
            "Epoch : 35 || 0/368 ||                 loss : 0.667, acc : 0.781\n",
            "Epoch : 35 || 50/368 ||                 loss : 0.980, acc : 0.625\n",
            "Epoch : 35 || 100/368 ||                 loss : 0.679, acc : 0.797\n",
            "Epoch : 35 || 150/368 ||                 loss : 0.644, acc : 0.828\n",
            "Epoch : 35 || 200/368 ||                 loss : 0.937, acc : 0.672\n",
            "Epoch : 35 || 250/368 ||                 loss : 0.996, acc : 0.688\n",
            "Epoch : 35 || 300/368 ||                 loss : 1.002, acc : 0.672\n",
            "Epoch : 35 || 350/368 ||                 loss : 0.880, acc : 0.688\n",
            "Validation result of epoch 35/100 || val_loss : 1.249 val_acc : 0.639 \n",
            "Epoch : 36 || 0/368 ||                 loss : 0.948, acc : 0.750\n",
            "Epoch : 36 || 50/368 ||                 loss : 0.850, acc : 0.750\n",
            "Epoch : 36 || 100/368 ||                 loss : 0.732, acc : 0.797\n",
            "Epoch : 36 || 150/368 ||                 loss : 0.883, acc : 0.719\n",
            "Epoch : 36 || 200/368 ||                 loss : 0.863, acc : 0.750\n",
            "Epoch : 36 || 250/368 ||                 loss : 0.833, acc : 0.688\n",
            "Epoch : 36 || 300/368 ||                 loss : 1.167, acc : 0.672\n",
            "Epoch : 36 || 350/368 ||                 loss : 0.891, acc : 0.688\n",
            "Validation result of epoch 36/100 || val_loss : 1.248 val_acc : 0.643 \n",
            "Epoch : 37 || 0/368 ||                 loss : 1.095, acc : 0.609\n",
            "Epoch : 37 || 50/368 ||                 loss : 0.783, acc : 0.781\n",
            "Epoch : 37 || 100/368 ||                 loss : 0.790, acc : 0.766\n",
            "Epoch : 37 || 150/368 ||                 loss : 1.186, acc : 0.641\n",
            "Epoch : 37 || 200/368 ||                 loss : 0.838, acc : 0.734\n",
            "Epoch : 37 || 250/368 ||                 loss : 0.766, acc : 0.781\n",
            "Epoch : 37 || 300/368 ||                 loss : 0.851, acc : 0.750\n",
            "Epoch : 37 || 350/368 ||                 loss : 0.726, acc : 0.797\n",
            "Validation result of epoch 37/100 || val_loss : 1.262 val_acc : 0.640 \n",
            "Epoch : 38 || 0/368 ||                 loss : 0.823, acc : 0.750\n",
            "Epoch : 38 || 50/368 ||                 loss : 1.151, acc : 0.578\n",
            "Epoch : 38 || 100/368 ||                 loss : 0.752, acc : 0.781\n",
            "Epoch : 38 || 150/368 ||                 loss : 0.876, acc : 0.750\n",
            "Epoch : 38 || 200/368 ||                 loss : 0.908, acc : 0.656\n",
            "Epoch : 38 || 250/368 ||                 loss : 0.660, acc : 0.781\n",
            "Epoch : 38 || 300/368 ||                 loss : 1.020, acc : 0.750\n",
            "Epoch : 38 || 350/368 ||                 loss : 0.929, acc : 0.703\n",
            "Validation result of epoch 38/100 || val_loss : 1.256 val_acc : 0.647 \n",
            "Epoch : 39 || 0/368 ||                 loss : 0.709, acc : 0.812\n",
            "Epoch : 39 || 50/368 ||                 loss : 0.800, acc : 0.797\n",
            "Epoch : 39 || 100/368 ||                 loss : 0.967, acc : 0.688\n",
            "Epoch : 39 || 150/368 ||                 loss : 0.958, acc : 0.703\n",
            "Epoch : 39 || 200/368 ||                 loss : 0.706, acc : 0.766\n",
            "Epoch : 39 || 250/368 ||                 loss : 0.791, acc : 0.719\n",
            "Epoch : 39 || 300/368 ||                 loss : 1.070, acc : 0.656\n",
            "Epoch : 39 || 350/368 ||                 loss : 0.718, acc : 0.797\n",
            "Validation result of epoch 39/100 || val_loss : 1.263 val_acc : 0.643 \n",
            "Epoch : 40 || 0/368 ||                 loss : 1.163, acc : 0.625\n",
            "Epoch : 40 || 50/368 ||                 loss : 0.738, acc : 0.719\n",
            "Epoch : 40 || 100/368 ||                 loss : 0.785, acc : 0.797\n",
            "Epoch : 40 || 150/368 ||                 loss : 0.855, acc : 0.750\n",
            "Epoch : 40 || 200/368 ||                 loss : 0.826, acc : 0.719\n",
            "Epoch : 40 || 250/368 ||                 loss : 0.956, acc : 0.703\n",
            "Epoch : 40 || 300/368 ||                 loss : 0.666, acc : 0.734\n",
            "Epoch : 40 || 350/368 ||                 loss : 0.667, acc : 0.812\n",
            "Validation result of epoch 40/100 || val_loss : 1.263 val_acc : 0.642 \n",
            "Epoch : 41 || 0/368 ||                 loss : 0.648, acc : 0.797\n",
            "Epoch : 41 || 50/368 ||                 loss : 0.799, acc : 0.781\n",
            "Epoch : 41 || 100/368 ||                 loss : 0.791, acc : 0.750\n",
            "Epoch : 41 || 150/368 ||                 loss : 0.623, acc : 0.812\n",
            "Epoch : 41 || 200/368 ||                 loss : 0.710, acc : 0.812\n",
            "Epoch : 41 || 250/368 ||                 loss : 0.994, acc : 0.672\n",
            "Epoch : 41 || 300/368 ||                 loss : 0.877, acc : 0.719\n",
            "Epoch : 41 || 350/368 ||                 loss : 0.947, acc : 0.703\n",
            "Validation result of epoch 41/100 || val_loss : 1.270 val_acc : 0.641 \n",
            "Epoch : 42 || 0/368 ||                 loss : 0.928, acc : 0.734\n",
            "Epoch : 42 || 50/368 ||                 loss : 0.663, acc : 0.781\n",
            "Epoch : 42 || 100/368 ||                 loss : 0.814, acc : 0.719\n",
            "Epoch : 42 || 150/368 ||                 loss : 0.651, acc : 0.781\n",
            "Epoch : 42 || 200/368 ||                 loss : 0.925, acc : 0.703\n",
            "Epoch : 42 || 250/368 ||                 loss : 0.963, acc : 0.688\n",
            "Epoch : 42 || 300/368 ||                 loss : 0.795, acc : 0.734\n",
            "Epoch : 42 || 350/368 ||                 loss : 0.787, acc : 0.781\n",
            "Validation result of epoch 42/100 || val_loss : 1.275 val_acc : 0.639 \n",
            "Epoch : 43 || 0/368 ||                 loss : 0.799, acc : 0.750\n",
            "Epoch : 43 || 50/368 ||                 loss : 0.843, acc : 0.703\n",
            "Epoch : 43 || 100/368 ||                 loss : 0.857, acc : 0.781\n",
            "Epoch : 43 || 150/368 ||                 loss : 0.776, acc : 0.812\n",
            "Epoch : 43 || 200/368 ||                 loss : 0.920, acc : 0.703\n",
            "Epoch : 43 || 250/368 ||                 loss : 0.514, acc : 0.859\n",
            "Epoch : 43 || 300/368 ||                 loss : 0.986, acc : 0.672\n",
            "Epoch : 43 || 350/368 ||                 loss : 0.785, acc : 0.719\n",
            "Validation result of epoch 43/100 || val_loss : 1.269 val_acc : 0.643 \n",
            "Epoch : 44 || 0/368 ||                 loss : 0.855, acc : 0.719\n",
            "Epoch : 44 || 50/368 ||                 loss : 0.709, acc : 0.781\n",
            "Epoch : 44 || 100/368 ||                 loss : 0.827, acc : 0.719\n",
            "Epoch : 44 || 150/368 ||                 loss : 0.934, acc : 0.672\n",
            "Epoch : 44 || 200/368 ||                 loss : 0.710, acc : 0.750\n",
            "Epoch : 44 || 250/368 ||                 loss : 1.158, acc : 0.688\n",
            "Epoch : 44 || 300/368 ||                 loss : 1.105, acc : 0.625\n",
            "Epoch : 44 || 350/368 ||                 loss : 1.073, acc : 0.641\n",
            "Validation result of epoch 44/100 || val_loss : 1.264 val_acc : 0.642 \n",
            "Epoch : 45 || 0/368 ||                 loss : 0.569, acc : 0.828\n",
            "Epoch : 45 || 50/368 ||                 loss : 0.922, acc : 0.750\n",
            "Epoch : 45 || 100/368 ||                 loss : 0.677, acc : 0.781\n",
            "Epoch : 45 || 150/368 ||                 loss : 0.922, acc : 0.719\n",
            "Epoch : 45 || 200/368 ||                 loss : 0.799, acc : 0.781\n",
            "Epoch : 45 || 250/368 ||                 loss : 0.695, acc : 0.797\n",
            "Epoch : 45 || 300/368 ||                 loss : 0.942, acc : 0.750\n",
            "Epoch : 45 || 350/368 ||                 loss : 0.809, acc : 0.797\n",
            "Validation result of epoch 45/100 || val_loss : 1.255 val_acc : 0.643 \n",
            "Epoch : 46 || 0/368 ||                 loss : 0.643, acc : 0.797\n",
            "Epoch : 46 || 50/368 ||                 loss : 0.773, acc : 0.734\n",
            "Epoch : 46 || 100/368 ||                 loss : 1.020, acc : 0.703\n",
            "Epoch : 46 || 150/368 ||                 loss : 0.663, acc : 0.781\n",
            "Epoch : 46 || 200/368 ||                 loss : 0.644, acc : 0.828\n",
            "Epoch : 46 || 250/368 ||                 loss : 0.823, acc : 0.766\n",
            "Epoch : 46 || 300/368 ||                 loss : 0.790, acc : 0.766\n",
            "Epoch : 46 || 350/368 ||                 loss : 0.857, acc : 0.703\n",
            "Validation result of epoch 46/100 || val_loss : 1.260 val_acc : 0.647 \n",
            "Epoch : 47 || 0/368 ||                 loss : 0.923, acc : 0.656\n",
            "Epoch : 47 || 50/368 ||                 loss : 1.049, acc : 0.703\n",
            "Epoch : 47 || 100/368 ||                 loss : 0.811, acc : 0.688\n",
            "Epoch : 47 || 150/368 ||                 loss : 0.687, acc : 0.781\n",
            "Epoch : 47 || 200/368 ||                 loss : 0.651, acc : 0.781\n",
            "Epoch : 47 || 250/368 ||                 loss : 0.679, acc : 0.797\n",
            "Epoch : 47 || 300/368 ||                 loss : 1.085, acc : 0.656\n",
            "Epoch : 47 || 350/368 ||                 loss : 0.985, acc : 0.688\n",
            "Validation result of epoch 47/100 || val_loss : 1.262 val_acc : 0.645 \n",
            "Epoch : 48 || 0/368 ||                 loss : 0.777, acc : 0.797\n",
            "Epoch : 48 || 50/368 ||                 loss : 0.653, acc : 0.766\n",
            "Epoch : 48 || 100/368 ||                 loss : 0.831, acc : 0.781\n",
            "Epoch : 48 || 150/368 ||                 loss : 1.073, acc : 0.672\n",
            "Epoch : 48 || 200/368 ||                 loss : 0.851, acc : 0.750\n",
            "Epoch : 48 || 250/368 ||                 loss : 1.254, acc : 0.656\n",
            "Epoch : 48 || 300/368 ||                 loss : 0.843, acc : 0.688\n",
            "Epoch : 48 || 350/368 ||                 loss : 0.896, acc : 0.750\n",
            "Validation result of epoch 48/100 || val_loss : 1.269 val_acc : 0.645 \n",
            "Epoch : 49 || 0/368 ||                 loss : 0.634, acc : 0.781\n",
            "Epoch : 49 || 50/368 ||                 loss : 0.602, acc : 0.781\n",
            "Epoch : 49 || 100/368 ||                 loss : 0.920, acc : 0.625\n",
            "Epoch : 49 || 150/368 ||                 loss : 0.995, acc : 0.703\n",
            "Epoch : 49 || 200/368 ||                 loss : 1.110, acc : 0.734\n",
            "Epoch : 49 || 250/368 ||                 loss : 0.824, acc : 0.766\n",
            "Epoch : 49 || 300/368 ||                 loss : 1.126, acc : 0.656\n",
            "Epoch : 49 || 350/368 ||                 loss : 0.853, acc : 0.734\n",
            "Validation result of epoch 49/100 || val_loss : 1.263 val_acc : 0.644 \n",
            "Epoch : 50 || 0/368 ||                 loss : 0.888, acc : 0.719\n",
            "Epoch : 50 || 50/368 ||                 loss : 0.577, acc : 0.859\n",
            "Epoch : 50 || 100/368 ||                 loss : 0.634, acc : 0.781\n",
            "Epoch : 50 || 150/368 ||                 loss : 0.900, acc : 0.734\n",
            "Epoch : 50 || 200/368 ||                 loss : 0.682, acc : 0.766\n",
            "Epoch : 50 || 250/368 ||                 loss : 0.606, acc : 0.844\n",
            "Epoch : 50 || 300/368 ||                 loss : 0.843, acc : 0.688\n",
            "Epoch : 50 || 350/368 ||                 loss : 0.631, acc : 0.844\n",
            "Validation result of epoch 50/100 || val_loss : 1.253 val_acc : 0.651 \n",
            "Epoch : 51 || 0/368 ||                 loss : 0.749, acc : 0.812\n",
            "Epoch : 51 || 50/368 ||                 loss : 0.708, acc : 0.781\n",
            "Epoch : 51 || 100/368 ||                 loss : 0.804, acc : 0.750\n",
            "Epoch : 51 || 150/368 ||                 loss : 0.851, acc : 0.750\n",
            "Epoch : 51 || 200/368 ||                 loss : 0.775, acc : 0.750\n",
            "Epoch : 51 || 250/368 ||                 loss : 0.796, acc : 0.766\n",
            "Epoch : 51 || 300/368 ||                 loss : 0.618, acc : 0.797\n",
            "Epoch : 51 || 350/368 ||                 loss : 0.716, acc : 0.750\n",
            "Validation result of epoch 51/100 || val_loss : 1.261 val_acc : 0.645 \n",
            "Epoch : 52 || 0/368 ||                 loss : 0.789, acc : 0.766\n",
            "Epoch : 52 || 50/368 ||                 loss : 0.742, acc : 0.812\n",
            "Epoch : 52 || 100/368 ||                 loss : 0.882, acc : 0.656\n",
            "Epoch : 52 || 150/368 ||                 loss : 0.799, acc : 0.734\n",
            "Epoch : 52 || 200/368 ||                 loss : 0.648, acc : 0.812\n",
            "Epoch : 52 || 250/368 ||                 loss : 0.793, acc : 0.766\n",
            "Epoch : 52 || 300/368 ||                 loss : 0.586, acc : 0.797\n",
            "Epoch : 52 || 350/368 ||                 loss : 0.635, acc : 0.781\n",
            "Validation result of epoch 52/100 || val_loss : 1.262 val_acc : 0.648 \n",
            "Epoch : 53 || 0/368 ||                 loss : 0.541, acc : 0.844\n",
            "Epoch : 53 || 50/368 ||                 loss : 0.981, acc : 0.750\n",
            "Epoch : 53 || 100/368 ||                 loss : 0.716, acc : 0.812\n",
            "Epoch : 53 || 150/368 ||                 loss : 0.617, acc : 0.797\n",
            "Epoch : 53 || 200/368 ||                 loss : 1.052, acc : 0.656\n",
            "Epoch : 53 || 250/368 ||                 loss : 0.621, acc : 0.781\n",
            "Epoch : 53 || 300/368 ||                 loss : 0.524, acc : 0.859\n",
            "Epoch : 53 || 350/368 ||                 loss : 0.762, acc : 0.766\n",
            "Validation result of epoch 53/100 || val_loss : 1.261 val_acc : 0.645 \n",
            "Epoch : 54 || 0/368 ||                 loss : 0.565, acc : 0.812\n",
            "Epoch : 54 || 50/368 ||                 loss : 0.805, acc : 0.734\n",
            "Epoch : 54 || 100/368 ||                 loss : 0.845, acc : 0.734\n",
            "Epoch : 54 || 150/368 ||                 loss : 0.645, acc : 0.812\n",
            "Epoch : 54 || 200/368 ||                 loss : 0.859, acc : 0.719\n",
            "Epoch : 54 || 250/368 ||                 loss : 0.713, acc : 0.750\n",
            "Epoch : 54 || 300/368 ||                 loss : 0.579, acc : 0.812\n",
            "Epoch : 54 || 350/368 ||                 loss : 0.571, acc : 0.812\n",
            "Validation result of epoch 54/100 || val_loss : 1.267 val_acc : 0.649 \n",
            "Epoch : 55 || 0/368 ||                 loss : 0.823, acc : 0.719\n",
            "Epoch : 55 || 50/368 ||                 loss : 0.689, acc : 0.828\n",
            "Epoch : 55 || 100/368 ||                 loss : 0.930, acc : 0.750\n",
            "Epoch : 55 || 150/368 ||                 loss : 1.061, acc : 0.719\n",
            "Epoch : 55 || 200/368 ||                 loss : 0.666, acc : 0.781\n",
            "Epoch : 55 || 250/368 ||                 loss : 0.708, acc : 0.797\n",
            "Epoch : 55 || 300/368 ||                 loss : 0.936, acc : 0.688\n",
            "Epoch : 55 || 350/368 ||                 loss : 0.700, acc : 0.797\n",
            "Validation result of epoch 55/100 || val_loss : 1.269 val_acc : 0.650 \n",
            "Epoch : 56 || 0/368 ||                 loss : 0.555, acc : 0.812\n",
            "Epoch : 56 || 50/368 ||                 loss : 1.002, acc : 0.703\n",
            "Epoch : 56 || 100/368 ||                 loss : 0.607, acc : 0.781\n",
            "Epoch : 56 || 150/368 ||                 loss : 0.679, acc : 0.766\n",
            "Epoch : 56 || 200/368 ||                 loss : 0.749, acc : 0.734\n",
            "Epoch : 56 || 250/368 ||                 loss : 0.872, acc : 0.688\n",
            "Epoch : 56 || 300/368 ||                 loss : 0.707, acc : 0.797\n",
            "Epoch : 56 || 350/368 ||                 loss : 0.606, acc : 0.781\n",
            "Validation result of epoch 56/100 || val_loss : 1.271 val_acc : 0.650 \n",
            "Epoch : 57 || 0/368 ||                 loss : 0.712, acc : 0.766\n",
            "Epoch : 57 || 50/368 ||                 loss : 0.805, acc : 0.734\n",
            "Epoch : 57 || 100/368 ||                 loss : 0.706, acc : 0.766\n",
            "Epoch : 57 || 150/368 ||                 loss : 0.736, acc : 0.750\n",
            "Epoch : 57 || 200/368 ||                 loss : 0.845, acc : 0.734\n",
            "Epoch : 57 || 250/368 ||                 loss : 0.647, acc : 0.781\n",
            "Epoch : 57 || 300/368 ||                 loss : 0.716, acc : 0.719\n",
            "Epoch : 57 || 350/368 ||                 loss : 0.779, acc : 0.703\n",
            "Validation result of epoch 57/100 || val_loss : 1.260 val_acc : 0.649 \n",
            "Epoch : 58 || 0/368 ||                 loss : 0.461, acc : 0.875\n",
            "Epoch : 58 || 50/368 ||                 loss : 0.605, acc : 0.812\n",
            "Epoch : 58 || 100/368 ||                 loss : 0.710, acc : 0.781\n",
            "Epoch : 58 || 150/368 ||                 loss : 0.716, acc : 0.797\n",
            "Epoch : 58 || 200/368 ||                 loss : 0.819, acc : 0.734\n",
            "Epoch : 58 || 250/368 ||                 loss : 0.743, acc : 0.766\n",
            "Epoch : 58 || 300/368 ||                 loss : 0.801, acc : 0.750\n",
            "Epoch : 58 || 350/368 ||                 loss : 0.580, acc : 0.781\n",
            "Validation result of epoch 58/100 || val_loss : 1.272 val_acc : 0.645 \n",
            "Epoch : 59 || 0/368 ||                 loss : 0.810, acc : 0.688\n",
            "Epoch : 59 || 50/368 ||                 loss : 0.746, acc : 0.750\n",
            "Epoch : 59 || 100/368 ||                 loss : 0.774, acc : 0.797\n",
            "Epoch : 59 || 150/368 ||                 loss : 0.597, acc : 0.797\n",
            "Epoch : 59 || 200/368 ||                 loss : 0.544, acc : 0.812\n",
            "Epoch : 59 || 250/368 ||                 loss : 0.791, acc : 0.766\n",
            "Epoch : 59 || 300/368 ||                 loss : 0.659, acc : 0.812\n",
            "Epoch : 59 || 350/368 ||                 loss : 0.609, acc : 0.797\n",
            "Validation result of epoch 59/100 || val_loss : 1.279 val_acc : 0.646 \n",
            "Epoch : 60 || 0/368 ||                 loss : 0.714, acc : 0.797\n",
            "Epoch : 60 || 50/368 ||                 loss : 0.767, acc : 0.734\n",
            "Epoch : 60 || 100/368 ||                 loss : 0.766, acc : 0.781\n",
            "Epoch : 60 || 150/368 ||                 loss : 0.888, acc : 0.641\n",
            "Epoch : 60 || 200/368 ||                 loss : 0.499, acc : 0.812\n",
            "Epoch : 60 || 250/368 ||                 loss : 0.678, acc : 0.812\n",
            "Epoch : 60 || 300/368 ||                 loss : 0.823, acc : 0.750\n",
            "Epoch : 60 || 350/368 ||                 loss : 0.642, acc : 0.828\n",
            "Validation result of epoch 60/100 || val_loss : 1.283 val_acc : 0.644 \n",
            "Epoch : 61 || 0/368 ||                 loss : 0.729, acc : 0.797\n",
            "Epoch : 61 || 50/368 ||                 loss : 0.600, acc : 0.844\n",
            "Epoch : 61 || 100/368 ||                 loss : 0.631, acc : 0.828\n",
            "Epoch : 61 || 150/368 ||                 loss : 0.989, acc : 0.641\n",
            "Epoch : 61 || 200/368 ||                 loss : 0.873, acc : 0.656\n",
            "Epoch : 61 || 250/368 ||                 loss : 0.702, acc : 0.812\n",
            "Epoch : 61 || 300/368 ||                 loss : 0.641, acc : 0.766\n",
            "Epoch : 61 || 350/368 ||                 loss : 0.765, acc : 0.766\n",
            "Validation result of epoch 61/100 || val_loss : 1.269 val_acc : 0.646 \n",
            "Epoch : 62 || 0/368 ||                 loss : 0.680, acc : 0.781\n",
            "Epoch : 62 || 50/368 ||                 loss : 0.608, acc : 0.828\n",
            "Epoch : 62 || 100/368 ||                 loss : 0.760, acc : 0.766\n",
            "Epoch : 62 || 150/368 ||                 loss : 0.845, acc : 0.703\n",
            "Epoch : 62 || 200/368 ||                 loss : 0.584, acc : 0.828\n",
            "Epoch : 62 || 250/368 ||                 loss : 0.823, acc : 0.703\n",
            "Epoch : 62 || 300/368 ||                 loss : 0.677, acc : 0.766\n",
            "Epoch : 62 || 350/368 ||                 loss : 0.620, acc : 0.828\n",
            "Validation result of epoch 62/100 || val_loss : 1.278 val_acc : 0.649 \n",
            "Epoch : 63 || 0/368 ||                 loss : 0.773, acc : 0.719\n",
            "Epoch : 63 || 50/368 ||                 loss : 0.655, acc : 0.797\n",
            "Epoch : 63 || 100/368 ||                 loss : 0.492, acc : 0.844\n",
            "Epoch : 63 || 150/368 ||                 loss : 0.729, acc : 0.734\n",
            "Epoch : 63 || 200/368 ||                 loss : 0.803, acc : 0.734\n",
            "Epoch : 63 || 250/368 ||                 loss : 0.635, acc : 0.828\n",
            "Epoch : 63 || 300/368 ||                 loss : 0.662, acc : 0.781\n",
            "Epoch : 63 || 350/368 ||                 loss : 0.951, acc : 0.719\n",
            "Validation result of epoch 63/100 || val_loss : 1.278 val_acc : 0.647 \n",
            "Epoch : 64 || 0/368 ||                 loss : 0.818, acc : 0.688\n",
            "Epoch : 64 || 50/368 ||                 loss : 0.753, acc : 0.781\n",
            "Epoch : 64 || 100/368 ||                 loss : 0.599, acc : 0.781\n",
            "Epoch : 64 || 150/368 ||                 loss : 0.622, acc : 0.797\n",
            "Epoch : 64 || 200/368 ||                 loss : 0.643, acc : 0.812\n",
            "Epoch : 64 || 250/368 ||                 loss : 0.669, acc : 0.766\n",
            "Epoch : 64 || 300/368 ||                 loss : 0.848, acc : 0.750\n",
            "Epoch : 64 || 350/368 ||                 loss : 0.666, acc : 0.812\n",
            "Validation result of epoch 64/100 || val_loss : 1.282 val_acc : 0.645 \n",
            "Epoch : 65 || 0/368 ||                 loss : 0.754, acc : 0.719\n",
            "Epoch : 65 || 50/368 ||                 loss : 0.623, acc : 0.812\n",
            "Epoch : 65 || 100/368 ||                 loss : 0.660, acc : 0.719\n",
            "Epoch : 65 || 150/368 ||                 loss : 0.851, acc : 0.781\n",
            "Epoch : 65 || 200/368 ||                 loss : 0.723, acc : 0.766\n",
            "Epoch : 65 || 250/368 ||                 loss : 0.942, acc : 0.688\n",
            "Epoch : 65 || 300/368 ||                 loss : 0.681, acc : 0.812\n",
            "Epoch : 65 || 350/368 ||                 loss : 0.761, acc : 0.766\n",
            "Validation result of epoch 65/100 || val_loss : 1.286 val_acc : 0.646 \n",
            "Epoch : 66 || 0/368 ||                 loss : 0.722, acc : 0.781\n",
            "Epoch : 66 || 50/368 ||                 loss : 0.609, acc : 0.797\n",
            "Epoch : 66 || 100/368 ||                 loss : 0.642, acc : 0.797\n",
            "Epoch : 66 || 150/368 ||                 loss : 0.800, acc : 0.734\n",
            "Epoch : 66 || 200/368 ||                 loss : 0.813, acc : 0.734\n",
            "Epoch : 66 || 250/368 ||                 loss : 0.638, acc : 0.781\n",
            "Epoch : 66 || 300/368 ||                 loss : 0.738, acc : 0.734\n",
            "Epoch : 66 || 350/368 ||                 loss : 0.849, acc : 0.688\n",
            "Validation result of epoch 66/100 || val_loss : 1.281 val_acc : 0.644 \n",
            "Epoch : 67 || 0/368 ||                 loss : 0.723, acc : 0.703\n",
            "Epoch : 67 || 50/368 ||                 loss : 0.626, acc : 0.781\n",
            "Epoch : 67 || 100/368 ||                 loss : 0.812, acc : 0.766\n",
            "Epoch : 67 || 150/368 ||                 loss : 0.519, acc : 0.828\n",
            "Epoch : 67 || 200/368 ||                 loss : 0.952, acc : 0.750\n",
            "Epoch : 67 || 250/368 ||                 loss : 0.514, acc : 0.859\n",
            "Epoch : 67 || 300/368 ||                 loss : 0.479, acc : 0.844\n",
            "Epoch : 67 || 350/368 ||                 loss : 0.795, acc : 0.719\n",
            "Validation result of epoch 67/100 || val_loss : 1.287 val_acc : 0.646 \n",
            "Epoch : 68 || 0/368 ||                 loss : 0.618, acc : 0.797\n",
            "Epoch : 68 || 50/368 ||                 loss : 0.834, acc : 0.719\n",
            "Epoch : 68 || 100/368 ||                 loss : 0.523, acc : 0.812\n",
            "Epoch : 68 || 150/368 ||                 loss : 0.410, acc : 0.906\n",
            "Epoch : 68 || 200/368 ||                 loss : 0.598, acc : 0.812\n",
            "Epoch : 68 || 250/368 ||                 loss : 0.465, acc : 0.844\n",
            "Epoch : 68 || 300/368 ||                 loss : 0.845, acc : 0.734\n",
            "Epoch : 68 || 350/368 ||                 loss : 0.745, acc : 0.812\n",
            "Validation result of epoch 68/100 || val_loss : 1.292 val_acc : 0.644 \n",
            "Epoch : 69 || 0/368 ||                 loss : 0.693, acc : 0.797\n",
            "Epoch : 69 || 50/368 ||                 loss : 0.814, acc : 0.719\n",
            "Epoch : 69 || 100/368 ||                 loss : 0.541, acc : 0.844\n",
            "Epoch : 69 || 150/368 ||                 loss : 0.560, acc : 0.859\n",
            "Epoch : 69 || 200/368 ||                 loss : 0.667, acc : 0.766\n",
            "Epoch : 69 || 250/368 ||                 loss : 0.687, acc : 0.719\n",
            "Epoch : 69 || 300/368 ||                 loss : 0.610, acc : 0.766\n",
            "Epoch : 69 || 350/368 ||                 loss : 0.694, acc : 0.828\n",
            "Validation result of epoch 69/100 || val_loss : 1.286 val_acc : 0.644 \n",
            "Epoch : 70 || 0/368 ||                 loss : 0.531, acc : 0.859\n",
            "Epoch : 70 || 50/368 ||                 loss : 0.610, acc : 0.812\n",
            "Epoch : 70 || 100/368 ||                 loss : 0.631, acc : 0.828\n",
            "Epoch : 70 || 150/368 ||                 loss : 0.827, acc : 0.750\n",
            "Epoch : 70 || 200/368 ||                 loss : 0.792, acc : 0.750\n",
            "Epoch : 70 || 250/368 ||                 loss : 0.997, acc : 0.734\n",
            "Epoch : 70 || 300/368 ||                 loss : 0.819, acc : 0.734\n",
            "Epoch : 70 || 350/368 ||                 loss : 0.609, acc : 0.781\n",
            "Validation result of epoch 70/100 || val_loss : 1.285 val_acc : 0.645 \n",
            "Epoch : 71 || 0/368 ||                 loss : 0.689, acc : 0.781\n",
            "Epoch : 71 || 50/368 ||                 loss : 0.775, acc : 0.734\n",
            "Epoch : 71 || 100/368 ||                 loss : 0.605, acc : 0.828\n",
            "Epoch : 71 || 150/368 ||                 loss : 0.733, acc : 0.797\n",
            "Epoch : 71 || 200/368 ||                 loss : 0.646, acc : 0.781\n",
            "Epoch : 71 || 250/368 ||                 loss : 0.762, acc : 0.750\n",
            "Epoch : 71 || 300/368 ||                 loss : 0.681, acc : 0.750\n",
            "Epoch : 71 || 350/368 ||                 loss : 0.671, acc : 0.797\n",
            "Validation result of epoch 71/100 || val_loss : 1.290 val_acc : 0.645 \n",
            "Epoch : 72 || 0/368 ||                 loss : 0.548, acc : 0.844\n",
            "Epoch : 72 || 50/368 ||                 loss : 0.568, acc : 0.828\n",
            "Epoch : 72 || 100/368 ||                 loss : 0.867, acc : 0.734\n",
            "Epoch : 72 || 150/368 ||                 loss : 0.721, acc : 0.750\n",
            "Epoch : 72 || 200/368 ||                 loss : 0.564, acc : 0.781\n",
            "Epoch : 72 || 250/368 ||                 loss : 0.751, acc : 0.828\n",
            "Epoch : 72 || 300/368 ||                 loss : 0.886, acc : 0.719\n",
            "Epoch : 72 || 350/368 ||                 loss : 0.975, acc : 0.672\n",
            "Validation result of epoch 72/100 || val_loss : 1.288 val_acc : 0.647 \n",
            "Epoch : 73 || 0/368 ||                 loss : 0.769, acc : 0.750\n",
            "Epoch : 73 || 50/368 ||                 loss : 0.707, acc : 0.766\n",
            "Epoch : 73 || 100/368 ||                 loss : 0.737, acc : 0.734\n",
            "Epoch : 73 || 150/368 ||                 loss : 0.614, acc : 0.781\n",
            "Epoch : 73 || 200/368 ||                 loss : 0.497, acc : 0.844\n",
            "Epoch : 73 || 250/368 ||                 loss : 0.331, acc : 0.906\n",
            "Epoch : 73 || 300/368 ||                 loss : 0.588, acc : 0.859\n",
            "Epoch : 73 || 350/368 ||                 loss : 0.574, acc : 0.781\n",
            "Validation result of epoch 73/100 || val_loss : 1.300 val_acc : 0.643 \n",
            "Epoch : 74 || 0/368 ||                 loss : 0.641, acc : 0.766\n",
            "Epoch : 74 || 50/368 ||                 loss : 0.668, acc : 0.766\n",
            "Epoch : 74 || 100/368 ||                 loss : 0.691, acc : 0.781\n",
            "Epoch : 74 || 150/368 ||                 loss : 0.559, acc : 0.875\n",
            "Epoch : 74 || 200/368 ||                 loss : 0.575, acc : 0.828\n",
            "Epoch : 74 || 250/368 ||                 loss : 0.475, acc : 0.859\n",
            "Epoch : 74 || 300/368 ||                 loss : 0.609, acc : 0.828\n",
            "Epoch : 74 || 350/368 ||                 loss : 0.638, acc : 0.812\n",
            "Validation result of epoch 74/100 || val_loss : 1.284 val_acc : 0.645 \n",
            "Epoch : 75 || 0/368 ||                 loss : 0.790, acc : 0.703\n",
            "Epoch : 75 || 50/368 ||                 loss : 0.694, acc : 0.781\n",
            "Epoch : 75 || 100/368 ||                 loss : 0.427, acc : 0.906\n",
            "Epoch : 75 || 150/368 ||                 loss : 0.638, acc : 0.797\n",
            "Epoch : 75 || 200/368 ||                 loss : 0.680, acc : 0.844\n",
            "Epoch : 75 || 250/368 ||                 loss : 0.641, acc : 0.797\n",
            "Epoch : 75 || 300/368 ||                 loss : 0.832, acc : 0.703\n",
            "Epoch : 75 || 350/368 ||                 loss : 0.669, acc : 0.750\n",
            "Validation result of epoch 75/100 || val_loss : 1.286 val_acc : 0.643 \n",
            "Epoch : 76 || 0/368 ||                 loss : 0.559, acc : 0.797\n",
            "Epoch : 76 || 50/368 ||                 loss : 0.586, acc : 0.828\n",
            "Epoch : 76 || 100/368 ||                 loss : 0.564, acc : 0.797\n",
            "Epoch : 76 || 150/368 ||                 loss : 0.497, acc : 0.844\n",
            "Epoch : 76 || 200/368 ||                 loss : 0.671, acc : 0.781\n",
            "Epoch : 76 || 250/368 ||                 loss : 0.464, acc : 0.859\n",
            "Epoch : 76 || 300/368 ||                 loss : 0.631, acc : 0.750\n",
            "Epoch : 76 || 350/368 ||                 loss : 0.485, acc : 0.844\n",
            "Validation result of epoch 76/100 || val_loss : 1.293 val_acc : 0.645 \n",
            "Epoch : 77 || 0/368 ||                 loss : 0.600, acc : 0.781\n",
            "Epoch : 77 || 50/368 ||                 loss : 0.533, acc : 0.812\n",
            "Epoch : 77 || 100/368 ||                 loss : 0.603, acc : 0.812\n",
            "Epoch : 77 || 150/368 ||                 loss : 0.815, acc : 0.781\n",
            "Epoch : 77 || 200/368 ||                 loss : 0.755, acc : 0.766\n",
            "Epoch : 77 || 250/368 ||                 loss : 0.589, acc : 0.828\n",
            "Epoch : 77 || 300/368 ||                 loss : 0.698, acc : 0.797\n",
            "Epoch : 77 || 350/368 ||                 loss : 0.663, acc : 0.781\n",
            "Validation result of epoch 77/100 || val_loss : 1.295 val_acc : 0.646 \n",
            "Epoch : 78 || 0/368 ||                 loss : 0.619, acc : 0.797\n",
            "Epoch : 78 || 50/368 ||                 loss : 0.608, acc : 0.844\n",
            "Epoch : 78 || 100/368 ||                 loss : 0.765, acc : 0.750\n",
            "Epoch : 78 || 150/368 ||                 loss : 0.730, acc : 0.719\n",
            "Epoch : 78 || 200/368 ||                 loss : 0.603, acc : 0.797\n",
            "Epoch : 78 || 250/368 ||                 loss : 0.560, acc : 0.766\n",
            "Epoch : 78 || 300/368 ||                 loss : 0.643, acc : 0.812\n",
            "Epoch : 78 || 350/368 ||                 loss : 0.764, acc : 0.734\n",
            "Validation result of epoch 78/100 || val_loss : 1.302 val_acc : 0.648 \n",
            "Epoch : 79 || 0/368 ||                 loss : 0.713, acc : 0.781\n",
            "Epoch : 79 || 50/368 ||                 loss : 0.467, acc : 0.828\n",
            "Epoch : 79 || 100/368 ||                 loss : 0.652, acc : 0.797\n",
            "Epoch : 79 || 150/368 ||                 loss : 0.704, acc : 0.766\n",
            "Epoch : 79 || 200/368 ||                 loss : 0.934, acc : 0.625\n",
            "Epoch : 79 || 250/368 ||                 loss : 0.833, acc : 0.750\n",
            "Epoch : 79 || 300/368 ||                 loss : 0.591, acc : 0.781\n",
            "Epoch : 79 || 350/368 ||                 loss : 0.816, acc : 0.750\n",
            "Validation result of epoch 79/100 || val_loss : 1.296 val_acc : 0.646 \n",
            "Epoch : 80 || 0/368 ||                 loss : 0.628, acc : 0.812\n",
            "Epoch : 80 || 50/368 ||                 loss : 0.899, acc : 0.688\n",
            "Epoch : 80 || 100/368 ||                 loss : 0.976, acc : 0.719\n",
            "Epoch : 80 || 150/368 ||                 loss : 0.468, acc : 0.859\n",
            "Epoch : 80 || 200/368 ||                 loss : 0.501, acc : 0.797\n",
            "Epoch : 80 || 250/368 ||                 loss : 0.294, acc : 0.953\n",
            "Epoch : 80 || 300/368 ||                 loss : 0.867, acc : 0.734\n",
            "Epoch : 80 || 350/368 ||                 loss : 0.532, acc : 0.859\n",
            "Validation result of epoch 80/100 || val_loss : 1.287 val_acc : 0.647 \n",
            "Epoch : 81 || 0/368 ||                 loss : 0.760, acc : 0.766\n",
            "Epoch : 81 || 50/368 ||                 loss : 0.756, acc : 0.734\n",
            "Epoch : 81 || 100/368 ||                 loss : 0.697, acc : 0.750\n",
            "Epoch : 81 || 150/368 ||                 loss : 0.455, acc : 0.844\n",
            "Epoch : 81 || 200/368 ||                 loss : 0.857, acc : 0.781\n",
            "Epoch : 81 || 250/368 ||                 loss : 0.805, acc : 0.719\n",
            "Epoch : 81 || 300/368 ||                 loss : 0.557, acc : 0.797\n",
            "Epoch : 81 || 350/368 ||                 loss : 0.630, acc : 0.797\n",
            "Validation result of epoch 81/100 || val_loss : 1.299 val_acc : 0.645 \n",
            "Epoch : 82 || 0/368 ||                 loss : 0.516, acc : 0.828\n",
            "Epoch : 82 || 50/368 ||                 loss : 0.509, acc : 0.844\n",
            "Epoch : 82 || 100/368 ||                 loss : 0.520, acc : 0.859\n",
            "Epoch : 82 || 150/368 ||                 loss : 0.637, acc : 0.812\n",
            "Epoch : 82 || 200/368 ||                 loss : 0.569, acc : 0.797\n",
            "Epoch : 82 || 250/368 ||                 loss : 0.474, acc : 0.875\n",
            "Epoch : 82 || 300/368 ||                 loss : 0.617, acc : 0.766\n",
            "Epoch : 82 || 350/368 ||                 loss : 0.775, acc : 0.750\n",
            "Validation result of epoch 82/100 || val_loss : 1.304 val_acc : 0.647 \n",
            "Epoch : 83 || 0/368 ||                 loss : 0.596, acc : 0.828\n",
            "Epoch : 83 || 50/368 ||                 loss : 0.541, acc : 0.859\n",
            "Epoch : 83 || 100/368 ||                 loss : 0.519, acc : 0.828\n",
            "Epoch : 83 || 150/368 ||                 loss : 0.608, acc : 0.812\n",
            "Epoch : 83 || 200/368 ||                 loss : 0.528, acc : 0.828\n",
            "Epoch : 83 || 250/368 ||                 loss : 0.745, acc : 0.781\n",
            "Epoch : 83 || 300/368 ||                 loss : 0.445, acc : 0.875\n",
            "Epoch : 83 || 350/368 ||                 loss : 0.658, acc : 0.781\n",
            "Validation result of epoch 83/100 || val_loss : 1.294 val_acc : 0.647 \n",
            "Epoch : 84 || 0/368 ||                 loss : 0.615, acc : 0.766\n",
            "Epoch : 84 || 50/368 ||                 loss : 0.641, acc : 0.781\n",
            "Epoch : 84 || 100/368 ||                 loss : 0.879, acc : 0.719\n",
            "Epoch : 84 || 150/368 ||                 loss : 0.540, acc : 0.828\n",
            "Epoch : 84 || 200/368 ||                 loss : 0.499, acc : 0.844\n",
            "Epoch : 84 || 250/368 ||                 loss : 0.545, acc : 0.844\n",
            "Epoch : 84 || 300/368 ||                 loss : 0.576, acc : 0.828\n",
            "Epoch : 84 || 350/368 ||                 loss : 0.741, acc : 0.781\n",
            "Validation result of epoch 84/100 || val_loss : 1.294 val_acc : 0.648 \n",
            "Epoch : 85 || 0/368 ||                 loss : 0.385, acc : 0.891\n",
            "Epoch : 85 || 50/368 ||                 loss : 0.681, acc : 0.812\n",
            "Epoch : 85 || 100/368 ||                 loss : 0.561, acc : 0.797\n",
            "Epoch : 85 || 150/368 ||                 loss : 0.495, acc : 0.828\n",
            "Epoch : 85 || 200/368 ||                 loss : 0.538, acc : 0.828\n",
            "Epoch : 85 || 250/368 ||                 loss : 0.498, acc : 0.812\n",
            "Epoch : 85 || 300/368 ||                 loss : 0.693, acc : 0.797\n",
            "Epoch : 85 || 350/368 ||                 loss : 0.640, acc : 0.812\n",
            "Validation result of epoch 85/100 || val_loss : 1.290 val_acc : 0.649 \n",
            "Epoch : 86 || 0/368 ||                 loss : 0.891, acc : 0.703\n",
            "Epoch : 86 || 50/368 ||                 loss : 0.849, acc : 0.766\n",
            "Epoch : 86 || 100/368 ||                 loss : 0.608, acc : 0.781\n",
            "Epoch : 86 || 150/368 ||                 loss : 0.722, acc : 0.719\n",
            "Epoch : 86 || 200/368 ||                 loss : 0.573, acc : 0.828\n",
            "Epoch : 86 || 250/368 ||                 loss : 0.688, acc : 0.688\n",
            "Epoch : 86 || 300/368 ||                 loss : 0.632, acc : 0.766\n",
            "Epoch : 86 || 350/368 ||                 loss : 0.450, acc : 0.859\n",
            "Validation result of epoch 86/100 || val_loss : 1.303 val_acc : 0.647 \n",
            "Epoch : 87 || 0/368 ||                 loss : 0.734, acc : 0.734\n",
            "Epoch : 87 || 50/368 ||                 loss : 0.732, acc : 0.719\n",
            "Epoch : 87 || 100/368 ||                 loss : 0.685, acc : 0.781\n",
            "Epoch : 87 || 150/368 ||                 loss : 0.781, acc : 0.734\n",
            "Epoch : 87 || 200/368 ||                 loss : 0.601, acc : 0.750\n",
            "Epoch : 87 || 250/368 ||                 loss : 0.881, acc : 0.734\n",
            "Epoch : 87 || 300/368 ||                 loss : 0.645, acc : 0.828\n",
            "Epoch : 87 || 350/368 ||                 loss : 0.569, acc : 0.844\n",
            "Validation result of epoch 87/100 || val_loss : 1.293 val_acc : 0.646 \n",
            "Epoch : 88 || 0/368 ||                 loss : 0.634, acc : 0.797\n",
            "Epoch : 88 || 50/368 ||                 loss : 0.894, acc : 0.703\n",
            "Epoch : 88 || 100/368 ||                 loss : 0.624, acc : 0.812\n",
            "Epoch : 88 || 150/368 ||                 loss : 0.482, acc : 0.859\n",
            "Epoch : 88 || 200/368 ||                 loss : 0.748, acc : 0.734\n",
            "Epoch : 88 || 250/368 ||                 loss : 0.597, acc : 0.844\n",
            "Epoch : 88 || 300/368 ||                 loss : 0.489, acc : 0.875\n",
            "Epoch : 88 || 350/368 ||                 loss : 0.597, acc : 0.797\n",
            "Validation result of epoch 88/100 || val_loss : 1.295 val_acc : 0.647 \n",
            "Epoch : 89 || 0/368 ||                 loss : 0.591, acc : 0.828\n",
            "Epoch : 89 || 50/368 ||                 loss : 0.380, acc : 0.875\n",
            "Epoch : 89 || 100/368 ||                 loss : 0.538, acc : 0.797\n",
            "Epoch : 89 || 150/368 ||                 loss : 0.618, acc : 0.766\n",
            "Epoch : 89 || 200/368 ||                 loss : 0.788, acc : 0.734\n",
            "Epoch : 89 || 250/368 ||                 loss : 0.621, acc : 0.828\n",
            "Epoch : 89 || 300/368 ||                 loss : 0.547, acc : 0.844\n",
            "Epoch : 89 || 350/368 ||                 loss : 0.563, acc : 0.828\n",
            "Validation result of epoch 89/100 || val_loss : 1.294 val_acc : 0.648 \n",
            "Epoch : 90 || 0/368 ||                 loss : 0.517, acc : 0.797\n",
            "Epoch : 90 || 50/368 ||                 loss : 0.840, acc : 0.719\n",
            "Epoch : 90 || 100/368 ||                 loss : 0.623, acc : 0.797\n",
            "Epoch : 90 || 150/368 ||                 loss : 0.854, acc : 0.812\n",
            "Epoch : 90 || 200/368 ||                 loss : 0.632, acc : 0.766\n",
            "Epoch : 90 || 250/368 ||                 loss : 0.672, acc : 0.797\n",
            "Epoch : 90 || 300/368 ||                 loss : 0.543, acc : 0.859\n",
            "Epoch : 90 || 350/368 ||                 loss : 0.579, acc : 0.812\n",
            "Validation result of epoch 90/100 || val_loss : 1.288 val_acc : 0.647 \n",
            "Epoch : 91 || 0/368 ||                 loss : 0.577, acc : 0.828\n",
            "Epoch : 91 || 50/368 ||                 loss : 0.461, acc : 0.844\n",
            "Epoch : 91 || 100/368 ||                 loss : 0.530, acc : 0.828\n",
            "Epoch : 91 || 150/368 ||                 loss : 0.716, acc : 0.750\n",
            "Epoch : 91 || 200/368 ||                 loss : 0.581, acc : 0.797\n",
            "Epoch : 91 || 250/368 ||                 loss : 0.322, acc : 0.906\n",
            "Epoch : 91 || 300/368 ||                 loss : 0.577, acc : 0.844\n",
            "Epoch : 91 || 350/368 ||                 loss : 0.747, acc : 0.781\n",
            "Validation result of epoch 91/100 || val_loss : 1.296 val_acc : 0.645 \n",
            "Epoch : 92 || 0/368 ||                 loss : 0.453, acc : 0.844\n",
            "Epoch : 92 || 50/368 ||                 loss : 0.677, acc : 0.828\n",
            "Epoch : 92 || 100/368 ||                 loss : 0.563, acc : 0.891\n",
            "Epoch : 92 || 150/368 ||                 loss : 0.541, acc : 0.844\n",
            "Epoch : 92 || 200/368 ||                 loss : 0.519, acc : 0.828\n",
            "Epoch : 92 || 250/368 ||                 loss : 0.523, acc : 0.812\n",
            "Epoch : 92 || 300/368 ||                 loss : 0.602, acc : 0.766\n",
            "Epoch : 92 || 350/368 ||                 loss : 0.964, acc : 0.719\n",
            "Validation result of epoch 92/100 || val_loss : 1.295 val_acc : 0.649 \n",
            "Epoch : 93 || 0/368 ||                 loss : 0.621, acc : 0.797\n",
            "Epoch : 93 || 50/368 ||                 loss : 0.582, acc : 0.812\n",
            "Epoch : 93 || 100/368 ||                 loss : 0.552, acc : 0.828\n",
            "Epoch : 93 || 150/368 ||                 loss : 0.720, acc : 0.844\n",
            "Epoch : 93 || 200/368 ||                 loss : 0.677, acc : 0.812\n",
            "Epoch : 93 || 250/368 ||                 loss : 0.605, acc : 0.781\n",
            "Epoch : 93 || 300/368 ||                 loss : 0.710, acc : 0.750\n",
            "Epoch : 93 || 350/368 ||                 loss : 0.793, acc : 0.766\n",
            "Validation result of epoch 93/100 || val_loss : 1.292 val_acc : 0.648 \n",
            "Epoch : 94 || 0/368 ||                 loss : 0.486, acc : 0.891\n",
            "Epoch : 94 || 50/368 ||                 loss : 0.692, acc : 0.781\n",
            "Epoch : 94 || 100/368 ||                 loss : 0.494, acc : 0.828\n",
            "Epoch : 94 || 150/368 ||                 loss : 0.557, acc : 0.812\n",
            "Epoch : 94 || 200/368 ||                 loss : 0.540, acc : 0.859\n",
            "Epoch : 94 || 250/368 ||                 loss : 0.301, acc : 0.906\n",
            "Epoch : 94 || 300/368 ||                 loss : 0.593, acc : 0.844\n",
            "Epoch : 94 || 350/368 ||                 loss : 0.539, acc : 0.828\n",
            "Validation result of epoch 94/100 || val_loss : 1.310 val_acc : 0.648 \n",
            "Epoch : 95 || 0/368 ||                 loss : 0.597, acc : 0.797\n",
            "Epoch : 95 || 50/368 ||                 loss : 0.527, acc : 0.844\n",
            "Epoch : 95 || 100/368 ||                 loss : 0.517, acc : 0.781\n",
            "Epoch : 95 || 150/368 ||                 loss : 0.914, acc : 0.719\n",
            "Epoch : 95 || 200/368 ||                 loss : 0.590, acc : 0.797\n",
            "Epoch : 95 || 250/368 ||                 loss : 0.579, acc : 0.844\n",
            "Epoch : 95 || 300/368 ||                 loss : 0.824, acc : 0.734\n",
            "Epoch : 95 || 350/368 ||                 loss : 0.805, acc : 0.750\n",
            "Validation result of epoch 95/100 || val_loss : 1.295 val_acc : 0.647 \n",
            "Epoch : 96 || 0/368 ||                 loss : 0.607, acc : 0.781\n",
            "Epoch : 96 || 50/368 ||                 loss : 0.580, acc : 0.781\n",
            "Epoch : 96 || 100/368 ||                 loss : 0.544, acc : 0.844\n",
            "Epoch : 96 || 150/368 ||                 loss : 0.963, acc : 0.719\n",
            "Epoch : 96 || 200/368 ||                 loss : 0.499, acc : 0.812\n",
            "Epoch : 96 || 250/368 ||                 loss : 0.686, acc : 0.766\n",
            "Epoch : 96 || 300/368 ||                 loss : 0.453, acc : 0.797\n",
            "Epoch : 96 || 350/368 ||                 loss : 0.715, acc : 0.812\n",
            "Validation result of epoch 96/100 || val_loss : 1.306 val_acc : 0.648 \n",
            "Epoch : 97 || 0/368 ||                 loss : 0.523, acc : 0.828\n",
            "Epoch : 97 || 50/368 ||                 loss : 0.474, acc : 0.859\n",
            "Epoch : 97 || 100/368 ||                 loss : 0.649, acc : 0.703\n",
            "Epoch : 97 || 150/368 ||                 loss : 0.655, acc : 0.812\n",
            "Epoch : 97 || 200/368 ||                 loss : 0.689, acc : 0.828\n",
            "Epoch : 97 || 250/368 ||                 loss : 0.700, acc : 0.703\n",
            "Epoch : 97 || 300/368 ||                 loss : 0.564, acc : 0.844\n",
            "Epoch : 97 || 350/368 ||                 loss : 0.593, acc : 0.812\n",
            "Validation result of epoch 97/100 || val_loss : 1.296 val_acc : 0.647 \n",
            "Epoch : 98 || 0/368 ||                 loss : 0.641, acc : 0.781\n",
            "Epoch : 98 || 50/368 ||                 loss : 0.432, acc : 0.875\n",
            "Epoch : 98 || 100/368 ||                 loss : 0.525, acc : 0.812\n",
            "Epoch : 98 || 150/368 ||                 loss : 0.492, acc : 0.891\n",
            "Epoch : 98 || 200/368 ||                 loss : 0.798, acc : 0.734\n",
            "Epoch : 98 || 250/368 ||                 loss : 0.508, acc : 0.859\n",
            "Epoch : 98 || 300/368 ||                 loss : 0.496, acc : 0.828\n",
            "Epoch : 98 || 350/368 ||                 loss : 0.988, acc : 0.750\n",
            "Validation result of epoch 98/100 || val_loss : 1.312 val_acc : 0.650 \n",
            "Epoch : 99 || 0/368 ||                 loss : 0.392, acc : 0.891\n",
            "Epoch : 99 || 50/368 ||                 loss : 0.564, acc : 0.812\n",
            "Epoch : 99 || 100/368 ||                 loss : 0.548, acc : 0.812\n",
            "Epoch : 99 || 150/368 ||                 loss : 0.636, acc : 0.766\n",
            "Epoch : 99 || 200/368 ||                 loss : 0.582, acc : 0.844\n",
            "Epoch : 99 || 250/368 ||                 loss : 0.491, acc : 0.844\n",
            "Epoch : 99 || 300/368 ||                 loss : 0.645, acc : 0.828\n",
            "Epoch : 99 || 350/368 ||                 loss : 0.634, acc : 0.766\n",
            "Validation result of epoch 99/100 || val_loss : 1.302 val_acc : 0.646 \n",
            "Best validation accuracy CrossE of classifier network : 0.651 took 231.631 secs\n",
            "Best validation accuracy Micro of classifier network : 0.651 took 231.631 secs\n",
            "Best validation accuracy Macro of classifier network : 0.487 took 231.631 secs\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Visualize**"
      ],
      "metadata": {
        "id": "2QCMX1MKmnYl"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "plt.figure(figsize=(10,5))\n",
        "plt.title(\"Classification Training and Validation Loss\")\n",
        "plt.plot(train_loss_clf, label=\"train_loss\", color='red', linewidth=1.0)\n",
        "plt.plot(val_loss_clf, label=\"val_loss\", color='blue', linewidth=1.0)\n",
        "plt.xlabel(\"Iterations\")\n",
        "plt.ylabel(\"Loss\")\n",
        "plt.legend()\n",
        "\n",
        "\n",
        "plt.figure(figsize=(10,5))\n",
        "plt.title(\"Classification Training and Validation Accuracy\")\n",
        "plt.plot(train_accuracy_clf, label=\"train_acc\", color='green', linewidth=1.0)\n",
        "plt.plot(val_accuracy_clf, label=\"val_acc\", color='orange', linewidth=1.0)\n",
        "plt.xlabel(\"Iterations\")\n",
        "plt.ylabel(\"Loss\")\n",
        "plt.legend()\n",
        "\n",
        "plt.tight_layout(4)\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 644
        },
        "id": "p20GF8UBmXeV",
        "outputId": "f446f3c2-0a32-48f2-860a-fcd7b27cad9d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x360 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x360 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Test**"
      ],
      "metadata": {
        "id": "XX-ug_PRmsYD"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "model = Classifier(args.dim, len(labels))\n",
        "# model.load_state_dict(best_val_model)\n",
        "model.load_state_dict(torch.load(os.path.join(models_path, \"classifer\")))\n",
        "\n",
        "model = model.to(device)\n",
        "pred_origins = [0]*len(labels)\n",
        "origins = list(labels.keys())\n",
        "test_origins = []\n",
        "model.eval()\n",
        "\n",
        "for batch_idx, data in enumerate(test_clf_dataloader):\n",
        "  x = data['recipe'].to(device=device)\n",
        "  pred = model(x)\n",
        "  pred = pred.type(torch.float).to(device)\n",
        "  pred = torch.argmax(torch.log_softmax(pred, 1), 1).tolist()\n",
        "  for idx in pred:\n",
        "    test_origins.append(origins[idx])\n",
        "    pred_origins[idx] += 1"
      ],
      "metadata": {
        "id": "6fB6c4uCmmtE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "pd.DataFrame(test_origins, columns = [\"origin\"]).to_csv(os.path.join(results, \"test_classification_answer.csv\"))"
      ],
      "metadata": {
        "id": "j9xzdTt0a6L2"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "fig1 = px.bar(x = list(labels.keys()), y = pred_origins, text_auto='.2s')\n",
        "fig1.update_layout(\n",
        "    xaxis={'categoryorder':'total descending'},\n",
        "    xaxis_title=\"Origin\",\n",
        "    yaxis_title = \"Frequency\",\n",
        "    title=\"Predicted Origins and their frequency\",\n",
        "    template=\"plotly_dark\")\n",
        "fig1.update_traces(textfont_size=12, textangle=0, textposition=\"outside\", cliponaxis=False)\n",
        "fig1"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 542
        },
        "id": "s-3MWnzgkeYY",
        "outputId": "73e0ad3f-367d-473e-cd19-dccfee28b2d0"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<html>\n",
              "<head><meta charset=\"utf-8\" /></head>\n",
              "<body>\n",
              "    <div>            <script src=\"https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/MathJax.js?config=TeX-AMS-MML_SVG\"></script><script type=\"text/javascript\">if (window.MathJax) {MathJax.Hub.Config({SVG: {font: \"STIX-Web\"}});}</script>                <script type=\"text/javascript\">window.PlotlyConfig = {MathJaxConfig: 'local'};</script>\n",
              "        <script src=\"https://cdn.plot.ly/plotly-2.8.3.min.js\"></script>                <div id=\"50dd6296-5e85-4304-83c1-06fbd357a822\" class=\"plotly-graph-div\" style=\"height:525px; width:100%;\"></div>            <script type=\"text/javascript\">                                    window.PLOTLYENV=window.PLOTLYENV || {};                                    if (document.getElementById(\"50dd6296-5e85-4304-83c1-06fbd357a822\")) {                    Plotly.newPlot(                        \"50dd6296-5e85-4304-83c1-06fbd357a822\",                        [{\"alignmentgroup\":\"True\",\"hovertemplate\":\"x=%{x}<br>y=%{y}<extra></extra>\",\"legendgroup\":\"\",\"marker\":{\"color\":\"#636efa\",\"pattern\":{\"shape\":\"\"}},\"name\":\"\",\"offsetgroup\":\"\",\"orientation\":\"v\",\"showlegend\":false,\"textposition\":\"outside\",\"texttemplate\":\"%{y:.2s}\",\"x\":[\"brazilian\",\"british\",\"cajun_creole\",\"chinese\",\"filipino\",\"french\",\"greek\",\"indian\",\"irish\",\"italian\",\"jamaican\",\"japanese\",\"korean\",\"mexican\",\"moroccan\",\"russian\",\"southern_us\",\"spanish\",\"thai\",\"vietnamese\"],\"xaxis\":\"x\",\"y\":[27,36,129,262,58,274,86,307,35,946,38,96,70,695,88,27,494,48,144,64],\"yaxis\":\"y\",\"type\":\"bar\",\"textfont\":{\"size\":12},\"cliponaxis\":false,\"textangle\":0}],                        {\"template\":{\"data\":{\"barpolar\":[{\"marker\":{\"line\":{\"color\":\"rgb(17,17,17)\",\"width\":0.5},\"pattern\":{\"fillmode\":\"overlay\",\"size\":10,\"solidity\":0.2}},\"type\":\"barpolar\"}],\"bar\":[{\"error_x\":{\"color\":\"#f2f5fa\"},\"error_y\":{\"color\":\"#f2f5fa\"},\"marker\":{\"line\":{\"color\":\"rgb(17,17,17)\",\"width\":0.5},\"pattern\":{\"fillmode\":\"overlay\",\"size\":10,\"solidity\":0.2}},\"type\":\"bar\"}],\"carpet\":[{\"aaxis\":{\"endlinecolor\":\"#A2B1C6\",\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"minorgridcolor\":\"#506784\",\"startlinecolor\":\"#A2B1C6\"},\"baxis\":{\"endlinecolor\":\"#A2B1C6\",\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"minorgridcolor\":\"#506784\",\"startlinecolor\":\"#A2B1C6\"},\"type\":\"carpet\"}],\"choropleth\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"type\":\"choropleth\"}],\"contourcarpet\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"type\":\"contourcarpet\"}],\"contour\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"contour\"}],\"heatmapgl\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"heatmapgl\"}],\"heatmap\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"heatmap\"}],\"histogram2dcontour\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"histogram2dcontour\"}],\"histogram2d\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"histogram2d\"}],\"histogram\":[{\"marker\":{\"pattern\":{\"fillmode\":\"overlay\",\"size\":10,\"solidity\":0.2}},\"type\":\"histogram\"}],\"mesh3d\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"type\":\"mesh3d\"}],\"parcoords\":[{\"line\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"parcoords\"}],\"pie\":[{\"automargin\":true,\"type\":\"pie\"}],\"scatter3d\":[{\"line\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatter3d\"}],\"scattercarpet\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scattercarpet\"}],\"scattergeo\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scattergeo\"}],\"scattergl\":[{\"marker\":{\"line\":{\"color\":\"#283442\"}},\"type\":\"scattergl\"}],\"scattermapbox\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scattermapbox\"}],\"scatterpolargl\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatterpolargl\"}],\"scatterpolar\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatterpolar\"}],\"scatter\":[{\"marker\":{\"line\":{\"color\":\"#283442\"}},\"type\":\"scatter\"}],\"scatterternary\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatterternary\"}],\"surface\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"surface\"}],\"table\":[{\"cells\":{\"fill\":{\"color\":\"#506784\"},\"line\":{\"color\":\"rgb(17,17,17)\"}},\"header\":{\"fill\":{\"color\":\"#2a3f5f\"},\"line\":{\"color\":\"rgb(17,17,17)\"}},\"type\":\"table\"}]},\"layout\":{\"annotationdefaults\":{\"arrowcolor\":\"#f2f5fa\",\"arrowhead\":0,\"arrowwidth\":1},\"autotypenumbers\":\"strict\",\"coloraxis\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"colorscale\":{\"diverging\":[[0,\"#8e0152\"],[0.1,\"#c51b7d\"],[0.2,\"#de77ae\"],[0.3,\"#f1b6da\"],[0.4,\"#fde0ef\"],[0.5,\"#f7f7f7\"],[0.6,\"#e6f5d0\"],[0.7,\"#b8e186\"],[0.8,\"#7fbc41\"],[0.9,\"#4d9221\"],[1,\"#276419\"]],\"sequential\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"sequentialminus\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]]},\"colorway\":[\"#636efa\",\"#EF553B\",\"#00cc96\",\"#ab63fa\",\"#FFA15A\",\"#19d3f3\",\"#FF6692\",\"#B6E880\",\"#FF97FF\",\"#FECB52\"],\"font\":{\"color\":\"#f2f5fa\"},\"geo\":{\"bgcolor\":\"rgb(17,17,17)\",\"lakecolor\":\"rgb(17,17,17)\",\"landcolor\":\"rgb(17,17,17)\",\"showlakes\":true,\"showland\":true,\"subunitcolor\":\"#506784\"},\"hoverlabel\":{\"align\":\"left\"},\"hovermode\":\"closest\",\"mapbox\":{\"style\":\"dark\"},\"paper_bgcolor\":\"rgb(17,17,17)\",\"plot_bgcolor\":\"rgb(17,17,17)\",\"polar\":{\"angularaxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"},\"bgcolor\":\"rgb(17,17,17)\",\"radialaxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"}},\"scene\":{\"xaxis\":{\"backgroundcolor\":\"rgb(17,17,17)\",\"gridcolor\":\"#506784\",\"gridwidth\":2,\"linecolor\":\"#506784\",\"showbackground\":true,\"ticks\":\"\",\"zerolinecolor\":\"#C8D4E3\"},\"yaxis\":{\"backgroundcolor\":\"rgb(17,17,17)\",\"gridcolor\":\"#506784\",\"gridwidth\":2,\"linecolor\":\"#506784\",\"showbackground\":true,\"ticks\":\"\",\"zerolinecolor\":\"#C8D4E3\"},\"zaxis\":{\"backgroundcolor\":\"rgb(17,17,17)\",\"gridcolor\":\"#506784\",\"gridwidth\":2,\"linecolor\":\"#506784\",\"showbackground\":true,\"ticks\":\"\",\"zerolinecolor\":\"#C8D4E3\"}},\"shapedefaults\":{\"line\":{\"color\":\"#f2f5fa\"}},\"sliderdefaults\":{\"bgcolor\":\"#C8D4E3\",\"bordercolor\":\"rgb(17,17,17)\",\"borderwidth\":1,\"tickwidth\":0},\"ternary\":{\"aaxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"},\"baxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"},\"bgcolor\":\"rgb(17,17,17)\",\"caxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"}},\"title\":{\"x\":0.05},\"updatemenudefaults\":{\"bgcolor\":\"#506784\",\"borderwidth\":0},\"xaxis\":{\"automargin\":true,\"gridcolor\":\"#283442\",\"linecolor\":\"#506784\",\"ticks\":\"\",\"title\":{\"standoff\":15},\"zerolinecolor\":\"#283442\",\"zerolinewidth\":2},\"yaxis\":{\"automargin\":true,\"gridcolor\":\"#283442\",\"linecolor\":\"#506784\",\"ticks\":\"\",\"title\":{\"standoff\":15},\"zerolinecolor\":\"#283442\",\"zerolinewidth\":2}}},\"xaxis\":{\"anchor\":\"y\",\"domain\":[0.0,1.0],\"title\":{\"text\":\"Origin\"},\"categoryorder\":\"total descending\"},\"yaxis\":{\"anchor\":\"x\",\"domain\":[0.0,1.0],\"title\":{\"text\":\"Frequency\"}},\"legend\":{\"tracegroupgap\":0},\"margin\":{\"t\":60},\"barmode\":\"relative\",\"title\":{\"text\":\"Predicted Origins and their frequency\"}},                        {\"responsive\": true}                    ).then(function(){\n",
              "                            \n",
              "var gd = document.getElementById('50dd6296-5e85-4304-83c1-06fbd357a822');\n",
              "var x = new MutationObserver(function (mutations, observer) {{\n",
              "        var display = window.getComputedStyle(gd).display;\n",
              "        if (!display || display === 'none') {{\n",
              "            console.log([gd, 'removed!']);\n",
              "            Plotly.purge(gd);\n",
              "            observer.disconnect();\n",
              "        }}\n",
              "}});\n",
              "\n",
              "// Listen for the removal of the full notebook cells\n",
              "var notebookContainer = gd.closest('#notebook-container');\n",
              "if (notebookContainer) {{\n",
              "    x.observe(notebookContainer, {childList: true});\n",
              "}}\n",
              "\n",
              "// Listen for the clearing of the current output cell\n",
              "var outputEl = gd.closest('.output');\n",
              "if (outputEl) {{\n",
              "    x.observe(outputEl, {childList: true});\n",
              "}}\n",
              "\n",
              "                        })                };                            </script>        </div>\n",
              "</body>\n",
              "</html>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "fig2 = px.bar(x = list(labels.keys()), y = list(np.sum(node_features, axis = 0)), text_auto='.2s')\n",
        "fig2.update_layout(\n",
        "    xaxis={'categoryorder':'total descending'},\n",
        "    title=\"Each countries distribution\",\n",
        "    template=\"plotly_dark\")\n",
        "fig2.update_traces(textfont_size=12, textangle=0, textposition=\"outside\", cliponaxis=False)\n",
        "\n",
        "fig2"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 542
        },
        "id": "E2CtUn7yicrq",
        "outputId": "64c61574-185f-466e-aeb5-5926f4c4582d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<html>\n",
              "<head><meta charset=\"utf-8\" /></head>\n",
              "<body>\n",
              "    <div>            <script src=\"https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/MathJax.js?config=TeX-AMS-MML_SVG\"></script><script type=\"text/javascript\">if (window.MathJax) {MathJax.Hub.Config({SVG: {font: \"STIX-Web\"}});}</script>                <script type=\"text/javascript\">window.PlotlyConfig = {MathJaxConfig: 'local'};</script>\n",
              "        <script src=\"https://cdn.plot.ly/plotly-2.8.3.min.js\"></script>                <div id=\"31c49557-b62a-4f44-8c76-3e91915ecb32\" class=\"plotly-graph-div\" style=\"height:525px; width:100%;\"></div>            <script type=\"text/javascript\">                                    window.PLOTLYENV=window.PLOTLYENV || {};                                    if (document.getElementById(\"31c49557-b62a-4f44-8c76-3e91915ecb32\")) {                    Plotly.newPlot(                        \"31c49557-b62a-4f44-8c76-3e91915ecb32\",                        [{\"alignmentgroup\":\"True\",\"hovertemplate\":\"x=%{x}<br>y=%{y}<extra></extra>\",\"legendgroup\":\"\",\"marker\":{\"color\":\"#636efa\",\"pattern\":{\"shape\":\"\"}},\"name\":\"\",\"offsetgroup\":\"\",\"orientation\":\"v\",\"showlegend\":false,\"textposition\":\"outside\",\"texttemplate\":\"%{y:.2s}\",\"x\":[\"brazilian\",\"british\",\"cajun_creole\",\"chinese\",\"filipino\",\"french\",\"greek\",\"indian\",\"irish\",\"italian\",\"jamaican\",\"japanese\",\"korean\",\"mexican\",\"moroccan\",\"russian\",\"southern_us\",\"spanish\",\"thai\",\"vietnamese\"],\"xaxis\":\"x\",\"y\":[3556.0,6328.0,15349.0,25379.0,5889.0,20381.0,9532.0,30216.0,4876.0,61791.0,4750.0,11071.0,7240.0,55410.0,8518.0,3989.0,32383.0,8275.0,15036.0,8275.0],\"yaxis\":\"y\",\"type\":\"bar\",\"textfont\":{\"size\":12},\"cliponaxis\":false,\"textangle\":0}],                        {\"template\":{\"data\":{\"barpolar\":[{\"marker\":{\"line\":{\"color\":\"rgb(17,17,17)\",\"width\":0.5},\"pattern\":{\"fillmode\":\"overlay\",\"size\":10,\"solidity\":0.2}},\"type\":\"barpolar\"}],\"bar\":[{\"error_x\":{\"color\":\"#f2f5fa\"},\"error_y\":{\"color\":\"#f2f5fa\"},\"marker\":{\"line\":{\"color\":\"rgb(17,17,17)\",\"width\":0.5},\"pattern\":{\"fillmode\":\"overlay\",\"size\":10,\"solidity\":0.2}},\"type\":\"bar\"}],\"carpet\":[{\"aaxis\":{\"endlinecolor\":\"#A2B1C6\",\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"minorgridcolor\":\"#506784\",\"startlinecolor\":\"#A2B1C6\"},\"baxis\":{\"endlinecolor\":\"#A2B1C6\",\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"minorgridcolor\":\"#506784\",\"startlinecolor\":\"#A2B1C6\"},\"type\":\"carpet\"}],\"choropleth\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"type\":\"choropleth\"}],\"contourcarpet\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"type\":\"contourcarpet\"}],\"contour\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"contour\"}],\"heatmapgl\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"heatmapgl\"}],\"heatmap\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"heatmap\"}],\"histogram2dcontour\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"histogram2dcontour\"}],\"histogram2d\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"histogram2d\"}],\"histogram\":[{\"marker\":{\"pattern\":{\"fillmode\":\"overlay\",\"size\":10,\"solidity\":0.2}},\"type\":\"histogram\"}],\"mesh3d\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"type\":\"mesh3d\"}],\"parcoords\":[{\"line\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"parcoords\"}],\"pie\":[{\"automargin\":true,\"type\":\"pie\"}],\"scatter3d\":[{\"line\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatter3d\"}],\"scattercarpet\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scattercarpet\"}],\"scattergeo\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scattergeo\"}],\"scattergl\":[{\"marker\":{\"line\":{\"color\":\"#283442\"}},\"type\":\"scattergl\"}],\"scattermapbox\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scattermapbox\"}],\"scatterpolargl\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatterpolargl\"}],\"scatterpolar\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatterpolar\"}],\"scatter\":[{\"marker\":{\"line\":{\"color\":\"#283442\"}},\"type\":\"scatter\"}],\"scatterternary\":[{\"marker\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"type\":\"scatterternary\"}],\"surface\":[{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"},\"colorscale\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"type\":\"surface\"}],\"table\":[{\"cells\":{\"fill\":{\"color\":\"#506784\"},\"line\":{\"color\":\"rgb(17,17,17)\"}},\"header\":{\"fill\":{\"color\":\"#2a3f5f\"},\"line\":{\"color\":\"rgb(17,17,17)\"}},\"type\":\"table\"}]},\"layout\":{\"annotationdefaults\":{\"arrowcolor\":\"#f2f5fa\",\"arrowhead\":0,\"arrowwidth\":1},\"autotypenumbers\":\"strict\",\"coloraxis\":{\"colorbar\":{\"outlinewidth\":0,\"ticks\":\"\"}},\"colorscale\":{\"diverging\":[[0,\"#8e0152\"],[0.1,\"#c51b7d\"],[0.2,\"#de77ae\"],[0.3,\"#f1b6da\"],[0.4,\"#fde0ef\"],[0.5,\"#f7f7f7\"],[0.6,\"#e6f5d0\"],[0.7,\"#b8e186\"],[0.8,\"#7fbc41\"],[0.9,\"#4d9221\"],[1,\"#276419\"]],\"sequential\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]],\"sequentialminus\":[[0.0,\"#0d0887\"],[0.1111111111111111,\"#46039f\"],[0.2222222222222222,\"#7201a8\"],[0.3333333333333333,\"#9c179e\"],[0.4444444444444444,\"#bd3786\"],[0.5555555555555556,\"#d8576b\"],[0.6666666666666666,\"#ed7953\"],[0.7777777777777778,\"#fb9f3a\"],[0.8888888888888888,\"#fdca26\"],[1.0,\"#f0f921\"]]},\"colorway\":[\"#636efa\",\"#EF553B\",\"#00cc96\",\"#ab63fa\",\"#FFA15A\",\"#19d3f3\",\"#FF6692\",\"#B6E880\",\"#FF97FF\",\"#FECB52\"],\"font\":{\"color\":\"#f2f5fa\"},\"geo\":{\"bgcolor\":\"rgb(17,17,17)\",\"lakecolor\":\"rgb(17,17,17)\",\"landcolor\":\"rgb(17,17,17)\",\"showlakes\":true,\"showland\":true,\"subunitcolor\":\"#506784\"},\"hoverlabel\":{\"align\":\"left\"},\"hovermode\":\"closest\",\"mapbox\":{\"style\":\"dark\"},\"paper_bgcolor\":\"rgb(17,17,17)\",\"plot_bgcolor\":\"rgb(17,17,17)\",\"polar\":{\"angularaxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"},\"bgcolor\":\"rgb(17,17,17)\",\"radialaxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"}},\"scene\":{\"xaxis\":{\"backgroundcolor\":\"rgb(17,17,17)\",\"gridcolor\":\"#506784\",\"gridwidth\":2,\"linecolor\":\"#506784\",\"showbackground\":true,\"ticks\":\"\",\"zerolinecolor\":\"#C8D4E3\"},\"yaxis\":{\"backgroundcolor\":\"rgb(17,17,17)\",\"gridcolor\":\"#506784\",\"gridwidth\":2,\"linecolor\":\"#506784\",\"showbackground\":true,\"ticks\":\"\",\"zerolinecolor\":\"#C8D4E3\"},\"zaxis\":{\"backgroundcolor\":\"rgb(17,17,17)\",\"gridcolor\":\"#506784\",\"gridwidth\":2,\"linecolor\":\"#506784\",\"showbackground\":true,\"ticks\":\"\",\"zerolinecolor\":\"#C8D4E3\"}},\"shapedefaults\":{\"line\":{\"color\":\"#f2f5fa\"}},\"sliderdefaults\":{\"bgcolor\":\"#C8D4E3\",\"bordercolor\":\"rgb(17,17,17)\",\"borderwidth\":1,\"tickwidth\":0},\"ternary\":{\"aaxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"},\"baxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"},\"bgcolor\":\"rgb(17,17,17)\",\"caxis\":{\"gridcolor\":\"#506784\",\"linecolor\":\"#506784\",\"ticks\":\"\"}},\"title\":{\"x\":0.05},\"updatemenudefaults\":{\"bgcolor\":\"#506784\",\"borderwidth\":0},\"xaxis\":{\"automargin\":true,\"gridcolor\":\"#283442\",\"linecolor\":\"#506784\",\"ticks\":\"\",\"title\":{\"standoff\":15},\"zerolinecolor\":\"#283442\",\"zerolinewidth\":2},\"yaxis\":{\"automargin\":true,\"gridcolor\":\"#283442\",\"linecolor\":\"#506784\",\"ticks\":\"\",\"title\":{\"standoff\":15},\"zerolinecolor\":\"#283442\",\"zerolinewidth\":2}}},\"xaxis\":{\"anchor\":\"y\",\"domain\":[0.0,1.0],\"title\":{\"text\":\"x\"},\"categoryorder\":\"total descending\"},\"yaxis\":{\"anchor\":\"x\",\"domain\":[0.0,1.0],\"title\":{\"text\":\"y\"}},\"legend\":{\"tracegroupgap\":0},\"margin\":{\"t\":60},\"barmode\":\"relative\",\"title\":{\"text\":\"Each countries distribution\"}},                        {\"responsive\": true}                    ).then(function(){\n",
              "                            \n",
              "var gd = document.getElementById('31c49557-b62a-4f44-8c76-3e91915ecb32');\n",
              "var x = new MutationObserver(function (mutations, observer) {{\n",
              "        var display = window.getComputedStyle(gd).display;\n",
              "        if (!display || display === 'none') {{\n",
              "            console.log([gd, 'removed!']);\n",
              "            Plotly.purge(gd);\n",
              "            observer.disconnect();\n",
              "        }}\n",
              "}});\n",
              "\n",
              "// Listen for the removal of the full notebook cells\n",
              "var notebookContainer = gd.closest('#notebook-container');\n",
              "if (notebookContainer) {{\n",
              "    x.observe(notebookContainer, {childList: true});\n",
              "}}\n",
              "\n",
              "// Listen for the clearing of the current output cell\n",
              "var outputEl = gd.closest('.output');\n",
              "if (outputEl) {{\n",
              "    x.observe(outputEl, {childList: true});\n",
              "}}\n",
              "\n",
              "                        })                };                            </script>        </div>\n",
              "</body>\n",
              "</html>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# **Completion**"
      ],
      "metadata": {
        "id": "Jz-ErCwmnkZJ"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Data Loader**"
      ],
      "metadata": {
        "id": "ygc3cWZlFoOv"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def load_data_completion(embedding, dim, stage):\n",
        "  emb = []\n",
        "  target = []\n",
        "  if stage == \"train\":\n",
        "    with open (\"train.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      for i in range(len(data)):\n",
        "        ingredients = data[i].strip().split(\",\")[:-1]\n",
        "        ingredients = [int(x) for x in ingredients]\n",
        "        \n",
        "        # Randomly pop one ingredient and make it target\n",
        "        random.shuffle(ingredients)\n",
        "        missing = ingredients.pop()\n",
        "        \n",
        "        if len(ingredients) != 0:\n",
        "          merge = np.zeros((len(ingredients), dim))\n",
        "          for j in range(len(ingredients)):\n",
        "            merge[j, :] = embedding.iloc[ingredients[j], :]\n",
        "          merge = np.mean(merge, axis=0)\n",
        "        else:\n",
        "          merge = np.zeros(dim)\n",
        "        emb.append(merge)\n",
        "        target.append(embedding.iloc[missing, :])\n",
        "    \n",
        "  elif stage == \"val\":\n",
        "    with open (\"validation_completion_answer.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      for i in range(len(data)):\n",
        "        target.append(embedding.iloc[int(data[i]), :])\n",
        "    \n",
        "    with open (\"validation_completion_question.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      \n",
        "      for i in range(len(data)):\n",
        "        ingredients = data[i].strip().split(\",\")\n",
        "        ingredients = [int(x) for x in ingredients]\n",
        "        merge = np.zeros((len(ingredients), dim))\n",
        "        \n",
        "        for j in range(len(ingredients)):\n",
        "          merge[j, :] = embedding.iloc[ingredients[j], :]\n",
        "        merge = np.mean(merge, axis=0)\n",
        "        emb.append(merge)\n",
        "    \n",
        "  elif stage == \"test\":\n",
        "    with open (\"test_completion_question.csv\", \"r\") as f:\n",
        "      data = f.readlines()\n",
        "      for i in range(len(data)):\n",
        "        ingredients = data[i].strip().split(\",\")\n",
        "        ingredients = [int(x) for x in ingredients]\n",
        "        merge = np.zeros((len(ingredients), dim))\n",
        "        \n",
        "        for j in range(len(ingredients)):\n",
        "          merge[j, :] = embedding.iloc[ingredients[j], :]\n",
        "        merge = np.mean(merge, axis=0) \n",
        "        emb.append(merge)\n",
        "        target.append(np.zeros(dim))\n",
        "\n",
        "  return np.vstack(emb), np.vstack(target)"
      ],
      "metadata": {
        "id": "blIjCSTpv1Aa"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Model**"
      ],
      "metadata": {
        "id": "Z1DKwXbEE1Ep"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "class Generator(nn.Module):\n",
        "  def __init__(self, num_feature, latent_dim):\n",
        "    super(Generator, self).__init__()    \n",
        "    self.fc1 = nn.Linear(num_feature + latent_dim, 64)\n",
        "    self.bn1 = nn.BatchNorm1d(64)\n",
        "    \n",
        "    self.fc2 = nn.Linear(64, 32)\n",
        "    self.bn2 = nn.BatchNorm1d(32)\n",
        "\n",
        "    self.fc3 = nn.Linear(32, 2*num_feature)\n",
        "    self.bn3 = nn.BatchNorm1d(2*num_feature)\n",
        "\n",
        "    self.dp = nn.Dropout(0.5)\n",
        "    self.act = nn.LeakyReLU(0.2)\n",
        "\n",
        "  def forward(self, x):\n",
        "    output = self.dp(self.act(self.bn1(self.fc1(x))))\n",
        "    output = self.dp(self.act(self.bn2(self.fc2(output))))\n",
        "    output = self.bn3(self.fc3(output))\n",
        "    return torch.tanh(output)\n",
        "\n",
        "class Discriminator(nn.Module):\n",
        "  def __init__(self, num_feature):\n",
        "    super(Discriminator, self).__init__()\n",
        "    self.fc1 = nn.Linear(2*num_feature, 512)\n",
        "    self.bn1 = nn.BatchNorm1d(512)\n",
        "    \n",
        "    self.fc2 = nn.Linear(512, 64)\n",
        "    self.bn2 = nn.BatchNorm1d(64)\n",
        "\n",
        "    self.fc3 = nn.Linear(64, 1)\n",
        "\n",
        "    self.dp = nn.Dropout(0.5)\n",
        "    self.act = nn.LeakyReLU(0.2)\n",
        "\n",
        "  def forward(self, x):\n",
        "    output = self.dp(self.act(self.bn1(self.fc1(x))))\n",
        "    output = self.dp(self.act(self.bn2(self.fc2(output))))\n",
        "    output = self.fc3(output)\n",
        "\n",
        "    return torch.sigmoid(output)\n",
        "\n",
        "\n",
        "class GANclassifier(nn.Module):\n",
        "  def __init__(self, num_feature):\n",
        "    super(GANclassifier, self).__init__()\n",
        "    self.fc1 = nn.Linear(2*num_feature, 512)\n",
        "    self.bn1 = nn.BatchNorm1d(512)\n",
        "    \n",
        "    self.fc2 = nn.Linear(512, 256)\n",
        "    self.bn2 = nn.BatchNorm1d(256)\n",
        "\n",
        "    self.fc3 = nn.Linear(256, num_feature)\n",
        "    self.bn3 = nn.BatchNorm1d(num_feature)\n",
        "\n",
        "    self.dp = nn.Dropout(0.5)\n",
        "    self.act = nn.ReLU()\n",
        "\n",
        "  def forward(self, x):\n",
        "    output = self.dp(self.act(self.bn1(self.fc1(x))))\n",
        "    output = self.dp(self.act(self.bn2(self.fc2(output))))\n",
        "    output = self.act(self.bn3(self.fc3(output)))\n",
        "\n",
        "    return output"
      ],
      "metadata": {
        "id": "QP8syoprwLhu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Training**"
      ],
      "metadata": {
        "id": "zPY0WexcFykq"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "x_train_com_dataset, y_train_com_dataset = load_data_completion(embeddings, args.pred_dim, \"train\")\n",
        "x_val_com_dataset, y_val_com_dataset = load_data_completion(embeddings, args.pred_dim, \"val\")\n",
        "x_test_com_dataset, y_test_com_dataset = load_data_completion(embeddings, args.pred_dim, \"test\")\n",
        "\n",
        "train_com_dataset = MyDataset(x_train_com_dataset, y_train_com_dataset)\n",
        "val_com_dataset = MyDataset(x_val_com_dataset, y_val_com_dataset)\n",
        "test_com_dataset = MyDataset(x_test_com_dataset, y_test_com_dataset)\n",
        "\n",
        "\n",
        "train_com_dataloader = torch.utils.data.DataLoader(train_com_dataset, batch_size=args.batch_size, shuffle=True, num_workers=2)\n",
        "val_com_dataloader = torch.utils.data.DataLoader(val_com_dataset, batch_size=args.batch_size, shuffle=False, num_workers=2)\n",
        "test_com_dataloader = torch.utils.data.DataLoader(test_com_dataset, batch_size=args.batch_size, shuffle=False, num_workers=2)"
      ],
      "metadata": {
        "id": "xA4Qo1OgnmDE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "classifier_loss = torch.nn.MSELoss()\n",
        "adversarial_loss = torch.nn.BCELoss()\n",
        "generator = Generator(args.pred_dim, args.latent_dim).to(device)\n",
        "discriminator = Discriminator(args.pred_dim).to(device)\n",
        "classifier = GANclassifier(args.pred_dim).to(device)"
      ],
      "metadata": {
        "id": "zlWU4iuMwEUN"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "optimizer_G = torch.optim.Adam(generator.parameters(), lr=args.pred_lr, betas=(args.b1, args.b2))\n",
        "optimizer_D = torch.optim.Adam(discriminator.parameters(), lr=args.pred_lr, betas=(args.b1, args.b2))\n",
        "optimizer_C = torch.optim.Adam(classifier.parameters(), lr=args.pred_lr, weight_decay=0.0001)\n",
        "scheduler_pred = torch.optim.lr_scheduler.MultiStepLR(optimizer_C, milestones=[50,80], gamma=0.5)"
      ],
      "metadata": {
        "id": "oQZUp1KSv4Ju"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "G_losses = []\n",
        "D_losses = []\n",
        "C_losses = []\n",
        "V_losses = []\n",
        "for epoch in range(args.pred_epochs):\n",
        "  for batch_idx, data in enumerate(train_com_dataloader):\n",
        "    recipe_emb = data['recipe'].to(device)\n",
        "    mis_recipe = data['label'].to(device)\n",
        "\n",
        "    valid_label = Variable(torch.FloatTensor(recipe_emb.shape[0], 1).fill_(1.0), requires_grad=False).to(device)\n",
        "    fake_label = Variable(torch.FloatTensor(recipe_emb.shape[0], 1).fill_(0.0), requires_grad=False).to(device)\n",
        "\n",
        "    noise = Variable(torch.FloatTensor(np.random.normal(0, 1, (recipe_emb.shape[0], args.latent_dim)))).to(device)\n",
        "    # noise = Variable(torch.FloatTensor(np.full((recipe_emb.shape[0], args.latent_dim), 0.001))).to(device)\n",
        "    fake = generator(torch.cat((noise, recipe_emb), dim=1))\n",
        "\n",
        "\n",
        "    optimizer_D.zero_grad()\n",
        "    optimizer_C.zero_grad()\n",
        "\n",
        "    real_score_out = discriminator(torch.cat((mis_recipe, recipe_emb), dim=1))\n",
        "    fake_score_out = discriminator(fake.detach())\n",
        "    real_embedding = classifier(torch.cat((mis_recipe, recipe_emb), dim=1))\n",
        "\n",
        "    D_real_loss = adversarial_loss(real_score_out, valid_label)\n",
        "    D_fake_loss = adversarial_loss(fake_score_out, fake_label)\n",
        "    C_real_loss = classifier_loss(real_embedding, mis_recipe)\n",
        "  \n",
        "    d_loss = (D_real_loss + D_fake_loss + C_real_loss) / 3\n",
        "    d_loss.backward()\n",
        "\n",
        "    optimizer_D.step()\n",
        "    optimizer_C.step()\n",
        "\n",
        "    # -----------------\n",
        "    #  Train Generator\n",
        "    # -----------------\n",
        "\n",
        "    optimizer_G.zero_grad()\n",
        "    \n",
        "    fake_score_out = discriminator(fake)\n",
        "    fake_embedding = classifier(fake)\n",
        "    \n",
        "    GD_fake_loss = adversarial_loss(fake_score_out, valid_label)\n",
        "    GC_fake_loss = classifier_loss(fake_embedding, fake[:, :args.pred_dim])\n",
        "    \n",
        "    g_loss = (GD_fake_loss + GC_fake_loss) / 2\n",
        "    g_loss.backward()\n",
        "    \n",
        "    optimizer_G.step()\n",
        "\n",
        "    if batch_idx % 50 == 0:\n",
        "      print(\n",
        "      \"[Epoch %d/%d] [Batch %d/%d] [D loss: %f] [G loss: %f] [C loss: %f]\"\n",
        "      % (epoch, args.pred_epochs, batch_idx, len(train_com_dataloader), d_loss.item(), g_loss.item(), C_real_loss.item() )\n",
        "      )\n",
        "\n",
        "  # Save Losses for plotting later\n",
        "  G_losses.append(g_loss.item())\n",
        "  D_losses.append(d_loss.item())\n",
        "  C_losses.append(C_real_loss.item())\n",
        "\n",
        "  generator.eval()\n",
        "  with torch.no_grad():\n",
        "    val_loss = []\n",
        "    for batch_idx, data in enumerate(val_com_dataloader):\n",
        "      recipe_emb = data['recipe'].to(device=device)\n",
        "      mis_recipe = data['label'].to(device=device)\n",
        "\n",
        "      #   Feed `x` into the network, get an output, and keep it in a variable called `logit`. \n",
        "      noise = Variable(torch.FloatTensor(np.random.normal(0, 1, (recipe_emb.shape[0], args.latent_dim)))).to(device)\n",
        "      generated = generator(torch.cat((noise, recipe_emb), dim=1))\n",
        "      generated = generated.type(torch.float).to(device)\n",
        "      partition = classifier(generated)\n",
        "\n",
        "      loss = classifier_loss(partition, mis_recipe) # generated[:, :args.pred_dim]\n",
        "      val_loss.append(loss)\n",
        "    \n",
        "    mean_val_loss = sum(val_loss) / len(val_loss)\n",
        "    V_losses.append(mean_val_loss.item())\n",
        "\n",
        "torch.save(generator.state_dict(), os.path.join(models_path, \"generator\"))\n",
        "torch.save(classifier.state_dict(), os.path.join(models_path, \"classifier\"))"
      ],
      "metadata": {
        "id": "d-n8IkrLLr03",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "65538139-9231-4516-d6a4-2c420ed6db09"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[Epoch 0/50] [Batch 0/368] [D loss: 0.663270] [G loss: 0.796260] [C loss: 0.591416]\n",
            "[Epoch 0/50] [Batch 50/368] [D loss: 0.653194] [G loss: 0.783017] [C loss: 0.572487]\n",
            "[Epoch 0/50] [Batch 100/368] [D loss: 0.677698] [G loss: 0.768337] [C loss: 0.592268]\n",
            "[Epoch 0/50] [Batch 150/368] [D loss: 0.675891] [G loss: 0.774235] [C loss: 0.559228]\n",
            "[Epoch 0/50] [Batch 200/368] [D loss: 0.660864] [G loss: 0.740915] [C loss: 0.581235]\n",
            "[Epoch 0/50] [Batch 250/368] [D loss: 0.665006] [G loss: 0.769989] [C loss: 0.572586]\n",
            "[Epoch 0/50] [Batch 300/368] [D loss: 0.661349] [G loss: 0.762727] [C loss: 0.556886]\n",
            "[Epoch 0/50] [Batch 350/368] [D loss: 0.668065] [G loss: 0.758962] [C loss: 0.557161]\n",
            "[Epoch 1/50] [Batch 0/368] [D loss: 0.664351] [G loss: 0.639335] [C loss: 0.570210]\n",
            "[Epoch 1/50] [Batch 50/368] [D loss: 0.677022] [G loss: 0.642521] [C loss: 0.566458]\n",
            "[Epoch 1/50] [Batch 100/368] [D loss: 0.660281] [G loss: 0.656414] [C loss: 0.553670]\n",
            "[Epoch 1/50] [Batch 150/368] [D loss: 0.674070] [G loss: 0.617805] [C loss: 0.555560]\n",
            "[Epoch 1/50] [Batch 200/368] [D loss: 0.661281] [G loss: 0.644717] [C loss: 0.552062]\n",
            "[Epoch 1/50] [Batch 250/368] [D loss: 0.671282] [G loss: 0.615661] [C loss: 0.535037]\n",
            "[Epoch 1/50] [Batch 300/368] [D loss: 0.671002] [G loss: 0.617143] [C loss: 0.539556]\n",
            "[Epoch 1/50] [Batch 350/368] [D loss: 0.654078] [G loss: 0.643196] [C loss: 0.544532]\n",
            "[Epoch 2/50] [Batch 0/368] [D loss: 0.656167] [G loss: 0.634770] [C loss: 0.538013]\n",
            "[Epoch 2/50] [Batch 50/368] [D loss: 0.649089] [G loss: 0.643968] [C loss: 0.536831]\n",
            "[Epoch 2/50] [Batch 100/368] [D loss: 0.659803] [G loss: 0.625053] [C loss: 0.527748]\n",
            "[Epoch 2/50] [Batch 150/368] [D loss: 0.649154] [G loss: 0.609759] [C loss: 0.544883]\n",
            "[Epoch 2/50] [Batch 200/368] [D loss: 0.663739] [G loss: 0.600039] [C loss: 0.539110]\n",
            "[Epoch 2/50] [Batch 250/368] [D loss: 0.664360] [G loss: 0.619201] [C loss: 0.522028]\n",
            "[Epoch 2/50] [Batch 300/368] [D loss: 0.655064] [G loss: 0.605794] [C loss: 0.543895]\n",
            "[Epoch 2/50] [Batch 350/368] [D loss: 0.665600] [G loss: 0.599400] [C loss: 0.518611]\n",
            "[Epoch 3/50] [Batch 0/368] [D loss: 0.654172] [G loss: 0.592834] [C loss: 0.513622]\n",
            "[Epoch 3/50] [Batch 50/368] [D loss: 0.661382] [G loss: 0.596110] [C loss: 0.507266]\n",
            "[Epoch 3/50] [Batch 100/368] [D loss: 0.648903] [G loss: 0.589786] [C loss: 0.501406]\n",
            "[Epoch 3/50] [Batch 150/368] [D loss: 0.651630] [G loss: 0.590824] [C loss: 0.522421]\n",
            "[Epoch 3/50] [Batch 200/368] [D loss: 0.650068] [G loss: 0.599022] [C loss: 0.490236]\n",
            "[Epoch 3/50] [Batch 250/368] [D loss: 0.645554] [G loss: 0.588765] [C loss: 0.485511]\n",
            "[Epoch 3/50] [Batch 300/368] [D loss: 0.652569] [G loss: 0.594403] [C loss: 0.489046]\n",
            "[Epoch 3/50] [Batch 350/368] [D loss: 0.645369] [G loss: 0.590976] [C loss: 0.510046]\n",
            "[Epoch 4/50] [Batch 0/368] [D loss: 0.623572] [G loss: 0.565489] [C loss: 0.483296]\n",
            "[Epoch 4/50] [Batch 50/368] [D loss: 0.645233] [G loss: 0.595097] [C loss: 0.457825]\n",
            "[Epoch 4/50] [Batch 100/368] [D loss: 0.608235] [G loss: 0.589132] [C loss: 0.444195]\n",
            "[Epoch 4/50] [Batch 150/368] [D loss: 0.608555] [G loss: 0.565666] [C loss: 0.430490]\n",
            "[Epoch 4/50] [Batch 200/368] [D loss: 0.637058] [G loss: 0.594300] [C loss: 0.502397]\n",
            "[Epoch 4/50] [Batch 250/368] [D loss: 0.627429] [G loss: 0.588584] [C loss: 0.446965]\n",
            "[Epoch 4/50] [Batch 300/368] [D loss: 0.610258] [G loss: 0.585708] [C loss: 0.428378]\n",
            "[Epoch 4/50] [Batch 350/368] [D loss: 0.621868] [G loss: 0.555442] [C loss: 0.431969]\n",
            "[Epoch 5/50] [Batch 0/368] [D loss: 0.598462] [G loss: 0.546987] [C loss: 0.407856]\n",
            "[Epoch 5/50] [Batch 50/368] [D loss: 0.600459] [G loss: 0.569221] [C loss: 0.428291]\n",
            "[Epoch 5/50] [Batch 100/368] [D loss: 0.627213] [G loss: 0.558658] [C loss: 0.435845]\n",
            "[Epoch 5/50] [Batch 150/368] [D loss: 0.612672] [G loss: 0.560059] [C loss: 0.400288]\n",
            "[Epoch 5/50] [Batch 200/368] [D loss: 0.610614] [G loss: 0.545861] [C loss: 0.394786]\n",
            "[Epoch 5/50] [Batch 250/368] [D loss: 0.585243] [G loss: 0.551560] [C loss: 0.327748]\n",
            "[Epoch 5/50] [Batch 300/368] [D loss: 0.599586] [G loss: 0.537602] [C loss: 0.365339]\n",
            "[Epoch 5/50] [Batch 350/368] [D loss: 0.578698] [G loss: 0.558317] [C loss: 0.329960]\n",
            "[Epoch 6/50] [Batch 0/368] [D loss: 0.581422] [G loss: 0.532787] [C loss: 0.361408]\n",
            "[Epoch 6/50] [Batch 50/368] [D loss: 0.601335] [G loss: 0.532045] [C loss: 0.387733]\n",
            "[Epoch 6/50] [Batch 100/368] [D loss: 0.589257] [G loss: 0.537803] [C loss: 0.346056]\n",
            "[Epoch 6/50] [Batch 150/368] [D loss: 0.580563] [G loss: 0.535542] [C loss: 0.320935]\n",
            "[Epoch 6/50] [Batch 200/368] [D loss: 0.568599] [G loss: 0.531194] [C loss: 0.350758]\n",
            "[Epoch 6/50] [Batch 250/368] [D loss: 0.600006] [G loss: 0.529073] [C loss: 0.379158]\n",
            "[Epoch 6/50] [Batch 300/368] [D loss: 0.567887] [G loss: 0.520097] [C loss: 0.294181]\n",
            "[Epoch 6/50] [Batch 350/368] [D loss: 0.609198] [G loss: 0.503988] [C loss: 0.385660]\n",
            "[Epoch 7/50] [Batch 0/368] [D loss: 0.578136] [G loss: 0.534209] [C loss: 0.292674]\n",
            "[Epoch 7/50] [Batch 50/368] [D loss: 0.563472] [G loss: 0.499627] [C loss: 0.282837]\n",
            "[Epoch 7/50] [Batch 100/368] [D loss: 0.569455] [G loss: 0.492386] [C loss: 0.309205]\n",
            "[Epoch 7/50] [Batch 150/368] [D loss: 0.589157] [G loss: 0.499075] [C loss: 0.348172]\n",
            "[Epoch 7/50] [Batch 200/368] [D loss: 0.554247] [G loss: 0.480907] [C loss: 0.258571]\n",
            "[Epoch 7/50] [Batch 250/368] [D loss: 0.561826] [G loss: 0.474376] [C loss: 0.235610]\n",
            "[Epoch 7/50] [Batch 300/368] [D loss: 0.537721] [G loss: 0.478457] [C loss: 0.201947]\n",
            "[Epoch 7/50] [Batch 350/368] [D loss: 0.527466] [G loss: 0.487522] [C loss: 0.213768]\n",
            "[Epoch 8/50] [Batch 0/368] [D loss: 0.561983] [G loss: 0.491342] [C loss: 0.245160]\n",
            "[Epoch 8/50] [Batch 50/368] [D loss: 0.541438] [G loss: 0.485436] [C loss: 0.196988]\n",
            "[Epoch 8/50] [Batch 100/368] [D loss: 0.572203] [G loss: 0.481932] [C loss: 0.293207]\n",
            "[Epoch 8/50] [Batch 150/368] [D loss: 0.529371] [G loss: 0.491958] [C loss: 0.171725]\n",
            "[Epoch 8/50] [Batch 200/368] [D loss: 0.517002] [G loss: 0.433020] [C loss: 0.163342]\n",
            "[Epoch 8/50] [Batch 250/368] [D loss: 0.537242] [G loss: 0.482652] [C loss: 0.191029]\n",
            "[Epoch 8/50] [Batch 300/368] [D loss: 0.524939] [G loss: 0.447397] [C loss: 0.180109]\n",
            "[Epoch 8/50] [Batch 350/368] [D loss: 0.542759] [G loss: 0.464972] [C loss: 0.190858]\n",
            "[Epoch 9/50] [Batch 0/368] [D loss: 0.509514] [G loss: 0.461187] [C loss: 0.153141]\n",
            "[Epoch 9/50] [Batch 50/368] [D loss: 0.528413] [G loss: 0.450442] [C loss: 0.185193]\n",
            "[Epoch 9/50] [Batch 100/368] [D loss: 0.556208] [G loss: 0.451593] [C loss: 0.256771]\n",
            "[Epoch 9/50] [Batch 150/368] [D loss: 0.529751] [G loss: 0.454258] [C loss: 0.193183]\n",
            "[Epoch 9/50] [Batch 200/368] [D loss: 0.522778] [G loss: 0.449805] [C loss: 0.172957]\n",
            "[Epoch 9/50] [Batch 250/368] [D loss: 0.516311] [G loss: 0.457336] [C loss: 0.169211]\n",
            "[Epoch 9/50] [Batch 300/368] [D loss: 0.515182] [G loss: 0.452537] [C loss: 0.172290]\n",
            "[Epoch 9/50] [Batch 350/368] [D loss: 0.524528] [G loss: 0.435525] [C loss: 0.169183]\n",
            "[Epoch 10/50] [Batch 0/368] [D loss: 0.535021] [G loss: 0.448612] [C loss: 0.171591]\n",
            "[Epoch 10/50] [Batch 50/368] [D loss: 0.518378] [G loss: 0.429772] [C loss: 0.147738]\n",
            "[Epoch 10/50] [Batch 100/368] [D loss: 0.557351] [G loss: 0.436988] [C loss: 0.277983]\n",
            "[Epoch 10/50] [Batch 150/368] [D loss: 0.530878] [G loss: 0.421381] [C loss: 0.164941]\n",
            "[Epoch 10/50] [Batch 200/368] [D loss: 0.536178] [G loss: 0.432894] [C loss: 0.205057]\n",
            "[Epoch 10/50] [Batch 250/368] [D loss: 0.515158] [G loss: 0.410048] [C loss: 0.175647]\n",
            "[Epoch 10/50] [Batch 300/368] [D loss: 0.496863] [G loss: 0.419978] [C loss: 0.140724]\n",
            "[Epoch 10/50] [Batch 350/368] [D loss: 0.523118] [G loss: 0.429872] [C loss: 0.167884]\n",
            "[Epoch 11/50] [Batch 0/368] [D loss: 0.524722] [G loss: 0.430598] [C loss: 0.166794]\n",
            "[Epoch 11/50] [Batch 50/368] [D loss: 0.518988] [G loss: 0.423361] [C loss: 0.150253]\n",
            "[Epoch 11/50] [Batch 100/368] [D loss: 0.534667] [G loss: 0.443445] [C loss: 0.150272]\n",
            "[Epoch 11/50] [Batch 150/368] [D loss: 0.519258] [G loss: 0.401575] [C loss: 0.200749]\n",
            "[Epoch 11/50] [Batch 200/368] [D loss: 0.528930] [G loss: 0.458619] [C loss: 0.164341]\n",
            "[Epoch 11/50] [Batch 250/368] [D loss: 0.546766] [G loss: 0.422366] [C loss: 0.202980]\n",
            "[Epoch 11/50] [Batch 300/368] [D loss: 0.502427] [G loss: 0.427178] [C loss: 0.140532]\n",
            "[Epoch 11/50] [Batch 350/368] [D loss: 0.560761] [G loss: 0.414025] [C loss: 0.279693]\n",
            "[Epoch 12/50] [Batch 0/368] [D loss: 0.545439] [G loss: 0.423629] [C loss: 0.186864]\n",
            "[Epoch 12/50] [Batch 50/368] [D loss: 0.511783] [G loss: 0.421118] [C loss: 0.130594]\n",
            "[Epoch 12/50] [Batch 100/368] [D loss: 0.518843] [G loss: 0.418643] [C loss: 0.142096]\n",
            "[Epoch 12/50] [Batch 150/368] [D loss: 0.522946] [G loss: 0.412436] [C loss: 0.142421]\n",
            "[Epoch 12/50] [Batch 200/368] [D loss: 0.534113] [G loss: 0.420968] [C loss: 0.139913]\n",
            "[Epoch 12/50] [Batch 250/368] [D loss: 0.547077] [G loss: 0.423442] [C loss: 0.249165]\n",
            "[Epoch 12/50] [Batch 300/368] [D loss: 0.533244] [G loss: 0.424356] [C loss: 0.180321]\n",
            "[Epoch 12/50] [Batch 350/368] [D loss: 0.491221] [G loss: 0.401879] [C loss: 0.117791]\n",
            "[Epoch 13/50] [Batch 0/368] [D loss: 0.548160] [G loss: 0.422448] [C loss: 0.221979]\n",
            "[Epoch 13/50] [Batch 50/368] [D loss: 0.517533] [G loss: 0.407448] [C loss: 0.149834]\n",
            "[Epoch 13/50] [Batch 100/368] [D loss: 0.519509] [G loss: 0.404166] [C loss: 0.140425]\n",
            "[Epoch 13/50] [Batch 150/368] [D loss: 0.529864] [G loss: 0.417172] [C loss: 0.198383]\n",
            "[Epoch 13/50] [Batch 200/368] [D loss: 0.496428] [G loss: 0.414819] [C loss: 0.129760]\n",
            "[Epoch 13/50] [Batch 250/368] [D loss: 0.525353] [G loss: 0.412265] [C loss: 0.178468]\n",
            "[Epoch 13/50] [Batch 300/368] [D loss: 0.501338] [G loss: 0.400041] [C loss: 0.114752]\n",
            "[Epoch 13/50] [Batch 350/368] [D loss: 0.512821] [G loss: 0.422310] [C loss: 0.120563]\n",
            "[Epoch 14/50] [Batch 0/368] [D loss: 0.500651] [G loss: 0.410921] [C loss: 0.121028]\n",
            "[Epoch 14/50] [Batch 50/368] [D loss: 0.524666] [G loss: 0.400440] [C loss: 0.115868]\n",
            "[Epoch 14/50] [Batch 100/368] [D loss: 0.527347] [G loss: 0.424021] [C loss: 0.155962]\n",
            "[Epoch 14/50] [Batch 150/368] [D loss: 0.506909] [G loss: 0.398051] [C loss: 0.118032]\n",
            "[Epoch 14/50] [Batch 200/368] [D loss: 0.509435] [G loss: 0.406963] [C loss: 0.127598]\n",
            "[Epoch 14/50] [Batch 250/368] [D loss: 0.497340] [G loss: 0.412495] [C loss: 0.108011]\n",
            "[Epoch 14/50] [Batch 300/368] [D loss: 0.521719] [G loss: 0.407650] [C loss: 0.133136]\n",
            "[Epoch 14/50] [Batch 350/368] [D loss: 0.498240] [G loss: 0.410621] [C loss: 0.110328]\n",
            "[Epoch 15/50] [Batch 0/368] [D loss: 0.534636] [G loss: 0.394317] [C loss: 0.235693]\n",
            "[Epoch 15/50] [Batch 50/368] [D loss: 0.505659] [G loss: 0.400351] [C loss: 0.106801]\n",
            "[Epoch 15/50] [Batch 100/368] [D loss: 0.502281] [G loss: 0.399040] [C loss: 0.127263]\n",
            "[Epoch 15/50] [Batch 150/368] [D loss: 0.500053] [G loss: 0.401017] [C loss: 0.121610]\n",
            "[Epoch 15/50] [Batch 200/368] [D loss: 0.502700] [G loss: 0.394961] [C loss: 0.115905]\n",
            "[Epoch 15/50] [Batch 250/368] [D loss: 0.512717] [G loss: 0.405775] [C loss: 0.148686]\n",
            "[Epoch 15/50] [Batch 300/368] [D loss: 0.547183] [G loss: 0.405073] [C loss: 0.210882]\n",
            "[Epoch 15/50] [Batch 350/368] [D loss: 0.518678] [G loss: 0.406906] [C loss: 0.144983]\n",
            "[Epoch 16/50] [Batch 0/368] [D loss: 0.532049] [G loss: 0.392119] [C loss: 0.190088]\n",
            "[Epoch 16/50] [Batch 50/368] [D loss: 0.518953] [G loss: 0.404449] [C loss: 0.142399]\n",
            "[Epoch 16/50] [Batch 100/368] [D loss: 0.509377] [G loss: 0.389701] [C loss: 0.099089]\n",
            "[Epoch 16/50] [Batch 150/368] [D loss: 0.505767] [G loss: 0.396775] [C loss: 0.117088]\n",
            "[Epoch 16/50] [Batch 200/368] [D loss: 0.518138] [G loss: 0.409935] [C loss: 0.162433]\n",
            "[Epoch 16/50] [Batch 250/368] [D loss: 0.495592] [G loss: 0.383081] [C loss: 0.113050]\n",
            "[Epoch 16/50] [Batch 300/368] [D loss: 0.517315] [G loss: 0.396055] [C loss: 0.124950]\n",
            "[Epoch 16/50] [Batch 350/368] [D loss: 0.503131] [G loss: 0.382886] [C loss: 0.134496]\n",
            "[Epoch 17/50] [Batch 0/368] [D loss: 0.516991] [G loss: 0.408873] [C loss: 0.131733]\n",
            "[Epoch 17/50] [Batch 50/368] [D loss: 0.523355] [G loss: 0.401277] [C loss: 0.135914]\n",
            "[Epoch 17/50] [Batch 100/368] [D loss: 0.507460] [G loss: 0.386021] [C loss: 0.146215]\n",
            "[Epoch 17/50] [Batch 150/368] [D loss: 0.509705] [G loss: 0.388733] [C loss: 0.117179]\n",
            "[Epoch 17/50] [Batch 200/368] [D loss: 0.531180] [G loss: 0.408008] [C loss: 0.207382]\n",
            "[Epoch 17/50] [Batch 250/368] [D loss: 0.508598] [G loss: 0.396989] [C loss: 0.132840]\n",
            "[Epoch 17/50] [Batch 300/368] [D loss: 0.512582] [G loss: 0.389379] [C loss: 0.118292]\n",
            "[Epoch 17/50] [Batch 350/368] [D loss: 0.489816] [G loss: 0.393267] [C loss: 0.099064]\n",
            "[Epoch 18/50] [Batch 0/368] [D loss: 0.532032] [G loss: 0.396054] [C loss: 0.203964]\n",
            "[Epoch 18/50] [Batch 50/368] [D loss: 0.508599] [G loss: 0.398763] [C loss: 0.120252]\n",
            "[Epoch 18/50] [Batch 100/368] [D loss: 0.499209] [G loss: 0.392299] [C loss: 0.088646]\n",
            "[Epoch 18/50] [Batch 150/368] [D loss: 0.501080] [G loss: 0.390671] [C loss: 0.105879]\n",
            "[Epoch 18/50] [Batch 200/368] [D loss: 0.491233] [G loss: 0.389434] [C loss: 0.118840]\n",
            "[Epoch 18/50] [Batch 250/368] [D loss: 0.506230] [G loss: 0.383863] [C loss: 0.110088]\n",
            "[Epoch 18/50] [Batch 300/368] [D loss: 0.495582] [G loss: 0.399556] [C loss: 0.102230]\n",
            "[Epoch 18/50] [Batch 350/368] [D loss: 0.502275] [G loss: 0.382192] [C loss: 0.108253]\n",
            "[Epoch 19/50] [Batch 0/368] [D loss: 0.496786] [G loss: 0.388533] [C loss: 0.111913]\n",
            "[Epoch 19/50] [Batch 50/368] [D loss: 0.520185] [G loss: 0.391395] [C loss: 0.119486]\n",
            "[Epoch 19/50] [Batch 100/368] [D loss: 0.507053] [G loss: 0.394934] [C loss: 0.097593]\n",
            "[Epoch 19/50] [Batch 150/368] [D loss: 0.501991] [G loss: 0.389406] [C loss: 0.160605]\n",
            "[Epoch 19/50] [Batch 200/368] [D loss: 0.539553] [G loss: 0.382142] [C loss: 0.187890]\n",
            "[Epoch 19/50] [Batch 250/368] [D loss: 0.499971] [G loss: 0.396678] [C loss: 0.111687]\n",
            "[Epoch 19/50] [Batch 300/368] [D loss: 0.510480] [G loss: 0.388635] [C loss: 0.118692]\n",
            "[Epoch 19/50] [Batch 350/368] [D loss: 0.504642] [G loss: 0.377212] [C loss: 0.114432]\n",
            "[Epoch 20/50] [Batch 0/368] [D loss: 0.512731] [G loss: 0.381533] [C loss: 0.116834]\n",
            "[Epoch 20/50] [Batch 50/368] [D loss: 0.484716] [G loss: 0.377431] [C loss: 0.091900]\n",
            "[Epoch 20/50] [Batch 100/368] [D loss: 0.494347] [G loss: 0.386154] [C loss: 0.099912]\n",
            "[Epoch 20/50] [Batch 150/368] [D loss: 0.519729] [G loss: 0.385014] [C loss: 0.140010]\n",
            "[Epoch 20/50] [Batch 200/368] [D loss: 0.491057] [G loss: 0.393738] [C loss: 0.113823]\n",
            "[Epoch 20/50] [Batch 250/368] [D loss: 0.510952] [G loss: 0.379273] [C loss: 0.158832]\n",
            "[Epoch 20/50] [Batch 300/368] [D loss: 0.505990] [G loss: 0.399791] [C loss: 0.118907]\n",
            "[Epoch 20/50] [Batch 350/368] [D loss: 0.495762] [G loss: 0.398369] [C loss: 0.093328]\n",
            "[Epoch 21/50] [Batch 0/368] [D loss: 0.485870] [G loss: 0.386443] [C loss: 0.081669]\n",
            "[Epoch 21/50] [Batch 50/368] [D loss: 0.489609] [G loss: 0.385397] [C loss: 0.119860]\n",
            "[Epoch 21/50] [Batch 100/368] [D loss: 0.544856] [G loss: 0.379864] [C loss: 0.239116]\n",
            "[Epoch 21/50] [Batch 150/368] [D loss: 0.501155] [G loss: 0.386850] [C loss: 0.108098]\n",
            "[Epoch 21/50] [Batch 200/368] [D loss: 0.510051] [G loss: 0.390710] [C loss: 0.123603]\n",
            "[Epoch 21/50] [Batch 250/368] [D loss: 0.489984] [G loss: 0.381552] [C loss: 0.101036]\n",
            "[Epoch 21/50] [Batch 300/368] [D loss: 0.499847] [G loss: 0.379568] [C loss: 0.104982]\n",
            "[Epoch 21/50] [Batch 350/368] [D loss: 0.500114] [G loss: 0.386747] [C loss: 0.116387]\n",
            "[Epoch 22/50] [Batch 0/368] [D loss: 0.504714] [G loss: 0.381133] [C loss: 0.113332]\n",
            "[Epoch 22/50] [Batch 50/368] [D loss: 0.502149] [G loss: 0.381923] [C loss: 0.108414]\n",
            "[Epoch 22/50] [Batch 100/368] [D loss: 0.502610] [G loss: 0.393980] [C loss: 0.094945]\n",
            "[Epoch 22/50] [Batch 150/368] [D loss: 0.488070] [G loss: 0.380036] [C loss: 0.110354]\n",
            "[Epoch 22/50] [Batch 200/368] [D loss: 0.515923] [G loss: 0.391673] [C loss: 0.113285]\n",
            "[Epoch 22/50] [Batch 250/368] [D loss: 0.475938] [G loss: 0.384120] [C loss: 0.086898]\n",
            "[Epoch 22/50] [Batch 300/368] [D loss: 0.496760] [G loss: 0.379563] [C loss: 0.101454]\n",
            "[Epoch 22/50] [Batch 350/368] [D loss: 0.496884] [G loss: 0.393693] [C loss: 0.087519]\n",
            "[Epoch 23/50] [Batch 0/368] [D loss: 0.509419] [G loss: 0.381851] [C loss: 0.150225]\n",
            "[Epoch 23/50] [Batch 50/368] [D loss: 0.495453] [G loss: 0.362596] [C loss: 0.095124]\n",
            "[Epoch 23/50] [Batch 100/368] [D loss: 0.521852] [G loss: 0.382284] [C loss: 0.168952]\n",
            "[Epoch 23/50] [Batch 150/368] [D loss: 0.487821] [G loss: 0.396548] [C loss: 0.088364]\n",
            "[Epoch 23/50] [Batch 200/368] [D loss: 0.491873] [G loss: 0.376111] [C loss: 0.091375]\n",
            "[Epoch 23/50] [Batch 250/368] [D loss: 0.492779] [G loss: 0.388864] [C loss: 0.120841]\n",
            "[Epoch 23/50] [Batch 300/368] [D loss: 0.560549] [G loss: 0.392229] [C loss: 0.256160]\n",
            "[Epoch 23/50] [Batch 350/368] [D loss: 0.502470] [G loss: 0.395615] [C loss: 0.085820]\n",
            "[Epoch 24/50] [Batch 0/368] [D loss: 0.491172] [G loss: 0.388639] [C loss: 0.091223]\n",
            "[Epoch 24/50] [Batch 50/368] [D loss: 0.507297] [G loss: 0.372709] [C loss: 0.108077]\n",
            "[Epoch 24/50] [Batch 100/368] [D loss: 0.497526] [G loss: 0.386179] [C loss: 0.094615]\n",
            "[Epoch 24/50] [Batch 150/368] [D loss: 0.504270] [G loss: 0.387641] [C loss: 0.127128]\n",
            "[Epoch 24/50] [Batch 200/368] [D loss: 0.492033] [G loss: 0.369970] [C loss: 0.085088]\n",
            "[Epoch 24/50] [Batch 250/368] [D loss: 0.496736] [G loss: 0.371555] [C loss: 0.113457]\n",
            "[Epoch 24/50] [Batch 300/368] [D loss: 0.497536] [G loss: 0.379179] [C loss: 0.099387]\n",
            "[Epoch 24/50] [Batch 350/368] [D loss: 0.507939] [G loss: 0.387019] [C loss: 0.137700]\n",
            "[Epoch 25/50] [Batch 0/368] [D loss: 0.515019] [G loss: 0.371372] [C loss: 0.121296]\n",
            "[Epoch 25/50] [Batch 50/368] [D loss: 0.483815] [G loss: 0.382357] [C loss: 0.091041]\n",
            "[Epoch 25/50] [Batch 100/368] [D loss: 0.516914] [G loss: 0.388312] [C loss: 0.144916]\n",
            "[Epoch 25/50] [Batch 150/368] [D loss: 0.492056] [G loss: 0.379788] [C loss: 0.095932]\n",
            "[Epoch 25/50] [Batch 200/368] [D loss: 0.489545] [G loss: 0.374890] [C loss: 0.102820]\n",
            "[Epoch 25/50] [Batch 250/368] [D loss: 0.495387] [G loss: 0.384592] [C loss: 0.101511]\n",
            "[Epoch 25/50] [Batch 300/368] [D loss: 0.486410] [G loss: 0.371673] [C loss: 0.078833]\n",
            "[Epoch 25/50] [Batch 350/368] [D loss: 0.498112] [G loss: 0.377168] [C loss: 0.098288]\n",
            "[Epoch 26/50] [Batch 0/368] [D loss: 0.492164] [G loss: 0.384265] [C loss: 0.087213]\n",
            "[Epoch 26/50] [Batch 50/368] [D loss: 0.507150] [G loss: 0.384087] [C loss: 0.113945]\n",
            "[Epoch 26/50] [Batch 100/368] [D loss: 0.482432] [G loss: 0.380495] [C loss: 0.079922]\n",
            "[Epoch 26/50] [Batch 150/368] [D loss: 0.516124] [G loss: 0.383670] [C loss: 0.151592]\n",
            "[Epoch 26/50] [Batch 200/368] [D loss: 0.499870] [G loss: 0.390567] [C loss: 0.104396]\n",
            "[Epoch 26/50] [Batch 250/368] [D loss: 0.503946] [G loss: 0.377354] [C loss: 0.118326]\n",
            "[Epoch 26/50] [Batch 300/368] [D loss: 0.485027] [G loss: 0.378771] [C loss: 0.089134]\n",
            "[Epoch 26/50] [Batch 350/368] [D loss: 0.496926] [G loss: 0.377361] [C loss: 0.095004]\n",
            "[Epoch 27/50] [Batch 0/368] [D loss: 0.500445] [G loss: 0.389765] [C loss: 0.102275]\n",
            "[Epoch 27/50] [Batch 50/368] [D loss: 0.491257] [G loss: 0.386362] [C loss: 0.096544]\n",
            "[Epoch 27/50] [Batch 100/368] [D loss: 0.486299] [G loss: 0.366731] [C loss: 0.090937]\n",
            "[Epoch 27/50] [Batch 150/368] [D loss: 0.488939] [G loss: 0.377816] [C loss: 0.088588]\n",
            "[Epoch 27/50] [Batch 200/368] [D loss: 0.492924] [G loss: 0.377127] [C loss: 0.100229]\n",
            "[Epoch 27/50] [Batch 250/368] [D loss: 0.487251] [G loss: 0.373933] [C loss: 0.089237]\n",
            "[Epoch 27/50] [Batch 300/368] [D loss: 0.509499] [G loss: 0.378608] [C loss: 0.107076]\n",
            "[Epoch 27/50] [Batch 350/368] [D loss: 0.493719] [G loss: 0.379876] [C loss: 0.126636]\n",
            "[Epoch 28/50] [Batch 0/368] [D loss: 0.484200] [G loss: 0.372170] [C loss: 0.081617]\n",
            "[Epoch 28/50] [Batch 50/368] [D loss: 0.504170] [G loss: 0.370991] [C loss: 0.111651]\n",
            "[Epoch 28/50] [Batch 100/368] [D loss: 0.485407] [G loss: 0.359957] [C loss: 0.085028]\n",
            "[Epoch 28/50] [Batch 150/368] [D loss: 0.482592] [G loss: 0.365143] [C loss: 0.106671]\n",
            "[Epoch 28/50] [Batch 200/368] [D loss: 0.518377] [G loss: 0.381966] [C loss: 0.127330]\n",
            "[Epoch 28/50] [Batch 250/368] [D loss: 0.495300] [G loss: 0.374694] [C loss: 0.105111]\n",
            "[Epoch 28/50] [Batch 300/368] [D loss: 0.500256] [G loss: 0.375249] [C loss: 0.097017]\n",
            "[Epoch 28/50] [Batch 350/368] [D loss: 0.493464] [G loss: 0.379650] [C loss: 0.080110]\n",
            "[Epoch 29/50] [Batch 0/368] [D loss: 0.486866] [G loss: 0.371006] [C loss: 0.091750]\n",
            "[Epoch 29/50] [Batch 50/368] [D loss: 0.501415] [G loss: 0.380770] [C loss: 0.096040]\n",
            "[Epoch 29/50] [Batch 100/368] [D loss: 0.483631] [G loss: 0.375863] [C loss: 0.081443]\n",
            "[Epoch 29/50] [Batch 150/368] [D loss: 0.525073] [G loss: 0.382345] [C loss: 0.167539]\n",
            "[Epoch 29/50] [Batch 200/368] [D loss: 0.484877] [G loss: 0.372379] [C loss: 0.084044]\n",
            "[Epoch 29/50] [Batch 250/368] [D loss: 0.495239] [G loss: 0.375591] [C loss: 0.086289]\n",
            "[Epoch 29/50] [Batch 300/368] [D loss: 0.496074] [G loss: 0.373806] [C loss: 0.093411]\n",
            "[Epoch 29/50] [Batch 350/368] [D loss: 0.501775] [G loss: 0.370782] [C loss: 0.113722]\n",
            "[Epoch 30/50] [Batch 0/368] [D loss: 0.483229] [G loss: 0.387721] [C loss: 0.077419]\n",
            "[Epoch 30/50] [Batch 50/368] [D loss: 0.486690] [G loss: 0.383318] [C loss: 0.086995]\n",
            "[Epoch 30/50] [Batch 100/368] [D loss: 0.475114] [G loss: 0.379029] [C loss: 0.099670]\n",
            "[Epoch 30/50] [Batch 150/368] [D loss: 0.509917] [G loss: 0.375198] [C loss: 0.114650]\n",
            "[Epoch 30/50] [Batch 200/368] [D loss: 0.502920] [G loss: 0.377102] [C loss: 0.120979]\n",
            "[Epoch 30/50] [Batch 250/368] [D loss: 0.490874] [G loss: 0.392301] [C loss: 0.094761]\n",
            "[Epoch 30/50] [Batch 300/368] [D loss: 0.484902] [G loss: 0.390035] [C loss: 0.079055]\n",
            "[Epoch 30/50] [Batch 350/368] [D loss: 0.474868] [G loss: 0.367295] [C loss: 0.077450]\n",
            "[Epoch 31/50] [Batch 0/368] [D loss: 0.499161] [G loss: 0.372909] [C loss: 0.082364]\n",
            "[Epoch 31/50] [Batch 50/368] [D loss: 0.494761] [G loss: 0.373482] [C loss: 0.090983]\n",
            "[Epoch 31/50] [Batch 100/368] [D loss: 0.493166] [G loss: 0.377966] [C loss: 0.108728]\n",
            "[Epoch 31/50] [Batch 150/368] [D loss: 0.481274] [G loss: 0.373305] [C loss: 0.092364]\n",
            "[Epoch 31/50] [Batch 200/368] [D loss: 0.495349] [G loss: 0.367240] [C loss: 0.106816]\n",
            "[Epoch 31/50] [Batch 250/368] [D loss: 0.485231] [G loss: 0.370365] [C loss: 0.102096]\n",
            "[Epoch 31/50] [Batch 300/368] [D loss: 0.480705] [G loss: 0.376873] [C loss: 0.090602]\n",
            "[Epoch 31/50] [Batch 350/368] [D loss: 0.488396] [G loss: 0.372476] [C loss: 0.090342]\n",
            "[Epoch 32/50] [Batch 0/368] [D loss: 0.486608] [G loss: 0.375716] [C loss: 0.074613]\n",
            "[Epoch 32/50] [Batch 50/368] [D loss: 0.486637] [G loss: 0.370929] [C loss: 0.084471]\n",
            "[Epoch 32/50] [Batch 100/368] [D loss: 0.478438] [G loss: 0.381274] [C loss: 0.089047]\n",
            "[Epoch 32/50] [Batch 150/368] [D loss: 0.485458] [G loss: 0.379874] [C loss: 0.086649]\n",
            "[Epoch 32/50] [Batch 200/368] [D loss: 0.480795] [G loss: 0.382748] [C loss: 0.080578]\n",
            "[Epoch 32/50] [Batch 250/368] [D loss: 0.494126] [G loss: 0.366431] [C loss: 0.084854]\n",
            "[Epoch 32/50] [Batch 300/368] [D loss: 0.481710] [G loss: 0.382640] [C loss: 0.076859]\n",
            "[Epoch 32/50] [Batch 350/368] [D loss: 0.488719] [G loss: 0.377820] [C loss: 0.093494]\n",
            "[Epoch 33/50] [Batch 0/368] [D loss: 0.478690] [G loss: 0.377090] [C loss: 0.070235]\n",
            "[Epoch 33/50] [Batch 50/368] [D loss: 0.483392] [G loss: 0.389072] [C loss: 0.083271]\n",
            "[Epoch 33/50] [Batch 100/368] [D loss: 0.486119] [G loss: 0.376438] [C loss: 0.088768]\n",
            "[Epoch 33/50] [Batch 150/368] [D loss: 0.500719] [G loss: 0.385496] [C loss: 0.116714]\n",
            "[Epoch 33/50] [Batch 200/368] [D loss: 0.487539] [G loss: 0.381241] [C loss: 0.082414]\n",
            "[Epoch 33/50] [Batch 250/368] [D loss: 0.495738] [G loss: 0.373131] [C loss: 0.109593]\n",
            "[Epoch 33/50] [Batch 300/368] [D loss: 0.490523] [G loss: 0.362402] [C loss: 0.072361]\n",
            "[Epoch 33/50] [Batch 350/368] [D loss: 0.480200] [G loss: 0.388058] [C loss: 0.078548]\n",
            "[Epoch 34/50] [Batch 0/368] [D loss: 0.478616] [G loss: 0.365658] [C loss: 0.073180]\n",
            "[Epoch 34/50] [Batch 50/368] [D loss: 0.491126] [G loss: 0.376769] [C loss: 0.088720]\n",
            "[Epoch 34/50] [Batch 100/368] [D loss: 0.475437] [G loss: 0.371789] [C loss: 0.078485]\n",
            "[Epoch 34/50] [Batch 150/368] [D loss: 0.489707] [G loss: 0.378658] [C loss: 0.074199]\n",
            "[Epoch 34/50] [Batch 200/368] [D loss: 0.482886] [G loss: 0.371166] [C loss: 0.073990]\n",
            "[Epoch 34/50] [Batch 250/368] [D loss: 0.494512] [G loss: 0.371207] [C loss: 0.077313]\n",
            "[Epoch 34/50] [Batch 300/368] [D loss: 0.500010] [G loss: 0.365021] [C loss: 0.094927]\n",
            "[Epoch 34/50] [Batch 350/368] [D loss: 0.504355] [G loss: 0.366640] [C loss: 0.116585]\n",
            "[Epoch 35/50] [Batch 0/368] [D loss: 0.486872] [G loss: 0.365728] [C loss: 0.071145]\n",
            "[Epoch 35/50] [Batch 50/368] [D loss: 0.491866] [G loss: 0.389477] [C loss: 0.095612]\n",
            "[Epoch 35/50] [Batch 100/368] [D loss: 0.484932] [G loss: 0.372221] [C loss: 0.090717]\n",
            "[Epoch 35/50] [Batch 150/368] [D loss: 0.493691] [G loss: 0.398583] [C loss: 0.107662]\n",
            "[Epoch 35/50] [Batch 200/368] [D loss: 0.486265] [G loss: 0.372870] [C loss: 0.077512]\n",
            "[Epoch 35/50] [Batch 250/368] [D loss: 0.488494] [G loss: 0.373115] [C loss: 0.090182]\n",
            "[Epoch 35/50] [Batch 300/368] [D loss: 0.482615] [G loss: 0.375488] [C loss: 0.100557]\n",
            "[Epoch 35/50] [Batch 350/368] [D loss: 0.480000] [G loss: 0.378374] [C loss: 0.070273]\n",
            "[Epoch 36/50] [Batch 0/368] [D loss: 0.481916] [G loss: 0.370552] [C loss: 0.077101]\n",
            "[Epoch 36/50] [Batch 50/368] [D loss: 0.476928] [G loss: 0.373765] [C loss: 0.102127]\n",
            "[Epoch 36/50] [Batch 100/368] [D loss: 0.496443] [G loss: 0.378166] [C loss: 0.119869]\n",
            "[Epoch 36/50] [Batch 150/368] [D loss: 0.531077] [G loss: 0.389120] [C loss: 0.196854]\n",
            "[Epoch 36/50] [Batch 200/368] [D loss: 0.482957] [G loss: 0.360391] [C loss: 0.078681]\n",
            "[Epoch 36/50] [Batch 250/368] [D loss: 0.487143] [G loss: 0.373591] [C loss: 0.105342]\n",
            "[Epoch 36/50] [Batch 300/368] [D loss: 0.475888] [G loss: 0.370646] [C loss: 0.089547]\n",
            "[Epoch 36/50] [Batch 350/368] [D loss: 0.487158] [G loss: 0.368228] [C loss: 0.093097]\n",
            "[Epoch 37/50] [Batch 0/368] [D loss: 0.493533] [G loss: 0.373249] [C loss: 0.078212]\n",
            "[Epoch 37/50] [Batch 50/368] [D loss: 0.485828] [G loss: 0.371731] [C loss: 0.092789]\n",
            "[Epoch 37/50] [Batch 100/368] [D loss: 0.483170] [G loss: 0.374397] [C loss: 0.071623]\n",
            "[Epoch 37/50] [Batch 150/368] [D loss: 0.479197] [G loss: 0.373440] [C loss: 0.096893]\n",
            "[Epoch 37/50] [Batch 200/368] [D loss: 0.480938] [G loss: 0.375400] [C loss: 0.076409]\n",
            "[Epoch 37/50] [Batch 250/368] [D loss: 0.483920] [G loss: 0.383996] [C loss: 0.070363]\n",
            "[Epoch 37/50] [Batch 300/368] [D loss: 0.484486] [G loss: 0.377794] [C loss: 0.083261]\n",
            "[Epoch 37/50] [Batch 350/368] [D loss: 0.477129] [G loss: 0.380437] [C loss: 0.069600]\n",
            "[Epoch 38/50] [Batch 0/368] [D loss: 0.491898] [G loss: 0.374216] [C loss: 0.079197]\n",
            "[Epoch 38/50] [Batch 50/368] [D loss: 0.492737] [G loss: 0.369922] [C loss: 0.077490]\n",
            "[Epoch 38/50] [Batch 100/368] [D loss: 0.486477] [G loss: 0.383613] [C loss: 0.094339]\n",
            "[Epoch 38/50] [Batch 150/368] [D loss: 0.496783] [G loss: 0.366643] [C loss: 0.088733]\n",
            "[Epoch 38/50] [Batch 200/368] [D loss: 0.514300] [G loss: 0.374040] [C loss: 0.179238]\n",
            "[Epoch 38/50] [Batch 250/368] [D loss: 0.515455] [G loss: 0.371962] [C loss: 0.164069]\n",
            "[Epoch 38/50] [Batch 300/368] [D loss: 0.484585] [G loss: 0.387819] [C loss: 0.082010]\n",
            "[Epoch 38/50] [Batch 350/368] [D loss: 0.492129] [G loss: 0.364206] [C loss: 0.087980]\n",
            "[Epoch 39/50] [Batch 0/368] [D loss: 0.479796] [G loss: 0.379689] [C loss: 0.074586]\n",
            "[Epoch 39/50] [Batch 50/368] [D loss: 0.499573] [G loss: 0.375757] [C loss: 0.095566]\n",
            "[Epoch 39/50] [Batch 100/368] [D loss: 0.475375] [G loss: 0.377207] [C loss: 0.076706]\n",
            "[Epoch 39/50] [Batch 150/368] [D loss: 0.496461] [G loss: 0.364095] [C loss: 0.089434]\n",
            "[Epoch 39/50] [Batch 200/368] [D loss: 0.489684] [G loss: 0.375552] [C loss: 0.069312]\n",
            "[Epoch 39/50] [Batch 250/368] [D loss: 0.487400] [G loss: 0.360708] [C loss: 0.082820]\n",
            "[Epoch 39/50] [Batch 300/368] [D loss: 0.470590] [G loss: 0.375866] [C loss: 0.072441]\n",
            "[Epoch 39/50] [Batch 350/368] [D loss: 0.487776] [G loss: 0.380395] [C loss: 0.100618]\n",
            "[Epoch 40/50] [Batch 0/368] [D loss: 0.489942] [G loss: 0.363853] [C loss: 0.088244]\n",
            "[Epoch 40/50] [Batch 50/368] [D loss: 0.483025] [G loss: 0.370591] [C loss: 0.082168]\n",
            "[Epoch 40/50] [Batch 100/368] [D loss: 0.484960] [G loss: 0.386931] [C loss: 0.088880]\n",
            "[Epoch 40/50] [Batch 150/368] [D loss: 0.476072] [G loss: 0.372345] [C loss: 0.079512]\n",
            "[Epoch 40/50] [Batch 200/368] [D loss: 0.477729] [G loss: 0.374334] [C loss: 0.081435]\n",
            "[Epoch 40/50] [Batch 250/368] [D loss: 0.479962] [G loss: 0.368717] [C loss: 0.071063]\n",
            "[Epoch 40/50] [Batch 300/368] [D loss: 0.483244] [G loss: 0.372664] [C loss: 0.074299]\n",
            "[Epoch 40/50] [Batch 350/368] [D loss: 0.484136] [G loss: 0.376239] [C loss: 0.063258]\n",
            "[Epoch 41/50] [Batch 0/368] [D loss: 0.507469] [G loss: 0.367919] [C loss: 0.111565]\n",
            "[Epoch 41/50] [Batch 50/368] [D loss: 0.497613] [G loss: 0.378059] [C loss: 0.106738]\n",
            "[Epoch 41/50] [Batch 100/368] [D loss: 0.497008] [G loss: 0.372705] [C loss: 0.081478]\n",
            "[Epoch 41/50] [Batch 150/368] [D loss: 0.471544] [G loss: 0.374509] [C loss: 0.096052]\n",
            "[Epoch 41/50] [Batch 200/368] [D loss: 0.477081] [G loss: 0.361368] [C loss: 0.064978]\n",
            "[Epoch 41/50] [Batch 250/368] [D loss: 0.475533] [G loss: 0.378737] [C loss: 0.064976]\n",
            "[Epoch 41/50] [Batch 300/368] [D loss: 0.478050] [G loss: 0.373460] [C loss: 0.084187]\n",
            "[Epoch 41/50] [Batch 350/368] [D loss: 0.489406] [G loss: 0.371812] [C loss: 0.080051]\n",
            "[Epoch 42/50] [Batch 0/368] [D loss: 0.471266] [G loss: 0.360174] [C loss: 0.074165]\n",
            "[Epoch 42/50] [Batch 50/368] [D loss: 0.508078] [G loss: 0.360124] [C loss: 0.137790]\n",
            "[Epoch 42/50] [Batch 100/368] [D loss: 0.485230] [G loss: 0.376438] [C loss: 0.077259]\n",
            "[Epoch 42/50] [Batch 150/368] [D loss: 0.466171] [G loss: 0.373671] [C loss: 0.076837]\n",
            "[Epoch 42/50] [Batch 200/368] [D loss: 0.482967] [G loss: 0.378950] [C loss: 0.091837]\n",
            "[Epoch 42/50] [Batch 250/368] [D loss: 0.477239] [G loss: 0.370168] [C loss: 0.073777]\n",
            "[Epoch 42/50] [Batch 300/368] [D loss: 0.495954] [G loss: 0.366384] [C loss: 0.092037]\n",
            "[Epoch 42/50] [Batch 350/368] [D loss: 0.496362] [G loss: 0.385666] [C loss: 0.082857]\n",
            "[Epoch 43/50] [Batch 0/368] [D loss: 0.483382] [G loss: 0.372744] [C loss: 0.070747]\n",
            "[Epoch 43/50] [Batch 50/368] [D loss: 0.480732] [G loss: 0.379637] [C loss: 0.075216]\n",
            "[Epoch 43/50] [Batch 100/368] [D loss: 0.476048] [G loss: 0.370840] [C loss: 0.079161]\n",
            "[Epoch 43/50] [Batch 150/368] [D loss: 0.480533] [G loss: 0.377710] [C loss: 0.086057]\n",
            "[Epoch 43/50] [Batch 200/368] [D loss: 0.495918] [G loss: 0.364684] [C loss: 0.100284]\n",
            "[Epoch 43/50] [Batch 250/368] [D loss: 0.491191] [G loss: 0.367243] [C loss: 0.081986]\n",
            "[Epoch 43/50] [Batch 300/368] [D loss: 0.476861] [G loss: 0.372697] [C loss: 0.091786]\n",
            "[Epoch 43/50] [Batch 350/368] [D loss: 0.473801] [G loss: 0.378646] [C loss: 0.084125]\n",
            "[Epoch 44/50] [Batch 0/368] [D loss: 0.475496] [G loss: 0.372193] [C loss: 0.081093]\n",
            "[Epoch 44/50] [Batch 50/368] [D loss: 0.481052] [G loss: 0.378530] [C loss: 0.073025]\n",
            "[Epoch 44/50] [Batch 100/368] [D loss: 0.477749] [G loss: 0.367068] [C loss: 0.075175]\n",
            "[Epoch 44/50] [Batch 150/368] [D loss: 0.485451] [G loss: 0.354664] [C loss: 0.078344]\n",
            "[Epoch 44/50] [Batch 200/368] [D loss: 0.497787] [G loss: 0.371127] [C loss: 0.110917]\n",
            "[Epoch 44/50] [Batch 250/368] [D loss: 0.472508] [G loss: 0.380252] [C loss: 0.092182]\n",
            "[Epoch 44/50] [Batch 300/368] [D loss: 0.476820] [G loss: 0.388566] [C loss: 0.080184]\n",
            "[Epoch 44/50] [Batch 350/368] [D loss: 0.502303] [G loss: 0.362383] [C loss: 0.098336]\n",
            "[Epoch 45/50] [Batch 0/368] [D loss: 0.515364] [G loss: 0.372399] [C loss: 0.168138]\n",
            "[Epoch 45/50] [Batch 50/368] [D loss: 0.490057] [G loss: 0.376049] [C loss: 0.099155]\n",
            "[Epoch 45/50] [Batch 100/368] [D loss: 0.521186] [G loss: 0.364656] [C loss: 0.159650]\n",
            "[Epoch 45/50] [Batch 150/368] [D loss: 0.478472] [G loss: 0.373707] [C loss: 0.086705]\n",
            "[Epoch 45/50] [Batch 200/368] [D loss: 0.481776] [G loss: 0.383320] [C loss: 0.069655]\n",
            "[Epoch 45/50] [Batch 250/368] [D loss: 0.477813] [G loss: 0.368083] [C loss: 0.072552]\n",
            "[Epoch 45/50] [Batch 300/368] [D loss: 0.486607] [G loss: 0.375056] [C loss: 0.070791]\n",
            "[Epoch 45/50] [Batch 350/368] [D loss: 0.474777] [G loss: 0.383677] [C loss: 0.068140]\n",
            "[Epoch 46/50] [Batch 0/368] [D loss: 0.484389] [G loss: 0.380298] [C loss: 0.075005]\n",
            "[Epoch 46/50] [Batch 50/368] [D loss: 0.478213] [G loss: 0.373230] [C loss: 0.063617]\n",
            "[Epoch 46/50] [Batch 100/368] [D loss: 0.480752] [G loss: 0.382853] [C loss: 0.072544]\n",
            "[Epoch 46/50] [Batch 150/368] [D loss: 0.483802] [G loss: 0.359712] [C loss: 0.066934]\n",
            "[Epoch 46/50] [Batch 200/368] [D loss: 0.494917] [G loss: 0.372889] [C loss: 0.089468]\n",
            "[Epoch 46/50] [Batch 250/368] [D loss: 0.480874] [G loss: 0.376359] [C loss: 0.075890]\n",
            "[Epoch 46/50] [Batch 300/368] [D loss: 0.486723] [G loss: 0.365045] [C loss: 0.073555]\n",
            "[Epoch 46/50] [Batch 350/368] [D loss: 0.476357] [G loss: 0.378926] [C loss: 0.078298]\n",
            "[Epoch 47/50] [Batch 0/368] [D loss: 0.496317] [G loss: 0.369191] [C loss: 0.083567]\n",
            "[Epoch 47/50] [Batch 50/368] [D loss: 0.477014] [G loss: 0.380113] [C loss: 0.063616]\n",
            "[Epoch 47/50] [Batch 100/368] [D loss: 0.495091] [G loss: 0.379863] [C loss: 0.117830]\n",
            "[Epoch 47/50] [Batch 150/368] [D loss: 0.489871] [G loss: 0.360343] [C loss: 0.064559]\n",
            "[Epoch 47/50] [Batch 200/368] [D loss: 0.483842] [G loss: 0.371701] [C loss: 0.062430]\n",
            "[Epoch 47/50] [Batch 250/368] [D loss: 0.487690] [G loss: 0.367017] [C loss: 0.096362]\n",
            "[Epoch 47/50] [Batch 300/368] [D loss: 0.470335] [G loss: 0.354959] [C loss: 0.070987]\n",
            "[Epoch 47/50] [Batch 350/368] [D loss: 0.493489] [G loss: 0.374721] [C loss: 0.070652]\n",
            "[Epoch 48/50] [Batch 0/368] [D loss: 0.484122] [G loss: 0.387773] [C loss: 0.068021]\n",
            "[Epoch 48/50] [Batch 50/368] [D loss: 0.485890] [G loss: 0.375595] [C loss: 0.115226]\n",
            "[Epoch 48/50] [Batch 100/368] [D loss: 0.481973] [G loss: 0.366116] [C loss: 0.075787]\n",
            "[Epoch 48/50] [Batch 150/368] [D loss: 0.485128] [G loss: 0.363567] [C loss: 0.097658]\n",
            "[Epoch 48/50] [Batch 200/368] [D loss: 0.478573] [G loss: 0.384659] [C loss: 0.069466]\n",
            "[Epoch 48/50] [Batch 250/368] [D loss: 0.477203] [G loss: 0.377164] [C loss: 0.075793]\n",
            "[Epoch 48/50] [Batch 300/368] [D loss: 0.484641] [G loss: 0.380214] [C loss: 0.078204]\n",
            "[Epoch 48/50] [Batch 350/368] [D loss: 0.477770] [G loss: 0.363568] [C loss: 0.082105]\n",
            "[Epoch 49/50] [Batch 0/368] [D loss: 0.469986] [G loss: 0.379271] [C loss: 0.079232]\n",
            "[Epoch 49/50] [Batch 50/368] [D loss: 0.484914] [G loss: 0.358521] [C loss: 0.083277]\n",
            "[Epoch 49/50] [Batch 100/368] [D loss: 0.485293] [G loss: 0.384897] [C loss: 0.073227]\n",
            "[Epoch 49/50] [Batch 150/368] [D loss: 0.465273] [G loss: 0.369245] [C loss: 0.083882]\n",
            "[Epoch 49/50] [Batch 200/368] [D loss: 0.464801] [G loss: 0.368127] [C loss: 0.087515]\n",
            "[Epoch 49/50] [Batch 250/368] [D loss: 0.491266] [G loss: 0.379202] [C loss: 0.072274]\n",
            "[Epoch 49/50] [Batch 300/368] [D loss: 0.481185] [G loss: 0.365877] [C loss: 0.085002]\n",
            "[Epoch 49/50] [Batch 350/368] [D loss: 0.480201] [G loss: 0.373005] [C loss: 0.084604]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.figure(figsize=(10,5))\n",
        "plt.title(\"GAN Generator, Discriminator, Embedding Loss During Training\")\n",
        "plt.plot(G_losses,label=\"Generator\")\n",
        "plt.plot(D_losses,label=\"Discriminator\")\n",
        "plt.plot(C_losses, label=\"Embedding\")\n",
        "plt.xlabel(\"iterations\")\n",
        "plt.ylabel(\"Loss\")\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 350
        },
        "id": "12b8A9Rcmf0H",
        "outputId": "440dfd55-3456-45d8-c499-a3341754f110"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x360 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "plt.figure(figsize=(10,5))\n",
        "plt.title(\"MSE between missing and predicted node\")\n",
        "plt.plot(V_losses, label=\"Validation\")\n",
        "plt.xlabel(\"iterations\")\n",
        "plt.ylabel(\"Loss\")\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 350
        },
        "id": "F1gxeinDp8jo",
        "outputId": "129d67f7-f8d9-496c-b965-f5ba2c156d29"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 720x360 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## **Test**"
      ],
      "metadata": {
        "id": "9WBf3cjshuDF"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "generator.load_state_dict(torch.load(os.path.join(models_path, \"generator\")))\n",
        "classifier.load_state_dict(torch.load(os.path.join(models_path, \"classifier\")))"
      ],
      "metadata": {
        "id": "jJHrRTh_54DM",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "5b75b0aa-7334-4e61-d64b-43780f755d04"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<All keys matched successfully>"
            ]
          },
          "metadata": {},
          "execution_count": 118
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "missing_nodes_embeddings = []\n",
        "generator.eval()\n",
        "classifier.eval()\n",
        "\n",
        "for data in val_com_dataloader:\n",
        "  recipe_emb = data['recipe'].to(device)\n",
        "  noise = Variable(torch.FloatTensor(np.random.normal(0, 1, (recipe_emb.shape[0], args.latent_dim)))).to(device)\n",
        "  # noise = Variable(torch.FloatTensor(np.full((recipe_emb.shape[0], args.latent_dim), 0.001))).to(device)\n",
        "  generated = generator(torch.cat((noise, recipe_emb), dim=1))\n",
        "  generated = generated.type(torch.float).to(device)\n",
        "  partition = classifier(generated)\n",
        "  missing_nodes_embeddings.append(partition)\n",
        "\n",
        "missing_nodes_embeddings = torch.cat(missing_nodes_embeddings, dim=0)"
      ],
      "metadata": {
        "id": "xL1cnbcn2pjF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "missing_nodes_embeddings = missing_nodes_embeddings.cpu().detach().numpy()"
      ],
      "metadata": {
        "id": "kuJVYJyptOJR"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def val_completion_missing():\n",
        "  target = []\n",
        "  ingredients = []\n",
        "  with open (\"validation_completion_answer.csv\", \"r\") as f:\n",
        "    data = f.readlines()\n",
        "    for missing in data:\n",
        "      target.append(int(missing.strip()))\n",
        "  \n",
        "  with open (\"validation_completion_question.csv\", \"r\") as f:\n",
        "    data = f.readlines()\n",
        "    for recipe in data:\n",
        "      ing = recipe.strip().split(\",\")\n",
        "      ing = [int(x) for x in ing]\n",
        "      ingredients.append(ing)\n",
        "      \n",
        "  return target, ingredients\n",
        "\n",
        "val_missing_nodes, val_other_ingredients = val_completion_missing()"
      ],
      "metadata": {
        "id": "d0r-1q8Kxv2K"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def L2_loop_function(X_test, X_train):\n",
        "  test_data_shape = X_test.shape[0]\n",
        "  train_data_shape = X_train.shape[0]\n",
        "  distances = np.zeros((test_data_shape, train_data_shape))\n",
        "\n",
        "  for i in range(test_data_shape):\n",
        "      distances[i, :] = np.sum((X_train - X_test[i, :]) ** 2, axis = 1) #no need to sqrt as it causes the same effect\n",
        "  return distances\n",
        "\n",
        "def L2_vectorized_function(X_test, X_train):\n",
        "  X_test_squared = np.sum(np.square(X_test), axis=1, keepdims=True)\n",
        "  X_train_squared = np.sum(np.square(X_train), axis=1, keepdims=True)\n",
        "  X_test_X_train = np.dot(X_test, X_train.T)\n",
        "  return X_test_squared - (2 * X_test_X_train) + X_train_squared.T\n",
        "\n",
        "\n",
        "def predict_nodes(distance, other_ingredients, k):\n",
        "  nodes = []\n",
        "  for i in range(distance.shape[0]):\n",
        "    indices = np.argsort(distance[i, :])\n",
        "    temp = []\n",
        "    for idx in indices:\n",
        "      if idx not in other_ingredients[i]:\n",
        "        temp.append(idx)\n",
        "      if len(temp) == k:\n",
        "        break\n",
        "    nodes.append(temp)\n",
        "  return nodes\n",
        "\n",
        "def predict_missing_node(X_test, X_train, other_ingredients, k=1, typ=0):\n",
        "  if typ == 0:\n",
        "      method = L2_vectorized_function(X_test, X_train)\n",
        "  else:\n",
        "      method = L2_loop_function(X_test, X_train)\n",
        "  return predict_nodes(method, other_ingredients, k)"
      ],
      "metadata": {
        "id": "li81WktQmdIO"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "pred_missing_nodes = predict_missing_node(missing_nodes_embeddings, embeddings.to_numpy(), val_other_ingredients, k=20, typ=0)"
      ],
      "metadata": {
        "id": "U_WEhsZ2q33j"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "count = 0\n",
        "for i, j in zip(val_missing_nodes, pred_missing_nodes):\n",
        "  if i in j:\n",
        "    count += 1\n",
        "print(f\"Matched: {count}/{len(val_missing_nodes)} Validation accuracy: {(count/len(val_missing_nodes))*100:.4} %\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fxwCEaIzzFuh",
        "outputId": "461a03eb-00b9-4797-e540-57882b5a81e5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Matched: 14/7848 Validation accuracy: 0.1784 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "NqUFcRgtmUTZ"
      },
      "source": [
        "## **KNN**"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "output = predict_missing_node(x_val_com_dataset, embeddings.to_numpy(), val_other_ingredients, k=200, typ=0)"
      ],
      "metadata": {
        "id": "z6EEokFUSdJ8"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "count = 0\n",
        "for i, j in zip(val_missing_nodes, output):\n",
        "  if i == j:\n",
        "    count += 1\n",
        "print(f\"Matched: {count}/{len(val_missing_nodes)} Validation accuracy: {(count/len(val_missing_nodes))*100:.4} %\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nPXHhJhhVr3o",
        "outputId": "a4912b6a-06b5-407a-8d23-5b5b7bab29b4"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Matched: 0/7848 Validation accuracy: 0.0 %\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "output_ = predict_missing_node(x_test_com_dataset, scaled_node_features, val_other_ingredients, k=20, typ=0)"
      ],
      "metadata": {
        "id": "lgCxEZh_z4xG"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "pd.DataFrame(output_.tolist(), columns = [\"ingredient\"]).to_csv(os.path.join(results, \"test_completion_answer.csv\"))"
      ],
      "metadata": {
        "id": "WaFAWkL4LOyO"
      },
      "execution_count": null,
      "outputs": []
    }
  ],
  "metadata": {
    "colab": {
      "collapsed_sections": [],
      "machine_shape": "hm",
      "name": "term_project.ipynb",
      "provenance": [],
      "toc_visible": true,
      "authorship_tag": "ABX9TyO3h6/EFJGmOU49/gU28MhT",
      "include_colab_link": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "nbformat": 4,
  "nbformat_minor": 0
}