MNIST_Classification.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "MNIST_Classification.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "authorship_tag": "ABX9TyNjZGjFdFIfByuKgppXQPEt",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "widgets": {
      "application/vnd.jupyter.widget-state+json": {
        "381bcc66e2244834bce63c20278cf042": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "VBoxModel",
          "state": {
            "_view_name": "VBoxView",
            "_dom_classes": [],
            "_model_name": "VBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_5e3582c81b074b8581a2ec9476979092",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_a810f5ab524340118978c0395d75a390",
              "IPY_MODEL_b52c9116fee94bf990445481ca82a0b7"
            ]
          }
        },
        "5e3582c81b074b8581a2ec9476979092": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "a810f5ab524340118978c0395d75a390": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "LabelModel",
          "state": {
            "_view_name": "LabelView",
            "style": "IPY_MODEL_8b9578ca96284cb29da8ddd09c4f044a",
            "_dom_classes": [],
            "description": "",
            "_model_name": "LabelModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 0.01MB of 0.01MB uploaded (0.00MB deduped)\r",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_547b4e975ff04d64867df12544ef123c"
          }
        },
        "b52c9116fee94bf990445481ca82a0b7": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_16d3c0ad09e84733afc58ac26ea7e6be",
            "_dom_classes": [],
            "description": "",
            "_model_name": "FloatProgressModel",
            "bar_style": "",
            "max": 1,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 1,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_6525e2e8f63442c48130ea943e11cab8"
          }
        },
        "8b9578ca96284cb29da8ddd09c4f044a": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "547b4e975ff04d64867df12544ef123c": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "16d3c0ad09e84733afc58ac26ea7e6be": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "6525e2e8f63442c48130ea943e11cab8": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "1b00178b18b14324acbffdf3c407d3aa": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "VBoxModel",
          "state": {
            "_view_name": "VBoxView",
            "_dom_classes": [],
            "_model_name": "VBoxModel",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "box_style": "",
            "layout": "IPY_MODEL_68408edec4484fed8fabeaa1054b4046",
            "_model_module": "@jupyter-widgets/controls",
            "children": [
              "IPY_MODEL_2094d189aa8c42a8bb58f0760561d770",
              "IPY_MODEL_4ca6198b2601407c859a569c5f63beb9"
            ]
          }
        },
        "68408edec4484fed8fabeaa1054b4046": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "2094d189aa8c42a8bb58f0760561d770": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "LabelModel",
          "state": {
            "_view_name": "LabelView",
            "style": "IPY_MODEL_0b4d662530194c41b8272cba9abbb976",
            "_dom_classes": [],
            "description": "",
            "_model_name": "LabelModel",
            "placeholder": "​",
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": " 0.01MB of 0.01MB uploaded (0.00MB deduped)\r",
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_f616c84340b14ca88c305438b41aaa35"
          }
        },
        "4ca6198b2601407c859a569c5f63beb9": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "FloatProgressModel",
          "state": {
            "_view_name": "ProgressView",
            "style": "IPY_MODEL_71ccbea484b94887bc18c54e3196381e",
            "_dom_classes": [],
            "description": "",
            "_model_name": "FloatProgressModel",
            "bar_style": "",
            "max": 1,
            "_view_module": "@jupyter-widgets/controls",
            "_model_module_version": "1.5.0",
            "value": 1,
            "_view_count": null,
            "_view_module_version": "1.5.0",
            "orientation": "horizontal",
            "min": 0,
            "description_tooltip": null,
            "_model_module": "@jupyter-widgets/controls",
            "layout": "IPY_MODEL_e29f2ba3bb5a464ea7880da39a53f4ce"
          }
        },
        "0b4d662530194c41b8272cba9abbb976": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "DescriptionStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "DescriptionStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "f616c84340b14ca88c305438b41aaa35": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        },
        "71ccbea484b94887bc18c54e3196381e": {
          "model_module": "@jupyter-widgets/controls",
          "model_name": "ProgressStyleModel",
          "state": {
            "_view_name": "StyleView",
            "_model_name": "ProgressStyleModel",
            "description_width": "",
            "_view_module": "@jupyter-widgets/base",
            "_model_module_version": "1.5.0",
            "_view_count": null,
            "_view_module_version": "1.2.0",
            "bar_color": null,
            "_model_module": "@jupyter-widgets/controls"
          }
        },
        "e29f2ba3bb5a464ea7880da39a53f4ce": {
          "model_module": "@jupyter-widgets/base",
          "model_name": "LayoutModel",
          "state": {
            "_view_name": "LayoutView",
            "grid_template_rows": null,
            "right": null,
            "justify_content": null,
            "_view_module": "@jupyter-widgets/base",
            "overflow": null,
            "_model_module_version": "1.2.0",
            "_view_count": null,
            "flex_flow": null,
            "width": null,
            "min_width": null,
            "border": null,
            "align_items": null,
            "bottom": null,
            "_model_module": "@jupyter-widgets/base",
            "top": null,
            "grid_column": null,
            "overflow_y": null,
            "overflow_x": null,
            "grid_auto_flow": null,
            "grid_area": null,
            "grid_template_columns": null,
            "flex": null,
            "_model_name": "LayoutModel",
            "justify_items": null,
            "grid_row": null,
            "max_height": null,
            "align_content": null,
            "visibility": null,
            "align_self": null,
            "height": null,
            "min_height": null,
            "padding": null,
            "grid_auto_rows": null,
            "grid_gap": null,
            "max_width": null,
            "order": null,
            "_view_module_version": "1.2.0",
            "grid_template_areas": null,
            "object_position": null,
            "object_fit": null,
            "grid_auto_columns": null,
            "margin": null,
            "display": null,
            "left": null
          }
        }
      }
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/paddy3696/NN_from_scratch/blob/main/MNIST_Classification.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "aYsxVcU3jimN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "2cfdedb2-6d05-4500-d4f6-e95bee7bd1d7"
      },
      "source": [
        "%pip install wandb -q"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[K     |████████████████████████████████| 2.0MB 8.7MB/s \n",
            "\u001b[K     |████████████████████████████████| 163kB 43.7MB/s \n",
            "\u001b[K     |████████████████████████████████| 133kB 38.3MB/s \n",
            "\u001b[K     |████████████████████████████████| 102kB 10.0MB/s \n",
            "\u001b[K     |████████████████████████████████| 71kB 7.7MB/s \n",
            "\u001b[?25h  Building wheel for pathtools (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Building wheel for subprocess32 (setup.py) ... \u001b[?25l\u001b[?25hdone\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jqPJLUzrO2uD"
      },
      "source": [
        "import numpy as np\n",
        "import pandas as pd\n",
        "import matplotlib.pyplot as plt\n",
        "from keras.datasets import fashion_mnist\n",
        "from sklearn.metrics import classification_report, accuracy_score, confusion_matrix\n",
        "from sklearn.model_selection import train_test_split\n",
        "\n",
        "# Ignore excessive warnings\n",
        "import logging\n",
        "logging.propagate = False \n",
        "logging.getLogger().setLevel(logging.ERROR)\n",
        "\n",
        "# WandB – Import the wandb library\n",
        "import wandb\n",
        "#wandb.init(project='mnist_classification', entity='paddy3696')"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "45EVqmwTiYtO"
      },
      "source": [
        "def one_hot(y, n_labels):\n",
        "    mat = np.zeros((len(y), n_labels))\n",
        "    for i, val in enumerate(y):\n",
        "        mat[i, val] = 1\n",
        "    return mat.T"
      ],
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "LZf4LhTynL-M"
      },
      "source": [
        "class MyNN(object):\n",
        "\n",
        "    def __init__(self,network_size,network_fns,wb_init='he',learning_rate=1e-3,\n",
        "                 optimizer = 'NADAM',keep_proba=1,loss_fn = 'crossE',regularize='l2',batch_size=32,\n",
        "                 max_epoch=30,alpha=0,beta_1=0.9,beta_2=0.999,seed=None,verbose=0):\n",
        "        \n",
        "        self.network_size = network_size\n",
        "        self.network_fns = network_fns\n",
        "        self.wb_init = wb_init\n",
        "        self.loss_fn = loss_fn\n",
        "        self.learning_rate = learning_rate\n",
        "        self.alpha = alpha\n",
        "        self.beta_1 = beta_1\n",
        "        self.beta_2 = beta_2\n",
        "        self.optimizer = optimizer\n",
        "        self.batch_size = batch_size\n",
        "        self.keep_proba = keep_proba\n",
        "        self.regularize = regularize\n",
        "        self.seed = seed\n",
        "        self.max_epoch = max_epoch\n",
        "        self.verbose = verbose\n",
        "\n",
        "    @staticmethod\n",
        "    def weights_init(network_size,wb_init='he',seed=None):\n",
        "        \n",
        "        np.random.seed(seed)\n",
        "        values = {}\n",
        "        opt_values = {}\n",
        "        L = len(network_size)          \n",
        "        if  wb_init == 'he':\n",
        "            for l in range(1, L):\n",
        "                values['W' + str(l)] = np.random.normal(0,np.sqrt(2.0/network_size[l-1]),(network_size[l], network_size[l-1]))\n",
        "                values['b' + str(l)] = np.random.normal(0,np.sqrt(2.0/network_size[l-1]),(network_size[l], 1))  \n",
        "               \n",
        "        elif wb_init == 'xavier_uniform':\n",
        "            for l in range(1, L):\n",
        "                values['W' + str(l)] = np.random.uniform(-(np.sqrt(6.0/(network_size[l]+network_size[l-1]))),\n",
        "                                                        (np.sqrt(6.0/(network_size[l]+network_size[l-1]))),\n",
        "                                                        (network_size[l], network_size[l-1]))\n",
        "                values['b' + str(l)] = np.random.uniform(-(np.sqrt(6.0/(network_size[l]+network_size[l-1]))),\n",
        "                                                        (np.sqrt(6.0/(network_size[l]+network_size[l-1]))),\n",
        "                                                        (network_size[l], 1))\n",
        "\n",
        "        return values\n",
        "    \n",
        "    @staticmethod\n",
        "    def ReLU(X,alpha=0,derivative=False):\n",
        "        X = np.array(X,dtype=np.float64)\n",
        "        if derivative == False:\n",
        "            return np.where(X<0,alpha*X,X)\n",
        "        elif derivative == True:\n",
        "            X_relu = np.ones_like(X,dtype=np.float64)\n",
        "            X_relu[X < 0] = alpha\n",
        "            return X_relu\n",
        "\n",
        "    @staticmethod\n",
        "    def sigmoid(X,derivative=False):\n",
        "        if derivative == False:\n",
        "            out = 1 / (1 + np.exp(-np.array(X)))\n",
        "        elif derivative == True:\n",
        "            s = 1 / (1 + np.exp(-np.array(X)))\n",
        "            out = s*(1-s)\n",
        "        return out\n",
        "    \n",
        "    @staticmethod\n",
        "    def Tanh(X,derivative=False):\n",
        "        X = np.array(X)\n",
        "        if derivative == False:\n",
        "            return np.tanh(X)\n",
        "        if derivative == True:\n",
        "            return 1 - (np.tanh(X))**2\n",
        "            \n",
        "    @staticmethod\n",
        "    def softmax(X):\n",
        "        return np.exp(X) / np.sum(np.exp(X),axis=0)\n",
        "    \n",
        "    @staticmethod\n",
        "    def forward_prop(X, network_fns,values,keep_prob=1,seed=None):\n",
        "    \n",
        "        if seed != None:\n",
        "            np.random.seed(seed)\n",
        "        caches = []\n",
        "        A = X\n",
        "        L = len(network_fns)\n",
        "        for l,act_fn in enumerate(network_fns,start=1):\n",
        "            A_prev = A \n",
        "        \n",
        "            Z = np.dot(values['W' + str(l)],A_prev)+values['b' + str(l)]\n",
        "            \n",
        "            if type(act_fn) is tuple:\n",
        "                \n",
        "                if  act_fn[0] == \"relu\":\n",
        "                    A = MyNN.ReLU(Z,act_fn[1])\n",
        "            else:\n",
        "                if act_fn == \"sigmoid\":\n",
        "                    A = MyNN.sigmoid(Z)\n",
        "                elif act_fn == \"tanh\":\n",
        "                    A = MyNN.Tanh(Z)\n",
        "                elif act_fn == \"softmax\":\n",
        "                    A = MyNN.softmax(Z)\n",
        "                elif  act_fn == \"relu\":\n",
        "                    A = MyNN.ReLU(Z)\n",
        "            \n",
        "            if keep_prob != 1 and l != L and l != 1:\n",
        "                D = np.random.rand(A.shape[0],A.shape[1])\n",
        "                D = (D<keep_prob)\n",
        "                A = np.multiply(A,D)\n",
        "                A = A / keep_prob\n",
        "                cache = ((A_prev, values['W' + str(l)],values['b' + str(l)],D), Z)\n",
        "                caches.append(cache)\n",
        "            else:\n",
        "                cache = ((A_prev, values['W' + str(l)],values['b' + str(l)]), Z)\n",
        "                #print(A.shape)\n",
        "                caches.append(cache)      \n",
        "        return A, caches\n",
        "\n",
        "    @staticmethod\n",
        "    def compute_loss(A, Y, values, alpha=0,regularize=None,loss_fn='crossE'):\n",
        "\n",
        "        m = Y.shape[1]\n",
        "    \n",
        "        if loss_fn == 'crossE':\n",
        "            loss = np.squeeze(-np.sum(np.multiply(np.log(A),Y))/m)    \n",
        "        elif loss_fn == 'mse':\n",
        "            loss = np.squeeze(np.sum((Y - A)**2)/m)       \n",
        "    \n",
        "        L = len(values)//2\n",
        "    \n",
        "        if regularize == 'l2' and alpha != 0:\n",
        "            sum_weights = 0\n",
        "            for l in range(1, L):\n",
        "                sum_weights = sum_weights + np.sum(np.square(values['W' + str(l)]))\n",
        "            loss = loss + sum_weights * (alpha/(2*m))\n",
        "        elif regularize == 'l1' and alpha != 0:\n",
        "            sum_weights = 0\n",
        "            for l in range(1, L):\n",
        "                sum_weights = sum_weights + np.sum(np.abs(values['W' + str(l)]))\n",
        "            loss = loss + sum_weights * (alpha/(2*m))\n",
        "        return loss\n",
        "\n",
        "    @staticmethod\n",
        "    def back_prop(AL, Y, caches, network_fns, keep_prob=1, regularize=None,alpha=0,loss_fn='crossE'):\n",
        "\n",
        "        grads = {}\n",
        "        L = len(caches)\n",
        "    \n",
        "        m = AL.shape[1]\n",
        "        Y = Y.reshape(AL.shape)\n",
        "    \n",
        "        # Initializing the backprop\n",
        "        if loss_fn == 'crossE':\n",
        "            dZL = AL - Y\n",
        "        elif loss_fn == 'mse':\n",
        "            dZL = 2 * (AL - Y)\n",
        "    \n",
        "        cache = caches[L-1]\n",
        "        linear_cache, activation_cache = cache\n",
        "        AL, W, b = linear_cache\n",
        "        grads[\"dW\" + str(L)] = np.dot(dZL,AL.T)/m\n",
        "        grads[\"db\" + str(L)] = np.sum(dZL,axis=1,keepdims=True)/m\n",
        "        grads[\"dA\" + str(L-1)] = np.dot(W.T,dZL)\n",
        "    \n",
        "            \n",
        "        v_dropout = 0\n",
        "        for l in reversed(range(L-1)):\n",
        "            cache = caches[l]\n",
        "            act_fn = network_fns[l]\n",
        "        \n",
        "            linear_cache, Z = cache\n",
        "            try:\n",
        "                A_prev, W, b = linear_cache\n",
        "            except:\n",
        "                A_prev, W, b, D = linear_cache\n",
        "                v_dropout = 1\n",
        "            \n",
        "            m = A_prev.shape[1]\n",
        "        \n",
        "            if keep_prob != 1 and v_dropout == 1:\n",
        "                dA_prev = np.multiply(grads[\"dA\" + str(l + 1)],D)\n",
        "                dA_prev = dA_prev/keep_prob\n",
        "                v_dropout = 0\n",
        "            else:\n",
        "                dA_prev = grads[\"dA\" + str(l + 1)]\n",
        "                v_dropout = 0\n",
        "            \n",
        "            \n",
        "            if type(act_fn) is tuple:\n",
        "                \n",
        "                if  act_fn[0] == \"relu\":\n",
        "                    dZ = np.multiply(dA_prev,MyNN.ReLU(Z,act_fn[1],derivative=True))\n",
        "\n",
        "            else:\n",
        "                if act_fn == \"sigmoid\":\n",
        "                    dZ = np.multiply(dA_prev,MyNN.sigmoid(Z,derivative=True))\n",
        "                elif act_fn == \"relu\":\n",
        "                    dZ = np.multiply(dA_prev,MyNN.ReLU(Z,derivative=True))\n",
        "                elif act_fn == \"tanh\":\n",
        "                    dZ = np.multiply(dA_prev,MyNN.Tanh(Z,derivative=True))\n",
        "            \n",
        "            grads[\"dA\" + str(l)] = np.dot(W.T,dZ)\n",
        "        \n",
        "            if regularize == 'l2':\n",
        "                grads[\"dW\" + str(l + 1)] = (np.dot(dZ,A_prev.T)/m)  + ((alpha * W)/m)\n",
        "            elif regularize == 'l1':\n",
        "                grads[\"dW\" + str(l + 1)] = (np.dot(dZ,A_prev.T)/m)  + ((alpha * np.sign(W+10**-8))/m)\n",
        "            else:\n",
        "                grads[\"dW\" + str(l + 1)] = (np.dot(dZ,A_prev.T)/m)\n",
        "            \n",
        "            grads[\"db\" + str(l + 1)] = np.sum(dZ,axis=1,keepdims=True)/m   \n",
        "        return grads\n",
        "    \n",
        "    @staticmethod\n",
        "    def update_values(values, grads,learning_rate,iter_no,method = 'SGD',opt_values=None,beta_1=0.9,beta_2=0.999):\n",
        "    \n",
        "        L = len(values) // 2 \n",
        "        if method == 'SGD':\n",
        "            for l in range(L):\n",
        "                values[\"W\" + str(l+1)] = values[\"W\" + str(l+1)] - learning_rate*grads[\"dW\" + str(l + 1)]\n",
        "                values[\"b\" + str(l+1)] = values[\"b\" + str(l+1)] - learning_rate*grads[\"db\" + str(l + 1)]\n",
        "            opt_values = None\n",
        "        elif method == 'SGDM':\n",
        "            for l in range(L):\n",
        "                opt_values['vdb'+str(l+1)] = beta_1*opt_values['vdb'+str(l+1)] + (1-beta_1)*grads[\"db\" + str(l + 1)]\n",
        "                opt_values['vdw'+str(l+1)] = beta_1*opt_values['vdw'+str(l+1)] + (1-beta_1)*grads[\"dW\" + str(l + 1)]\n",
        "                values[\"W\" + str(l+1)] = values[\"W\" + str(l+1)] - learning_rate*opt_values['vdw'+str(l+1)]\n",
        "                values[\"b\" + str(l+1)] = values[\"b\" + str(l+1)] - learning_rate*opt_values['vdb'+str(l+1)]\n",
        "        elif method == 'RMSP':\n",
        "            for l in range(L):\n",
        "                opt_values['sdb'+str(l+1)] = beta_2*opt_values['sdb'+str(l+1)] + \\\n",
        "                                                     (1-beta_2)*np.square(grads[\"db\" + str(l + 1)])\n",
        "                opt_values['sdw'+str(l+1)] = beta_2*opt_values['sdw'+str(l+1)] + \\\n",
        "                                                           (1-beta_2)*np.square(grads[\"dW\" + str(l + 1)])\n",
        "                values[\"W\" + str(l+1)] = values[\"W\" + str(l+1)] - \\\n",
        "                                       learning_rate*(grads[\"dW\" + str(l + 1)]/(np.sqrt(opt_values['sdw'+str(l+1)])+10**-8))\n",
        "                values[\"b\" + str(l+1)] = values[\"b\" + str(l+1)] - \\\n",
        "                                       learning_rate*(grads[\"db\" + str(l + 1)]/(np.sqrt(opt_values['sdb'+str(l+1)])+10**-8))\n",
        "        elif method == 'ADAM':\n",
        "            for l in range(L):\n",
        "                opt_values['vdb'+str(l+1)] = beta_1*opt_values['vdb'+str(l+1)] + (1-beta_1)*grads[\"db\" + str(l + 1)]\n",
        "                opt_values['vdw'+str(l+1)] = beta_1*opt_values['vdw'+str(l+1)] + (1-beta_1)*grads[\"dW\" + str(l + 1)]\n",
        "                opt_values['sdb'+str(l+1)] = beta_2*opt_values['sdb'+str(l+1)] + \\\n",
        "                                                                  (1-beta_2)*np.square(grads[\"db\" + str(l + 1)])\n",
        "                opt_values['sdw'+str(l+1)] = beta_2*opt_values['sdw'+str(l+1)] + \\\n",
        "                                                                   (1-beta_2)*np.square(grads[\"dW\" + str(l + 1)])\n",
        "            \n",
        "                learning_rate = learning_rate * np.sqrt((1-beta_2**iter_no)/((1-beta_1**iter_no)+10**-8))\n",
        "                values[\"W\" + str(l+1)] = values[\"W\" + str(l+1)] - \\\n",
        "                                       learning_rate*(opt_values['vdw'+str(l+1)]/\\\n",
        "                                                      (np.sqrt(opt_values['sdw'+str(l+1)])+10**-8))\n",
        "                values[\"b\" + str(l+1)] = values[\"b\" + str(l+1)] - \\\n",
        "                                       learning_rate*(opt_values['vdb'+str(l+1)]/\\\n",
        "                                                      (np.sqrt(opt_values['sdb'+str(l+1)])+10**-8))\n",
        "        \n",
        "        elif method == 'NADAM':\n",
        "            for l in range(L):\n",
        "                opt_values['vdb'+str(l+1)] = beta_1*opt_values['vdb'+str(l+1)] + (1-beta_1)*grads[\"db\" + str(l + 1)]\n",
        "                opt_values['vdw'+str(l+1)] = beta_1*opt_values['vdw'+str(l+1)] + (1-beta_1)*grads[\"dW\" + str(l + 1)]\n",
        "                opt_values['sdb'+str(l+1)] = beta_2*opt_values['sdb'+str(l+1)] + \\\n",
        "                                                                  (1-beta_2)*np.square(grads[\"db\" + str(l + 1)])\n",
        "                opt_values['sdw'+str(l+1)] = beta_2*opt_values['sdw'+str(l+1)] + \\\n",
        "                                                                   (1-beta_2)*np.square(grads[\"dW\" + str(l + 1)])\n",
        "            \n",
        "                \n",
        "                learning_rate = learning_rate * np.sqrt((1-beta_2**iter_no)/((1-beta_1**iter_no)+10**-8))\n",
        "                values[\"W\" + str(l+1)] = values[\"W\" + str(l+1)] - \\\n",
        "                                       learning_rate*((beta_1*opt_values['vdw'+str(l+1)])/\\\n",
        "                                                      (np.sqrt(opt_values['sdw'+str(l+1)])+10**-8))\n",
        "                values[\"b\" + str(l+1)] = values[\"b\" + str(l+1)] - \\\n",
        "                                       learning_rate*((beta_1*opt_values['vdb'+str(l+1)]+ (1-beta_1)*grads[\"db\" + str(l + 1)])/\\\n",
        "                                                      (np.sqrt(opt_values['sdb'+str(l+1)])+10**-8))\n",
        "        \n",
        "        return values,opt_values\n",
        "    \n",
        "    def fit(self,X,y,x_valid,y_valid):\n",
        "\n",
        "        x_train = X\n",
        "        y_train = y\n",
        "        X_valid = X\n",
        "        Y_valid = y\n",
        "        y = one_hot(y, 10)\n",
        "        np.random.seed(self.seed)\n",
        "        self.grads = {}\n",
        "        self.losss = []\n",
        "        M = X.shape[1]\n",
        "        opt_values = {}\n",
        "        \n",
        "        if self.verbose == 1:\n",
        "            print('Initilizing Weights...')\n",
        "        self.values = self.weights_init(self.network_size,self.wb_init,self.seed)\n",
        "        self.iter_no = 0\n",
        "        idx = np.arange(0,M)\n",
        "        \n",
        "        if self.optimizer != 'SGD':\n",
        "            for l in range(1, len(self.network_size)):\n",
        "                opt_values['vdw' + str(l)] = np.zeros((self.network_size[l], self.network_size[l-1]))\n",
        "                opt_values['vdb' + str(l)] = np.zeros((self.network_size[l], 1))\n",
        "                opt_values['sdw' + str(l)] = np.zeros((self.network_size[l], self.network_size[l-1]))\n",
        "                opt_values['sdb' + str(l)] = np.zeros((self.network_size[l], 1)) \n",
        "        \n",
        "        if self.verbose == 1:\n",
        "            print('Starting Training...')\n",
        "        acc_epoch = []\n",
        "        \n",
        "        for epoch_no in range(1,self.max_epoch+1):\n",
        "            \n",
        "            np.random.shuffle(idx)\n",
        "            X = X[:,idx]\n",
        "            y = y[:,idx]\n",
        "            for i in range(0,M, self.batch_size):\n",
        "                self.iter_no = self.iter_no + 1\n",
        "                X_batch = X[:,i:i + self.batch_size]\n",
        "                y_batch = y[:,i:i + self.batch_size]\n",
        "                # Forward prop:\n",
        "                AL, cache = self.forward_prop(X_batch,self.network_fns,self.values,self.keep_proba,self.seed)\n",
        "                #loss\n",
        "                loss = self.compute_loss(AL, y_batch, self.values,self.alpha,self.regularize,self.loss_fn)\n",
        "                self.losss.append(loss)\n",
        "                \n",
        "                \n",
        "                #back prop\n",
        "                grads = self.back_prop(AL, y_batch, cache,self.network_fns,self.keep_proba,self.regularize,self.alpha,self.loss_fn)\n",
        "                \n",
        "                #update params\n",
        "                self.values,opt_values = self.update_values(self.values,grads,self.learning_rate,\n",
        "                                                                        self.iter_no-1,self.optimizer,\n",
        "                                                                        opt_values,self.beta_1,self.beta_2)\n",
        "                                \n",
        "                if self.verbose == 1:\n",
        "                    if self.iter_no % 1000 == 0:\n",
        "                        \n",
        "                        print(\"loss after iteration {}: {}\".format(self.iter_no, loss))\n",
        "                        #wandb.log({'Training Error' : loss}\n",
        "                \n",
        "                #wandb.log({'Training Loss' : loss})\n",
        "                \n",
        "\n",
        "            y_pred_train = self.predict(x_train)\n",
        "            y_pred_valid = self.predict(x_valid)\n",
        "\n",
        "            train_acc = accuracy_score(y_train, y_pred_train)\n",
        "            valid_acc = accuracy_score(y_valid, y_pred_valid)\n",
        "            train_loss = np.mean(loss)\n",
        "\n",
        "            if self.verbose == 1:\n",
        "              print('Training Accuracy: %.2f%%' % (train_acc * 100))\n",
        "              print('Validation Accuracy: %.2f%%' % (valid_acc * 100))\n",
        "              print('Train loss per epoch: %.2f' % (train_loss))\n",
        "            \n",
        "            wandb.log({'Training_accuracy' : train_acc})\n",
        "            wandb.log({'Validation_accuracy' : valid_acc})\n",
        "            wandb.log({'Training_Loss' : loss})\n",
        "            acc_epoch.append(valid_acc)\n",
        "\n",
        "        acc_best = max(acc_epoch)\n",
        "        if self.verbose == 1:\n",
        "          print('Best Validation Accuracy: %.2f%%' % (acc_best * 100))\n",
        "          \n",
        "        wandb.log({'accuracy' : acc_best})\n",
        "        return self\n",
        "    def predict(self,X,proba=False):\n",
        "        \n",
        "        out, _ = self.forward_prop(X,self.network_fns,self.values,self.keep_proba,self.seed)\n",
        "        if proba == True:\n",
        "            return out.T\n",
        "        else:\n",
        "            return np.argmax(out, axis=0)"
      ],
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 417
        },
        "id": "x07iER-4Y5os",
        "outputId": "e128a0b9-4cd9-4d2d-b07d-c58bffdb29f6"
      },
      "source": [
        "(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data()\n",
        "\n",
        "class_num = 10\n",
        "num_row = 2\n",
        "num_col = 5# plot images\n",
        "fig, axes = plt.subplots(num_row, num_col, figsize=(1.5*num_col,2*num_row))\n",
        "for i in range(class_num):\n",
        "  ax = axes[i//num_col, i%num_col]\n",
        "  a = np.argmax(y_train == i)\n",
        "  ax.imshow(x_train[a], cmap='gray')\n",
        "  ax.set_title('Label: {}'.format(y_train[a]))\n",
        "plt.tight_layout()\n",
        "\n",
        "x_train, x_valid, y_train, y_valid = train_test_split(x_train, y_train, test_size=0.1, random_state=0, stratify=y_train)\n",
        "\n",
        "x_train = x_train.reshape((len(x_train), 28*28))\n",
        "x_train = x_train.astype('float32') / 255\n",
        "\n",
        "x_valid = x_valid.reshape((len(x_valid), 28*28))\n",
        "x_valid = x_valid.astype('float32') / 255\n",
        "\n",
        "# Preprocessing test data\n",
        "x_test = x_test.reshape((len(x_test), 28 * 28))\n",
        "x_test = x_test.astype('float32') / 225"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz\n",
            "32768/29515 [=================================] - 0s 0us/step\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz\n",
            "26427392/26421880 [==============================] - 0s 0us/step\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz\n",
            "8192/5148 [===============================================] - 0s 0us/step\n",
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz\n",
            "4423680/4422102 [==============================] - 0s 0us/step\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 540x288 with 10 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SpXPExPSGG0X"
      },
      "source": [
        "# Configure the sweep – specify the parameters to search through, the search strategy, the optimization metric et all.\n",
        "sweep_config = {\n",
        "    'method': 'bayes', #grid, random\n",
        "    'metric': {\n",
        "      'name': 'accuracy',\n",
        "      'goal': 'maximize'   \n",
        "    },\n",
        "    'parameters': {\n",
        "        'max_epoch': {\n",
        "            'values': [20, 30]\n",
        "        },\n",
        "        'wb_init': {\n",
        "            'values': ['he', 'xavier_uniform']\n",
        "        },\n",
        "        'batch_size': {\n",
        "            'values': [16, 32, 64]\n",
        "        },\n",
        "        'hidden_size': {\n",
        "            'values': [32, 64, 128]\n",
        "        },\n",
        "        'n_hidden': {\n",
        "            'values': [3,4,5]\n",
        "        },\n",
        "        'alpha': {\n",
        "            'values': [0, 0.0005, 0.5]\n",
        "        },\n",
        "        'learning_rate': {\n",
        "            'values': [1e-3, 1e-4]\n",
        "        },\n",
        "        'optimizer': {\n",
        "            'values': ['SGD','SGDM','RMSP','ADAM','NADAM'] \n",
        "        },\n",
        "        'activation': {\n",
        "            'values': ['relu','sigmoid','tanh']\n",
        "        },\n",
        "    }\n",
        "}"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 86
        },
        "id": "_5M_1mNXKsD5",
        "outputId": "93580b9a-1e4b-4899-ac12-cf1ac4e3fd57"
      },
      "source": [
        "sweep_id = wandb.sweep(sweep_config, entity=\"paddy3696\", project=\"mnist_classification\")"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "\n",
              "        window._wandbApiKey = new Promise((resolve, reject) => {\n",
              "            function loadScript(url) {\n",
              "            return new Promise(function(resolve, reject) {\n",
              "                let newScript = document.createElement(\"script\");\n",
              "                newScript.onerror = reject;\n",
              "                newScript.onload = resolve;\n",
              "                document.body.appendChild(newScript);\n",
              "                newScript.src = url;\n",
              "            });\n",
              "            }\n",
              "            loadScript(\"https://cdn.jsdelivr.net/npm/postmate/build/postmate.min.js\").then(() => {\n",
              "            const iframe = document.createElement('iframe')\n",
              "            iframe.style.cssText = \"width:0;height:0;border:none\"\n",
              "            document.body.appendChild(iframe)\n",
              "            const handshake = new Postmate({\n",
              "                container: iframe,\n",
              "                url: 'https://wandb.ai/authorize'\n",
              "            });\n",
              "            const timeout = setTimeout(() => reject(\"Couldn't auto authenticate\"), 5000)\n",
              "            handshake.then(function(child) {\n",
              "                child.on('authorize', data => {\n",
              "                    clearTimeout(timeout)\n",
              "                    resolve(data)\n",
              "                });\n",
              "            });\n",
              "            })\n",
              "        });\n",
              "    "
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\u001b[34m\u001b[1mwandb\u001b[0m: Appending key for api.wandb.ai to your netrc file: /root/.netrc\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Create sweep with ID: 2fbqxb69\n",
            "Sweep URL: https://wandb.ai/paddy3696/mnist_classification/sweeps/2fbqxb69\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RqE-vcfnH-tU"
      },
      "source": [
        "def train():\n",
        "  # Default values for hyper-parameters we're going to sweep over\n",
        "  config_defaults = {\n",
        "      'max_epoch': 20,\n",
        "      'n_hidden': 4,\n",
        "      'hidden_size': 64,\n",
        "      'batch_size': 64,\n",
        "      'alpha': 0.0005,\n",
        "      'loss_fn': 'crossE',\n",
        "      'wb_init': 'he',\n",
        "      'learning_rate': 1e-3,\n",
        "      'activation': 'tanh',\n",
        "      'optimizer': 'NADAM',\n",
        "      'seed': 9,\n",
        "      'verbose':0,\n",
        "      'regularize':'l2'\n",
        "  }\n",
        "\n",
        "  # Initialize a new wandb run\n",
        "  wandb.init(config=config_defaults)\n",
        "\n",
        "  \n",
        "    \n",
        "  # Config is a variable that holds and saves hyperparameters and inputs\n",
        "  config = wandb.config\n",
        "\n",
        "  X = x_train.T\n",
        "  X_valid = x_valid.T\n",
        "  Y_valid = y_valid\n",
        "  Y = y_train\n",
        "  #X_test = x_test.T\n",
        "  n_class= 10\n",
        "\n",
        "  if config.activation == 'relu':\n",
        "    config.wb_init = 'he'\n",
        "  else:\n",
        "    config.wb_init = 'xavier_uniform'\n",
        "\n",
        "  layers = []\n",
        "  for i in range(config.n_hidden+2):\n",
        "    if i == 0:\n",
        "        layers.append(X.shape[0])\n",
        "    elif i == config.n_hidden+1:\n",
        "        layers.append(n_class)\n",
        "    else:\n",
        "        layers.append(config.hidden_size)\n",
        "    i = i+1\n",
        "\n",
        "  #print(layers)\n",
        "\n",
        "  o_act = 'softmax'\n",
        "  if config.activation == 'relu':\n",
        "    actvn_fn = ('relu',0)\n",
        "  else:\n",
        "    actvn_fn = config.activation\n",
        "\n",
        "  act = []\n",
        "  for i in range(config.n_hidden+1):\n",
        "    if i == config.n_hidden:\n",
        "      act.append(o_act)\n",
        "    else:\n",
        "      act.append(actvn_fn)\n",
        "    i = i+1\n",
        "\n",
        "  #print(act)\n",
        "\n",
        "  wandb.run.name = str(config.optimizer) +'_'+ str(config.activation) + '_bs_' + str(config.batch_size)  \n",
        "\n",
        "  model = MyNN(network_size=layers,network_fns=act, wb_init = config.wb_init, loss_fn=config.loss_fn,\n",
        "               optimizer=config.optimizer,max_epoch=config.max_epoch,verbose=config.verbose,seed=config.seed,\n",
        "               regularize =config.regularize,alpha = config.alpha, learning_rate = config.learning_rate, batch_size = config.batch_size)\n",
        "  \n",
        "  \n",
        "  model.fit(X,Y,X_valid,Y_valid)"
      ],
      "execution_count": 13,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N49PTE5dLDDX",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000,
          "referenced_widgets": [
            "381bcc66e2244834bce63c20278cf042",
            "5e3582c81b074b8581a2ec9476979092",
            "a810f5ab524340118978c0395d75a390",
            "b52c9116fee94bf990445481ca82a0b7",
            "8b9578ca96284cb29da8ddd09c4f044a",
            "547b4e975ff04d64867df12544ef123c",
            "16d3c0ad09e84733afc58ac26ea7e6be",
            "6525e2e8f63442c48130ea943e11cab8",
            "1b00178b18b14324acbffdf3c407d3aa",
            "68408edec4484fed8fabeaa1054b4046",
            "2094d189aa8c42a8bb58f0760561d770",
            "4ca6198b2601407c859a569c5f63beb9",
            "0b4d662530194c41b8272cba9abbb976",
            "f616c84340b14ca88c305438b41aaa35",
            "71ccbea484b94887bc18c54e3196381e",
            "e29f2ba3bb5a464ea7880da39a53f4ce"
          ]
        },
        "outputId": "dd25eb08-2463-42e2-eb79-1537fd522f3e"
      },
      "source": [
        "wandb.agent(sweep_id, train,count=2)"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\u001b[34m\u001b[1mwandb\u001b[0m: Agent Starting Run: otjc8prz with config:\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tactivation: tanh\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \talpha: 0\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tbatch_size: 16\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \thidden_size: 32\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tlearning_rate: 0.001\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tmax_epoch: 20\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tn_hidden: 5\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \toptimizer: ADAM\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \twb_init: xavier_uniform\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "                Tracking run with wandb version 0.10.22<br/>\n",
              "                Syncing run <strong style=\"color:#cdcd00\">flowing-sweep-3</strong> to <a href=\"https://wandb.ai\" target=\"_blank\">Weights & Biases</a> <a href=\"https://docs.wandb.com/integrations/jupyter.html\" target=\"_blank\">(Documentation)</a>.<br/>\n",
              "                Project page: <a href=\"https://wandb.ai/paddy3696/mnist_classification\" target=\"_blank\">https://wandb.ai/paddy3696/mnist_classification</a><br/>\n",
              "                Sweep page: <a href=\"https://wandb.ai/paddy3696/mnist_classification/sweeps/2fbqxb69\" target=\"_blank\">https://wandb.ai/paddy3696/mnist_classification/sweeps/2fbqxb69</a><br/>\n",
              "Run page: <a href=\"https://wandb.ai/paddy3696/mnist_classification/runs/otjc8prz\" target=\"_blank\">https://wandb.ai/paddy3696/mnist_classification/runs/otjc8prz</a><br/>\n",
              "                Run data is saved locally in <code>/content/wandb/run-20210316_080729-otjc8prz</code><br/><br/>\n",
              "            "
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\u001b[34m\u001b[1mwandb\u001b[0m: \u001b[33mWARNING\u001b[0m Config item 'wb_init' was locked by 'sweep' (ignored update).\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<br/>Waiting for W&B process to finish, PID 271<br/>Program ended successfully."
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "381bcc66e2244834bce63c20278cf042",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "VBox(children=(Label(value=' 0.00MB of 0.00MB uploaded (0.00MB deduped)\\r'), FloatProgress(value=1.0, max=1.0)…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "Find user logs for this run at: <code>/content/wandb/run-20210316_080729-otjc8prz/logs/debug.log</code>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "Find internal logs for this run at: <code>/content/wandb/run-20210316_080729-otjc8prz/logs/debug-internal.log</code>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<h3>Run summary:</h3><br/><style>\n",
              "    table.wandb td:nth-child(1) { padding: 0 10px; text-align: right }\n",
              "    </style><table class=\"wandb\">\n",
              "<tr><td>Training_accuracy</td><td>0.9105</td></tr><tr><td>_runtime</td><td>192</td></tr><tr><td>_timestamp</td><td>1615882241</td></tr><tr><td>_step</td><td>60</td></tr><tr><td>Validation_accuracy</td><td>0.88417</td></tr><tr><td>Training_Loss</td><td>0.14905</td></tr><tr><td>accuracy</td><td>0.88417</td></tr></table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<h3>Run history:</h3><br/><style>\n",
              "    table.wandb td:nth-child(1) { padding: 0 10px; text-align: right }\n",
              "    </style><table class=\"wandb\">\n",
              "<tr><td>Training_accuracy</td><td>▁▃▄▄▅▅▅▆▆▆▅▆▇▇▇▇▇▇██</td></tr><tr><td>_runtime</td><td>▁▁▁▁▂▂▂▂▂▂▃▃▃▃▄▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▇▇▇▇▇▇████</td></tr><tr><td>_timestamp</td><td>▁▁▁▁▂▂▂▂▂▂▃▃▃▃▄▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▇▇▇▇▇▇████</td></tr><tr><td>_step</td><td>▁▁▁▁▂▂▂▂▂▃▃▃▃▃▃▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▆▆▇▇▇▇▇███</td></tr><tr><td>Validation_accuracy</td><td>▁▄▆▄▆▇▇▇▆▆▆▇▇▇█▇▇▇▇█</td></tr><tr><td>Training_Loss</td><td>▃▅▆▇█▅█▃▁▆▃▅▃▅▅▂▁▁▃▂</td></tr><tr><td>accuracy</td><td>▁</td></tr></table><br/>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "                    <br/>Synced <strong style=\"color:#cdcd00\">flowing-sweep-3</strong>: <a href=\"https://wandb.ai/paddy3696/mnist_classification/runs/otjc8prz\" target=\"_blank\">https://wandb.ai/paddy3696/mnist_classification/runs/otjc8prz</a><br/>\n",
              "                "
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\u001b[34m\u001b[1mwandb\u001b[0m: Agent Starting Run: d3r7yf9o with config:\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tactivation: relu\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \talpha: 0.0005\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tbatch_size: 64\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \thidden_size: 64\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tlearning_rate: 0.001\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tmax_epoch: 20\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \tn_hidden: 5\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \toptimizer: NADAM\n",
            "\u001b[34m\u001b[1mwandb\u001b[0m: \twb_init: he\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "                Tracking run with wandb version 0.10.22<br/>\n",
              "                Syncing run <strong style=\"color:#cdcd00\">golden-sweep-2</strong> to <a href=\"https://wandb.ai\" target=\"_blank\">Weights & Biases</a> <a href=\"https://docs.wandb.com/integrations/jupyter.html\" target=\"_blank\">(Documentation)</a>.<br/>\n",
              "                Project page: <a href=\"https://wandb.ai/paddy3696/mnist_classification\" target=\"_blank\">https://wandb.ai/paddy3696/mnist_classification</a><br/>\n",
              "                Sweep page: <a href=\"https://wandb.ai/paddy3696/mnist_classification/sweeps/2fbqxb69\" target=\"_blank\">https://wandb.ai/paddy3696/mnist_classification/sweeps/2fbqxb69</a><br/>\n",
              "Run page: <a href=\"https://wandb.ai/paddy3696/mnist_classification/runs/d3r7yf9o\" target=\"_blank\">https://wandb.ai/paddy3696/mnist_classification/runs/d3r7yf9o</a><br/>\n",
              "                Run data is saved locally in <code>/content/wandb/run-20210316_081046-d3r7yf9o</code><br/><br/>\n",
              "            "
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\u001b[34m\u001b[1mwandb\u001b[0m: \u001b[33mWARNING\u001b[0m Config item 'wb_init' was locked by 'sweep' (ignored update).\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<br/>Waiting for W&B process to finish, PID 316<br/>Program ended successfully."
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "1b00178b18b14324acbffdf3c407d3aa",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "VBox(children=(Label(value=' 0.00MB of 0.00MB uploaded (0.00MB deduped)\\r'), FloatProgress(value=1.0, max=1.0)…"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "Find user logs for this run at: <code>/content/wandb/run-20210316_081046-d3r7yf9o/logs/debug.log</code>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "Find internal logs for this run at: <code>/content/wandb/run-20210316_081046-d3r7yf9o/logs/debug-internal.log</code>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<h3>Run summary:</h3><br/><style>\n",
              "    table.wandb td:nth-child(1) { padding: 0 10px; text-align: right }\n",
              "    </style><table class=\"wandb\">\n",
              "<tr><td>Training_accuracy</td><td>0.92715</td></tr><tr><td>_runtime</td><td>136</td></tr><tr><td>_timestamp</td><td>1615882382</td></tr><tr><td>_step</td><td>60</td></tr><tr><td>Validation_accuracy</td><td>0.8935</td></tr><tr><td>Training_Loss</td><td>0.05787</td></tr><tr><td>accuracy</td><td>0.89567</td></tr></table>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "<h3>Run history:</h3><br/><style>\n",
              "    table.wandb td:nth-child(1) { padding: 0 10px; text-align: right }\n",
              "    </style><table class=\"wandb\">\n",
              "<tr><td>Training_accuracy</td><td>▁▂▃▄▅▅▅▆▆▆▆▇▇▇▇▇▇▆██</td></tr><tr><td>_runtime</td><td>▁▁▁▁▂▂▂▂▂▂▃▃▃▃▄▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▆▆▇▇▇▇████</td></tr><tr><td>_timestamp</td><td>▁▁▁▁▂▂▂▂▂▂▃▃▃▃▄▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▆▆▇▇▇▇████</td></tr><tr><td>_step</td><td>▁▁▁▁▂▂▂▂▂▃▃▃▃▃▃▄▄▄▄▄▅▅▅▅▅▅▆▆▆▆▆▆▇▇▇▇▇███</td></tr><tr><td>Validation_accuracy</td><td>▁▃▃▅▆▆▆▆▆▇▆▆▇▇▆▇▇▆██</td></tr><tr><td>Training_Loss</td><td>▇▆▄▅▆▄▃▃▃▄▄▅▃▃▄▄▂█▃▁</td></tr><tr><td>accuracy</td><td>▁</td></tr></table><br/>"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)"
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/html": [
              "\n",
              "                    <br/>Synced <strong style=\"color:#cdcd00\">golden-sweep-2</strong>: <a href=\"https://wandb.ai/paddy3696/mnist_classification/runs/d3r7yf9o\" target=\"_blank\">https://wandb.ai/paddy3696/mnist_classification/runs/d3r7yf9o</a><br/>\n",
              "                "
            ],
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    }
  ]
}