LSTM + Attention.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Task2_NLP.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/nthanhkhang/ChatBots-to-Question-Answer-systems/blob/main/Task2_NLP.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "UlOoeawDGuki"
      },
      "source": [
        "# I.Import Library"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pXxMKJcznI1F",
        "outputId": "a472fc6b-2039-4b43-b0bc-1f9a8267414a"
      },
      "source": [
        "import re\n",
        "import pandas as pd\n",
        "from nltk.corpus import stopwords\n",
        "\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from keras.preprocessing.text import Tokenizer\n",
        "from keras.preprocessing.sequence import pad_sequences\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns\n",
        "sns.set_style(\"whitegrid\")\n",
        "\n",
        "from keras.layers import Input,Embedding,Bidirectional,LSTM,Dense,Concatenate\n",
        "from keras.models import Model\n",
        "import nltk\n",
        "nltk.download('stopwords')\n",
        "nltk.download('punkt')\n",
        "import numpy as np"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[nltk_data] Downloading package stopwords to /root/nltk_data...\n",
            "[nltk_data]   Package stopwords is already up-to-date!\n",
            "[nltk_data] Downloading package punkt to /root/nltk_data...\n",
            "[nltk_data]   Package punkt is already up-to-date!\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yJ9MrGk6G8FU"
      },
      "source": [
        "# II.Read Data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pmF1pX90nL8y",
        "outputId": "e5a56602-ac4e-44bb-d319-d34a20b2a1ed"
      },
      "source": [
        "import zipfile\n",
        "with zipfile.ZipFile(\"DataChatbot.zip\",\"r\") as zf:\n",
        "    zf.extractall('DataChatbot')\n",
        "print(zf)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "<zipfile.ZipFile [closed]>\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "nMhIcCCxnWX9",
        "outputId": "96a59696-d63c-4f36-f983-e387a23ea664"
      },
      "source": [
        "questions=[]\n",
        "answers=[]\n",
        "error=0\n",
        "import glob\n",
        "import os\n",
        "path = 'DataChatbot'\n",
        "for filename in glob.glob(os.path.join(path, '*.txt')):\n",
        "   with open(os.path.join(os.getcwd(), filename), 'r') as f: # open in readonly mode\n",
        "      for line in f:\n",
        "        part = line.strip().strip('__eou_').split('__eou__')\n",
        "        questions.append(\", \".join(part[:-1]))\n",
        "        answers.append(part[-1])\n",
        "print('Question',questions[:5])\n",
        "print('Answers',answers[:5])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Question ['bạn đã có việc làm chưa ? ', 'bạn đã có việc làm chưa ? ', 'bạn đã đi làm hay chưa? ', 'hiện tại bạn làm ở đâu? ', 'địa chỉ nơi làm việc của bạn ở đâu? ']\n",
            "Answers [' chưa, mình đang thất nghiệp ', ' rồi, mình đang làm việc cho một công ty ', ' mình đã đi làm rồi ', ' hiện tại mình làm ở FPT software ', ' ở khu công nghệ cao ']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 197
        },
        "id": "oW2Bdj_Sndnw",
        "outputId": "b00b9c48-0901-439b-af84-ffc9f513461f"
      },
      "source": [
        "df = pd.DataFrame(questions)\n",
        "df.columns = [\"Questions\"]\n",
        "df['Answers']=df['Questions'].apply(len)\n",
        "df['Answers'] = pd.DataFrame(answers)\n",
        "df.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Questions</th>\n",
              "      <th>Answers</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>bạn đã có việc làm chưa ?</td>\n",
              "      <td>chưa, mình đang thất nghiệp</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>bạn đã có việc làm chưa ?</td>\n",
              "      <td>rồi, mình đang làm việc cho một công ty</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>bạn đã đi làm hay chưa?</td>\n",
              "      <td>mình đã đi làm rồi</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>hiện tại bạn làm ở đâu?</td>\n",
              "      <td>hiện tại mình làm ở FPT software</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>địa chỉ nơi làm việc của bạn ở đâu?</td>\n",
              "      <td>ở khu công nghệ cao</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                              Questions                                    Answers\n",
              "0            bạn đã có việc làm chưa ?                chưa, mình đang thất nghiệp \n",
              "1            bạn đã có việc làm chưa ?    rồi, mình đang làm việc cho một công ty \n",
              "2              bạn đã đi làm hay chưa?                         mình đã đi làm rồi \n",
              "3              hiện tại bạn làm ở đâu?           hiện tại mình làm ở FPT software \n",
              "4  địa chỉ nơi làm việc của bạn ở đâu?                        ở khu công nghệ cao "
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 6
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 384
        },
        "id": "zB7avYexntYz",
        "outputId": "76066c47-9298-4a2f-afd3-4aac3f6a175d"
      },
      "source": [
        "print(df.shape)\n",
        "print(\"=\"*60)\n",
        "print(df.info())\n",
        "print(\"=\"*60)\n",
        "df.describe()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(5855, 2)\n",
            "============================================================\n",
            "<class 'pandas.core.frame.DataFrame'>\n",
            "RangeIndex: 5855 entries, 0 to 5854\n",
            "Data columns (total 2 columns):\n",
            " #   Column     Non-Null Count  Dtype \n",
            "---  ------     --------------  ----- \n",
            " 0   Questions  5855 non-null   object\n",
            " 1   Answers    5855 non-null   object\n",
            "dtypes: object(2)\n",
            "memory usage: 91.6+ KB\n",
            "None\n",
            "============================================================\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Questions</th>\n",
              "      <th>Answers</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>count</th>\n",
              "      <td>5855</td>\n",
              "      <td>5855</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>unique</th>\n",
              "      <td>4919</td>\n",
              "      <td>5067</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>top</th>\n",
              "      <td>bạn sinh năm nào?</td>\n",
              "      <td></td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>freq</th>\n",
              "      <td>25</td>\n",
              "      <td>86</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                 Questions Answers\n",
              "count                 5855    5855\n",
              "unique                4919    5067\n",
              "top     bạn sinh năm nào?         \n",
              "freq                    25      86"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 167
        },
        "id": "A4qL4mlqnwjC",
        "outputId": "4211cbfc-07db-4844-aba3-de836273071c"
      },
      "source": [
        "df.drop_duplicates(subset=['Questions'],inplace=True)\n",
        "df.describe()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Questions</th>\n",
              "      <th>Answers</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>count</th>\n",
              "      <td>4919</td>\n",
              "      <td>4919</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>unique</th>\n",
              "      <td>4919</td>\n",
              "      <td>4632</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>top</th>\n",
              "      <td>Thông thường từ nhà bạn đến trường tốn bào lâu?</td>\n",
              "      <td></td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>freq</th>\n",
              "      <td>1</td>\n",
              "      <td>77</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                               Questions Answers\n",
              "count                                               4919    4919\n",
              "unique                                              4919    4632\n",
              "top     Thông thường từ nhà bạn đến trường tốn bào lâu?         \n",
              "freq                                                   1      77"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "iZFycvHbn2PM",
        "outputId": "7219eaaf-72b6-47b7-f5f3-6636c7aef0b4"
      },
      "source": [
        "print(df.isnull().sum())\n",
        "print('\\n')\n",
        "print(df.isnull().any())"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Questions    0\n",
            "Answers      0\n",
            "dtype: int64\n",
            "\n",
            "\n",
            "Questions    False\n",
            "Answers      False\n",
            "dtype: bool\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 227
        },
        "id": "Tc8kH-MKn4U1",
        "outputId": "95d6a7a6-0333-4b7a-a09c-c3aa92a57fa0"
      },
      "source": [
        "df.loc[55:60,:]"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Questions</th>\n",
              "      <th>Answers</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>55</th>\n",
              "      <td>bạn có đang đi làm không?</td>\n",
              "      <td>có, hiện tại mình đang đi dạy thêm lập trình ...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>56</th>\n",
              "      <td>bạn đang làm ở đâu?</td>\n",
              "      <td>mình đang làm ở học viện công nghệ TEKY</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>57</th>\n",
              "      <td>bạn đi làm có vui không?</td>\n",
              "      <td>chắc chắn là có rồi, làm việc mình thích cơ mà</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>58</th>\n",
              "      <td>đồng nghiệp của bạn như thế nào?</td>\n",
              "      <td>từ sếp lớn cho tới các bạn giảng viên ai cũng...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>59</th>\n",
              "      <td>lương của công việc hiện tại của bạn cao không?</td>\n",
              "      <td>lương cũng đủ để tiêu vặt</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>60</th>\n",
              "      <td>Mày đang làm cho công ty nào vậy ?</td>\n",
              "      <td>Tao đang làm cho một agency marketing nổi tiếng</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                           Questions                                            Answers\n",
              "55                        bạn có đang đi làm không?    có, hiện tại mình đang đi dạy thêm lập trình ...\n",
              "56                              bạn đang làm ở đâu?            mình đang làm ở học viện công nghệ TEKY \n",
              "57                         bạn đi làm có vui không?     chắc chắn là có rồi, làm việc mình thích cơ mà \n",
              "58                 đồng nghiệp của bạn như thế nào?    từ sếp lớn cho tới các bạn giảng viên ai cũng...\n",
              "59  lương của công việc hiện tại của bạn cao không?                          lương cũng đủ để tiêu vặt \n",
              "60               Mày đang làm cho công ty nào vậy ?    Tao đang làm cho một agency marketing nổi tiếng "
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "EaiJizDfFP55"
      },
      "source": [
        "# III.Text Preprocessing\n",
        "\n",
        "## Lowercase\n",
        "\n",
        "## Remove characters which are not english or numbers"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Rg1xDN2xn-tO"
      },
      "source": [
        "def pre_process(questions):\n",
        "    stop_words = stopwords.words(\"english\")\n",
        "    \n",
        "    # Remove non english words\n",
        "    questions = [re.sub('[^a-z(c++)(c#)]', ' ', x.lower()) for x in questions]\n",
        "    # Tokenlization\n",
        "    questions_tokens = [nltk.word_tokenize(t) for t in questions]\n",
        "    # Removing Stop Words\n",
        "    questions_stop = [[t for t in tokens if (t not in stop_words) and (3 < len(t.strip()) < 15)]\n",
        "                      for tokens in questions_tokens]\n",
        "    \n",
        "    questions_stop = pd.Series(questions_stop)\n",
        "    return questions_stop"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 407
        },
        "id": "GD8kqaldoBXQ",
        "outputId": "0caf5a24-f077-4e4d-f786-f6f1634a4c4f"
      },
      "source": [
        "# Initial preprocessing training data\n",
        "# Initial preprocessing training data\n",
        "questions = df['Questions']\n",
        "questions_pp = pre_process(questions)\n",
        "df"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Questions</th>\n",
              "      <th>Answers</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>bạn đã có việc làm chưa ?</td>\n",
              "      <td>chưa, mình đang thất nghiệp</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>bạn đã đi làm hay chưa?</td>\n",
              "      <td>mình đã đi làm rồi</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>hiện tại bạn làm ở đâu?</td>\n",
              "      <td>hiện tại mình làm ở FPT software</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>địa chỉ nơi làm việc của bạn ở đâu?</td>\n",
              "      <td>ở khu công nghệ cao</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5</th>\n",
              "      <td>bạn làm công việc gì?</td>\n",
              "      <td>mình đang là kĩ sư phần mêm</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5846</th>\n",
              "      <td>bạn có thích anime?</td>\n",
              "      <td>yes</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5847</th>\n",
              "      <td>bạn thích xem phim không ?</td>\n",
              "      <td>Thích chứ ông</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5852</th>\n",
              "      <td>Vào mỗi dịp rảnh rỗi, em thường làm gì nhất</td>\n",
              "      <td>Em thường không làm gì cả</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5853</th>\n",
              "      <td>Sao không thử nghe nhạc xem, nó giúp em ổn hơn...</td>\n",
              "      <td>Em chỉ nghe lúc em muốn thôi</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5854</th>\n",
              "      <td>Thế em ghét làm việc gì nhất</td>\n",
              "      <td>Ghét nhất là phải làm bài tập về nhà</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>4919 rows × 2 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "                                              Questions                                 Answers\n",
              "0                            bạn đã có việc làm chưa ?             chưa, mình đang thất nghiệp \n",
              "2                              bạn đã đi làm hay chưa?                      mình đã đi làm rồi \n",
              "3                              hiện tại bạn làm ở đâu?        hiện tại mình làm ở FPT software \n",
              "4                  địa chỉ nơi làm việc của bạn ở đâu?                     ở khu công nghệ cao \n",
              "5                                bạn làm công việc gì?             mình đang là kĩ sư phần mêm \n",
              "...                                                 ...                                     ...\n",
              "5846                               bạn có thích anime?                                     yes \n",
              "5847                        bạn thích xem phim không ?                           Thích chứ ông \n",
              "5852       Vào mỗi dịp rảnh rỗi, em thường làm gì nhất               Em thường không làm gì cả \n",
              "5853  Sao không thử nghe nhạc xem, nó giúp em ổn hơn...           Em chỉ nghe lúc em muốn thôi \n",
              "5854                      Thế em ghét làm việc gì nhất    Ghét nhất là phải làm bài tập về nhà \n",
              "\n",
              "[4919 rows x 2 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 197
        },
        "id": "cg_Qmm6MoXDk",
        "outputId": "32a7c29b-4bc0-4a57-ab80-82dfe7c70963"
      },
      "source": [
        "df['decoder_input'] = df.Answers.apply(lambda x: 'sos '+x)\n",
        "df['decoder_label'] = df.Answers.apply(lambda x: x+' eo>')\n",
        "df.head()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>Questions</th>\n",
              "      <th>Answers</th>\n",
              "      <th>decoder_input</th>\n",
              "      <th>decoder_label</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>bạn đã có việc làm chưa ?</td>\n",
              "      <td>chưa, mình đang thất nghiệp</td>\n",
              "      <td>sos  chưa, mình đang thất nghiệp</td>\n",
              "      <td>chưa, mình đang thất nghiệp  eo&gt;</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>bạn đã đi làm hay chưa?</td>\n",
              "      <td>mình đã đi làm rồi</td>\n",
              "      <td>sos  mình đã đi làm rồi</td>\n",
              "      <td>mình đã đi làm rồi  eo&gt;</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>hiện tại bạn làm ở đâu?</td>\n",
              "      <td>hiện tại mình làm ở FPT software</td>\n",
              "      <td>sos  hiện tại mình làm ở FPT software</td>\n",
              "      <td>hiện tại mình làm ở FPT software  eo&gt;</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>địa chỉ nơi làm việc của bạn ở đâu?</td>\n",
              "      <td>ở khu công nghệ cao</td>\n",
              "      <td>sos  ở khu công nghệ cao</td>\n",
              "      <td>ở khu công nghệ cao  eo&gt;</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>5</th>\n",
              "      <td>bạn làm công việc gì?</td>\n",
              "      <td>mình đang là kĩ sư phần mêm</td>\n",
              "      <td>sos  mình đang là kĩ sư phần mêm</td>\n",
              "      <td>mình đang là kĩ sư phần mêm  eo&gt;</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                              Questions  ...                           decoder_label\n",
              "0            bạn đã có việc làm chưa ?   ...        chưa, mình đang thất nghiệp  eo>\n",
              "2              bạn đã đi làm hay chưa?   ...                 mình đã đi làm rồi  eo>\n",
              "3              hiện tại bạn làm ở đâu?   ...   hiện tại mình làm ở FPT software  eo>\n",
              "4  địa chỉ nơi làm việc của bạn ở đâu?   ...                ở khu công nghệ cao  eo>\n",
              "5                bạn làm công việc gì?   ...        mình đang là kĩ sư phần mêm  eo>\n",
              "\n",
              "[5 rows x 4 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "Jgzbg0Juop9j",
        "outputId": "0979de92-70a2-432d-9471-e026346623bb"
      },
      "source": [
        "encoder_input = np.array(df.Questions)\n",
        "decoder_input = np.array(df.decoder_input)\n",
        "decoder_label = np.array(df.decoder_label)\n",
        "\n",
        "n_rows = df.shape[0]\n",
        "print(f\"{n_rows} rows\")\n",
        "\n",
        "indices = np.arange(n_rows)\n",
        "np.random.shuffle(indices)\n",
        "\n",
        "encoder_input = encoder_input[indices]\n",
        "decoder_input = decoder_input[indices]\n",
        "decoder_label = decoder_label[indices]\n",
        "\n",
        "train_size = 0.9\n",
        "\n",
        "train_encoder_input = encoder_input[:int(n_rows*train_size)]\n",
        "train_decoder_input = decoder_input[:int(n_rows*train_size)]\n",
        "train_decoder_label = decoder_label[:int(n_rows*train_size)]\n",
        "\n",
        "test_encoder_input = encoder_input[int(n_rows*train_size):]\n",
        "test_decoder_input = decoder_input[int(n_rows*train_size):]\n",
        "test_decoder_label = decoder_label[int(n_rows*train_size):]\n",
        "\n",
        "print(train_encoder_input.shape)\n",
        "print(train_decoder_input.shape)\n",
        "print(train_decoder_label.shape)\n",
        "\n",
        "print(test_encoder_input.shape)\n",
        "print(test_decoder_input.shape)\n",
        "print(test_decoder_label.shape)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "4919 rows\n",
            "(4427,)\n",
            "(4427,)\n",
            "(4427,)\n",
            "(492,)\n",
            "(492,)\n",
            "(492,)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "aJ7bnMy8o6uh",
        "outputId": "b73ff0d4-350d-4754-b602-30f4d4fe95b0"
      },
      "source": [
        "q_tok = Tokenizer()\n",
        "q_tok.fit_on_texts(train_encoder_input)\n",
        "print(len(q_tok.word_counts))\n",
        "\n",
        "a_tok = Tokenizer()\n",
        "a_tok.fit_on_texts(train_decoder_input)\n",
        "a_tok.fit_on_texts(train_decoder_label)\n",
        "print(len(a_tok.word_counts))\n",
        "\n",
        "train_encoder_input = q_tok.texts_to_sequences(train_encoder_input)\n",
        "test_encoder_input = q_tok.texts_to_sequences(test_encoder_input)\n",
        "\n",
        "train_decoder_input = a_tok.texts_to_sequences(train_decoder_input)\n",
        "test_decoder_input = a_tok.texts_to_sequences(test_decoder_input)\n",
        "\n",
        "train_decoder_label = a_tok.texts_to_sequences(train_decoder_label)\n",
        "test_decoder_label = a_tok.texts_to_sequences(test_decoder_label)\n",
        "\n",
        "question_word_size = len(q_tok.word_counts)\n",
        "question_vocab_size = question_word_size+1\n",
        "\n",
        "answer_word_size = len(a_tok.word_counts)\n",
        "answer_vocab_size = answer_word_size+1"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "2056\n",
            "2984\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 310
        },
        "id": "rqv9Hteyo9M_",
        "outputId": "f8bcf9fa-8dda-49fe-a6a5-32a118ce8121"
      },
      "source": [
        "question_lengths = [len(s) for s in train_encoder_input]\n",
        "print(f\"maximum question sequence length >> {np.max(question_lengths)}\")\n",
        "\n",
        "answer_lengths = [len(s) for s in train_decoder_input]\n",
        "print(f\"maximum answer sequence length >> {np.max(answer_lengths)}\")\n",
        "\n",
        "plt.subplot(2,1,1)\n",
        "plt.hist(question_lengths,bins=50)\n",
        "plt.show()\n",
        "\n",
        "plt.subplot(2,1,2)\n",
        "plt.hist(answer_lengths,bins=50)\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "maximum question sequence length >> 93\n",
            "maximum answer sequence length >> 157\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "c7zGAfsyo-Uw",
        "outputId": "d528144c-a44f-4efa-dd9b-05a95b6eb8a7"
      },
      "source": [
        "train_encoder_input = pad_sequences(train_encoder_input,padding='post')\n",
        "train_decoder_input = pad_sequences(train_decoder_input,padding='post')\n",
        "train_decoder_label = pad_sequences(train_decoder_label,padding='post')\n",
        "\n",
        "print(train_encoder_input.shape)\n",
        "print(train_decoder_input.shape)\n",
        "print(train_decoder_label.shape)\n",
        "\n",
        "question_sequence_size = train_encoder_input.shape[1]\n",
        "answer_sequence_size = train_decoder_input.shape[1]\n",
        "\n",
        "test_encoder_input = pad_sequences(test_encoder_input,padding='post',maxlen=question_sequence_size)\n",
        "test_decoder_input = pad_sequences(test_decoder_input,padding='post',maxlen=answer_sequence_size)\n",
        "test_decoder_label = pad_sequences(test_decoder_label,padding='post',maxlen=answer_sequence_size)\n",
        "\n",
        "print(test_encoder_input.shape)\n",
        "print(test_decoder_input.shape)\n",
        "print(test_decoder_label.shape)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(4427, 93)\n",
            "(4427, 157)\n",
            "(4427, 157)\n",
            "(492, 93)\n",
            "(492, 157)\n",
            "(492, 157)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gyx0L4ecFfgs"
      },
      "source": [
        "# IV.Encoder-Decoder Model Attention"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jNyXZooarCS1"
      },
      "source": [
        "embedding_size=256"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "36BL1IZ4rHCw"
      },
      "source": [
        "import tensorflow\n",
        "from tensorflow.keras.layers import Attention"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "axwTGRZRrKl6",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "50b4051e-60c9-4079-c5aa-9a96db96faa0"
      },
      "source": [
        "#trainer model\n",
        "\n",
        "hidden_size = 256\n",
        "\n",
        "#encoder part start\n",
        "encoder_input = Input(shape=[question_sequence_size])\n",
        "# encoder_embedding = Embedding(question_vocab_size,embedding_size,mask_zero=True,trainable=False,weights=[question_embedding_matrix])\n",
        "encoder_embedding = Embedding(question_vocab_size,embedding_size,mask_zero=True)\n",
        "encoder_embedded = encoder_embedding(encoder_input)\n",
        "\n",
        "lstm1 = LSTM(hidden_size,return_sequences=True,return_state=True,dropout=0.3,recurrent_dropout=0.3)\n",
        "encoder_output1,_,_= lstm1(encoder_embedded)\n",
        "lstm2 = LSTM(hidden_size,return_sequences=True,return_state=True,dropout=0.3,recurrent_dropout=0.3)\n",
        "encoder_output2,_,_ = lstm2(encoder_output1)\n",
        "lstm3 = LSTM(hidden_size,return_sequences=True,return_state=True,dropout=0.3,recurrent_dropout=0.3)\n",
        "encoder_output3,encoder_h3,encoder_c3= lstm3(encoder_output2)\n",
        "#encoder part done\n",
        "\n",
        "#decoder part start\n",
        "decoder_input = Input(shape=(None,))\n",
        "# decoder_embedding = Embedding(answer_vocab_size,embedding_size,mask_zero=True,trainable=False,weights=[answer_embedding_matrix])\n",
        "decoder_embedding = Embedding(answer_vocab_size,embedding_size,mask_zero=True)\n",
        "decoder_embedded = decoder_embedding(decoder_input)\n",
        "\n",
        "decoder_lstm = LSTM(hidden_size,return_sequences=True,return_state=True)\n",
        "decoder_output,_,_ = decoder_lstm(decoder_embedded,initial_state=[encoder_h3,encoder_c3])\n",
        "\n",
        "#Attention layer (query,value)\n",
        "attn_layer = Attention()\n",
        "attention_context = attn_layer([decoder_output,encoder_output3])\n",
        "\n",
        "decoder_output = Concatenate(axis=-1)([decoder_output,attention_context])\n",
        "\n",
        "dense1 = Dense(512,activation='relu')\n",
        "decoder_output = dense1(decoder_output)\n",
        "\n",
        "decoder_softmax = Dense(answer_vocab_size,activation='softmax')\n",
        "decoder_output = decoder_softmax(decoder_output)\n",
        "\n",
        "trainer = Model([encoder_input,decoder_input],decoder_output)\n",
        "loss = tensorflow.keras.losses.SparseCategoricalCrossentropy(from_logits=False)\n",
        "trainer.compile(loss=loss,optimizer='adam',metrics=['accuracy'])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:Layer lstm will not use cuDNN kernel since it doesn't meet the cuDNN kernel criteria. It will use generic GPU kernel as fallback when running on GPU\n",
            "WARNING:tensorflow:Layer lstm_1 will not use cuDNN kernel since it doesn't meet the cuDNN kernel criteria. It will use generic GPU kernel as fallback when running on GPU\n",
            "WARNING:tensorflow:Layer lstm_2 will not use cuDNN kernel since it doesn't meet the cuDNN kernel criteria. It will use generic GPU kernel as fallback when running on GPU\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WJCPO8tDrOWf",
        "outputId": "4509aa93-57bb-41fa-d2a2-4a2aca2c5db6"
      },
      "source": [
        "train_hist = trainer.fit([train_encoder_input,train_decoder_input],train_decoder_label,epochs=45,validation_split=0.1,batch_size=32)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/45\n",
            "125/125 [==============================] - 198s 1s/step - loss: 0.4223 - accuracy: 0.1044 - val_loss: 0.3450 - val_accuracy: 0.1405\n",
            "Epoch 2/45\n",
            "125/125 [==============================] - 153s 1s/step - loss: 0.3543 - accuracy: 0.1312 - val_loss: 0.3353 - val_accuracy: 0.1517\n",
            "Epoch 3/45\n",
            "125/125 [==============================] - 154s 1s/step - loss: 0.3464 - accuracy: 0.1423 - val_loss: 0.3215 - val_accuracy: 0.1633\n",
            "Epoch 4/45\n",
            "125/125 [==============================] - 153s 1s/step - loss: 0.3319 - accuracy: 0.1579 - val_loss: 0.3130 - val_accuracy: 0.1794\n",
            "Epoch 5/45\n",
            "125/125 [==============================] - 152s 1s/step - loss: 0.2998 - accuracy: 0.1781 - val_loss: 0.3084 - val_accuracy: 0.1937\n",
            "Epoch 6/45\n",
            "125/125 [==============================] - 150s 1s/step - loss: 0.2846 - accuracy: 0.1940 - val_loss: 0.3058 - val_accuracy: 0.2134\n",
            "Epoch 7/45\n",
            "125/125 [==============================] - 150s 1s/step - loss: 0.2749 - accuracy: 0.2141 - val_loss: 0.3047 - val_accuracy: 0.2214\n",
            "Epoch 8/45\n",
            "125/125 [==============================] - 149s 1s/step - loss: 0.2494 - accuracy: 0.2389 - val_loss: 0.3090 - val_accuracy: 0.2314\n",
            "Epoch 9/45\n",
            "125/125 [==============================] - 148s 1s/step - loss: 0.2321 - accuracy: 0.2673 - val_loss: 0.3119 - val_accuracy: 0.2348\n",
            "Epoch 10/45\n",
            "125/125 [==============================] - 149s 1s/step - loss: 0.2124 - accuracy: 0.2963 - val_loss: 0.3208 - val_accuracy: 0.2355\n",
            "Epoch 11/45\n",
            "125/125 [==============================] - 148s 1s/step - loss: 0.1936 - accuracy: 0.3423 - val_loss: 0.3275 - val_accuracy: 0.2438\n",
            "Epoch 12/45\n",
            "125/125 [==============================] - 148s 1s/step - loss: 0.1755 - accuracy: 0.3874 - val_loss: 0.3415 - val_accuracy: 0.2338\n",
            "Epoch 13/45\n",
            "125/125 [==============================] - 147s 1s/step - loss: 0.1573 - accuracy: 0.4251 - val_loss: 0.3528 - val_accuracy: 0.2428\n",
            "Epoch 14/45\n",
            "125/125 [==============================] - 145s 1s/step - loss: 0.1410 - accuracy: 0.4755 - val_loss: 0.3657 - val_accuracy: 0.2285\n",
            "Epoch 15/45\n",
            "125/125 [==============================] - 149s 1s/step - loss: 0.1243 - accuracy: 0.5237 - val_loss: 0.3806 - val_accuracy: 0.2273\n",
            "Epoch 16/45\n",
            "125/125 [==============================] - 150s 1s/step - loss: 0.1121 - accuracy: 0.5648 - val_loss: 0.3947 - val_accuracy: 0.2297\n",
            "Epoch 17/45\n",
            "125/125 [==============================] - 147s 1s/step - loss: 0.0984 - accuracy: 0.6140 - val_loss: 0.4135 - val_accuracy: 0.2353\n",
            "Epoch 18/45\n",
            "125/125 [==============================] - 148s 1s/step - loss: 0.0858 - accuracy: 0.6549 - val_loss: 0.4287 - val_accuracy: 0.2222\n",
            "Epoch 19/45\n",
            "125/125 [==============================] - 146s 1s/step - loss: 0.0757 - accuracy: 0.6947 - val_loss: 0.4448 - val_accuracy: 0.2173\n",
            "Epoch 20/45\n",
            "125/125 [==============================] - 147s 1s/step - loss: 0.0657 - accuracy: 0.7399 - val_loss: 0.4613 - val_accuracy: 0.2224\n",
            "Epoch 21/45\n",
            "125/125 [==============================] - 148s 1s/step - loss: 0.0561 - accuracy: 0.7726 - val_loss: 0.4794 - val_accuracy: 0.2299\n",
            "Epoch 22/45\n",
            "125/125 [==============================] - 146s 1s/step - loss: 0.0478 - accuracy: 0.8065 - val_loss: 0.4913 - val_accuracy: 0.2175\n",
            "Epoch 23/45\n",
            "125/125 [==============================] - 146s 1s/step - loss: 0.0421 - accuracy: 0.8301 - val_loss: 0.5071 - val_accuracy: 0.2236\n",
            "Epoch 24/45\n",
            "125/125 [==============================] - 146s 1s/step - loss: 0.0349 - accuracy: 0.8612 - val_loss: 0.5224 - val_accuracy: 0.2178\n",
            "Epoch 25/45\n",
            "125/125 [==============================] - 142s 1s/step - loss: 0.0302 - accuracy: 0.8824 - val_loss: 0.5354 - val_accuracy: 0.2168\n",
            "Epoch 26/45\n",
            "125/125 [==============================] - 143s 1s/step - loss: 0.0263 - accuracy: 0.8975 - val_loss: 0.5530 - val_accuracy: 0.2197\n",
            "Epoch 27/45\n",
            "125/125 [==============================] - 143s 1s/step - loss: 0.0220 - accuracy: 0.9169 - val_loss: 0.5657 - val_accuracy: 0.2146\n",
            "Epoch 28/45\n",
            "125/125 [==============================] - 145s 1s/step - loss: 0.0193 - accuracy: 0.9262 - val_loss: 0.5824 - val_accuracy: 0.2183\n",
            "Epoch 29/45\n",
            "125/125 [==============================] - 144s 1s/step - loss: 0.0173 - accuracy: 0.9358 - val_loss: 0.5889 - val_accuracy: 0.2154\n",
            "Epoch 30/45\n",
            "125/125 [==============================] - 144s 1s/step - loss: 0.0148 - accuracy: 0.9435 - val_loss: 0.6006 - val_accuracy: 0.2188\n",
            "Epoch 31/45\n",
            "125/125 [==============================] - 142s 1s/step - loss: 0.0132 - accuracy: 0.9484 - val_loss: 0.6155 - val_accuracy: 0.2158\n",
            "Epoch 32/45\n",
            "125/125 [==============================] - 142s 1s/step - loss: 0.0117 - accuracy: 0.9531 - val_loss: 0.6258 - val_accuracy: 0.2173\n",
            "Epoch 33/45\n",
            "125/125 [==============================] - 142s 1s/step - loss: 0.0102 - accuracy: 0.9632 - val_loss: 0.6332 - val_accuracy: 0.2146\n",
            "Epoch 34/45\n",
            "125/125 [==============================] - 142s 1s/step - loss: 0.0089 - accuracy: 0.9653 - val_loss: 0.6421 - val_accuracy: 0.2178\n",
            "Epoch 35/45\n",
            "125/125 [==============================] - 145s 1s/step - loss: 0.0082 - accuracy: 0.9699 - val_loss: 0.6477 - val_accuracy: 0.2193\n",
            "Epoch 36/45\n",
            "125/125 [==============================] - 145s 1s/step - loss: 0.0078 - accuracy: 0.9696 - val_loss: 0.6537 - val_accuracy: 0.2151\n",
            "Epoch 37/45\n",
            "125/125 [==============================] - 153s 1s/step - loss: 0.0071 - accuracy: 0.9741 - val_loss: 0.6656 - val_accuracy: 0.2185\n",
            "Epoch 38/45\n",
            "125/125 [==============================] - 154s 1s/step - loss: 0.0071 - accuracy: 0.9724 - val_loss: 0.6686 - val_accuracy: 0.2127\n",
            "Epoch 39/45\n",
            "125/125 [==============================] - 149s 1s/step - loss: 0.0076 - accuracy: 0.9709 - val_loss: 0.6704 - val_accuracy: 0.2175\n",
            "Epoch 40/45\n",
            "125/125 [==============================] - 149s 1s/step - loss: 0.0072 - accuracy: 0.9719 - val_loss: 0.6814 - val_accuracy: 0.2171\n",
            "Epoch 41/45\n",
            "125/125 [==============================] - 150s 1s/step - loss: 0.0066 - accuracy: 0.9745 - val_loss: 0.6776 - val_accuracy: 0.2112\n",
            "Epoch 42/45\n",
            "125/125 [==============================] - 150s 1s/step - loss: 0.0069 - accuracy: 0.9739 - val_loss: 0.6832 - val_accuracy: 0.2166\n",
            "Epoch 43/45\n",
            "125/125 [==============================] - 150s 1s/step - loss: 0.0081 - accuracy: 0.9693 - val_loss: 0.6925 - val_accuracy: 0.2158\n",
            "Epoch 44/45\n",
            "125/125 [==============================] - 151s 1s/step - loss: 0.0083 - accuracy: 0.9661 - val_loss: 0.6886 - val_accuracy: 0.2100\n",
            "Epoch 45/45\n",
            "125/125 [==============================] - 150s 1s/step - loss: 0.0067 - accuracy: 0.9729 - val_loss: 0.6946 - val_accuracy: 0.2093\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-CT-daJ3F_-B"
      },
      "source": [
        "## 1.Generator model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JzeDsYhOF8ON"
      },
      "source": [
        "from tensorflow.keras.utils import plot_model\n",
        "\n",
        "gen_encoder = Model(encoder_input,[encoder_output3,encoder_h3,encoder_c3])\n",
        "\n",
        "gen_decoder_value_input = Input(shape=(question_sequence_size,hidden_size))\n",
        "gen_decoder_h_input = Input(shape=[hidden_size])\n",
        "gen_decoder_c_input = Input(shape=[hidden_size])\n",
        "\n",
        "gen_decoder_embedded = decoder_embedding(decoder_input)\n",
        "gen_decoder_output,gen_decoder_h,gen_decoder_c = decoder_lstm(gen_decoder_embedded,initial_state=[gen_decoder_h_input,gen_decoder_c_input])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "LT36_3v8GGGC"
      },
      "source": [
        "## 2.Attention([querys,values])"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "X4TCQXwkGGwa"
      },
      "source": [
        "gen_attention_context = attn_layer([gen_decoder_output,gen_decoder_value_input])\n",
        "gen_decoder_output = Concatenate(axis=-1)([gen_decoder_output,gen_attention_context])\n",
        "\n",
        "gen_decoder_output = dense1(gen_decoder_output)\n",
        "gen_decoder_output = decoder_softmax(gen_decoder_output)\n",
        "\n",
        "gen_decoder = Model([decoder_input]+[gen_decoder_value_input,gen_decoder_h_input,gen_decoder_c_input],[gen_decoder_output]+[gen_decoder_h,gen_decoder_c])"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 466
        },
        "id": "xNNoeQOzIINm",
        "outputId": "89ab3220-14b9-471f-9e8d-6f55949d06f7"
      },
      "source": [
        "plot_model(gen_encoder)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAANsAAAHBCAYAAADguc4JAAAABmJLR0QA/wD/AP+gvaeTAAAgAElEQVR4nO3de1hU9b4/8PcaGJhLMIAg5AUUyBDzcjSLEK3s1La8lCKCirfn5PZyOmZZkGVtn5KyvEDbUp92nk6n3cNFMU23mufkNS/lMcJKMW9YhooJghtQLn5+f/hjdrMB5TJ8Zwber+dZf8xaa77fz1pr3rPWrJlZSxMRARG1trU6R1dA1F4wbESKMGxEijBsRIq427vBuLg4ezdJpNwDDzyA559/3q5t2j1s69atQ1RUFLp06WLvpomUOHjwYKu0a/ewAcBzzz2HcePGtUbTRK2utY7O+JmNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRh4dty5YtsFgs2LRpk6NLsYsbN24gNTUV0dHRzW7j4MGD6NmzJ3Q6HTRNQ2BgIBYtWmTHKlsuOzsboaGh0DQNmqYhKCgIiYmJji7LqbXK/9maoi1dSe/EiROYNm0a9u3bh759+za7naioKBw7dgzDhg3DF198gePHj8PHx8eOlbZcbGwsYmNjER4ejt9++w0XLlxwdElOz+F7tuHDh6OkpAQjR450dCmoqKho9h4pNzcXL730EmbNmoV+/frZuTLHa8m6oZscHjZnsmbNGhQWFjbruX379kV2djYmTpwIT09PO1fmeC1ZN3STQ8P21VdfITg4GJqm4b333gMArFy5EmazGSaTCRs3bsTjjz8Ob29vdOnSBenp6dbn/vnPf4bBYEDHjh0xc+ZM3HnnnTAYDIiOjsbXX39tnW/OnDnw8PBAUFCQddy///u/w2w2Q9M0/PbbbwCAuXPnYt68eTh16hQ0TUN4eHirLPO2bdvg7e2NlJSUJj/X1dfN3r17ERkZCYvFAoPBgN69e+OLL74AADz99NPWz39hYWHIyckBAEybNg0mkwkWiwWff/45AKCmpgavvfYagoODYTQa0adPH2RmZgIA3nnnHZhMJnh5eaGwsBDz5s1D586dcfz48WbVbFdiZwAkMzOz0fP/8ssvAkBWrFhhHffKK68IAPnyyy+lpKRECgsLZfDgwWI2m6WystI634wZM8RsNsvRo0fl2rVr8uOPP8rAgQPFy8tLfv75Z+t8EydOlMDAQJt+lyxZIgDk0qVL1nGxsbESFhbWnMW2cf/990vfvn3rnbZ582bx8vKS119//bbt/OEPfxAAUlxcbB3nbOsmLCxMLBbLbZdFRGTt2rWycOFCKSoqksuXL0tUVJR06NDBpg83Nzf59ddfbZ43YcIE+fzzz62PX3jhBfH09JR169ZJcXGxvPzyy6LT6eTQoUM26+jZZ5+VFStWyJgxY+TYsWONqlFEZOzYsTJ27NhGz99IWU59GBkdHQ1vb28EBAQgISEBZWVl+Pnnn23mcXd3R8+ePeHp6YnIyEisXLkSV69exUcffeSgqm9t+PDhKC0txauvvtqidlxx3YwdOxZ/+tOf4OvrCz8/P4waNQqXL1/GpUuXAACzZs1CTU2NTX2lpaU4dOgQnnjiCQDAtWvXsHLlSowePRqxsbHw8fHBggULoNfr6yzX4sWL8cwzzyA7OxsRERHqFrQBTh223/Pw8AAAVFVV3XK+e++9FyaTCXl5eSrKcgquum70ej2Am4eFADB06FD06NED//mf/2k9S52RkYGEhAS4ubkBAI4fP47y8nLcc8891naMRiOCgoKcZrka4jJhawpPT0/ruyXZcuS6+dvf/oaHHnoIAQEB8PT0RFJSks10TdMwc+ZMnD59Gl9++SUA4L//+7/xb//2b9Z5ysrKAAALFiywfsbTNA1nz55FeXm5uoVphjYXtqqqKly5coUXia2H6nWzZ88epKamAgB+/vlnjB49GkFBQfj6669RUlKCt99+u85zpk6dCoPBgA8//BDHjx+Ht7c3QkJCrNMDAgIAAKmpqRARm+HAgQNKlqu5HP6ltr3t2rULIoKoqCjrOHd399seYrUHqtfN4cOHYTabAQDff/89qqqqMHv2bISGhgK4uSf7Z76+voiPj0dGRga8vLwwffp0m+ldu3aFwWDAd9991yo1tyaX37PduHEDxcXFqK6uxpEjRzB37lwEBwdj6tSp1nnCw8NRVFSEDRs2oKqqCpcuXcLZs2frtOXn54eCggLk5+fj6tWrrfIi3Lp1a7NP/TeVo9ZNVVUVLl68iF27dlnDFhwcDAD43//9X1y7dg0nTpyw+Rri92bNmoXr169j8+bNdX7sYDAYMG3aNKSnp2PlypUoLS1FTU0Nzp07h/Pnzzd1Fall7/ObaMKp/xUrVkhQUJAAEJPJJKNGjZL3339fTCaTAJC77rpLTp06JR988IF4e3sLAAkJCZGffvpJRG6e3tbr9dK5c2dxd3cXb29veeqpp+TUqVM2/Vy+fFkefvhhMRgM0r17d/mP//gPefHFFwWAhIeHW0+Ff/vttxISEiJGo1FiYmLkwoULjV7uAwcOyKBBg+TOO+8UAAJAgoKCJDo6Wnbv3m2db8uWLeLl5SWLFi1qsK2DBw9Kr169RKfTWdtJSUlxqnWzatUqCQsLsy5rQ8P69eutfSUnJ4ufn5/4+PhIXFycvPfeewJAwsLCbL6OEBH5l3/5F5k/f3696+f69euSnJwswcHB4u7uLgEBARIbGys//vijvP3222I0GgWAdO3aVT755JNGb8NarXXq3+Hfs7XEjBkzxM/PT0lfrsbV180TTzwhp0+fdkjf7fJ7tsaoPW1MdbnSuvn9YemRI0dgMBjQvXt3B1Zkfy4fttaSl5dnc2q5oSEhIcHRpbYJycnJOHHiBH766SdMmzYNb7zxhqNLsjuXDdvLL7+Mjz76CCUlJejevTvWrVtn1/YjIiLqnFqub8jIyLBrv/bQ2uumNZhMJkREROBf//VfsXDhQkRGRjq6JLvTROz7hzJN05CZmcn7s5HLqr0/29q1a+3Z7FqX3bMRuRqGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSJFWueBPamqqvX8xTaTMwYMHbS6KZC9237ONHTuWl5FrRQUFBdZr3lPriIqKwgMPPGD3du3+fzZqXVlZWYiPj29T97VrJ/h/NiJVGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkVa5Z7aZB+//vorRo4ciaqqKuu4srIy3HHHHejdu7fNvP369cMnn3yiukRqAobNiXXu3BnXrl3DsWPH6kz74YcfbB7Hx8erKouaiYeRTm7y5Mlwd7/9eyLD5vwYNic3YcIE1NTUNDhd0zT0798fd911l8KqqDkYNicXHByMgQMHQqerf1O5ublh8uTJiqui5mDYXMDkyZOhaVq902pqahAXF6e4ImoOhs0FjBs3rt7xbm5uePDBB9GpUyfFFVFzMGwuICAgAA899BDc3NzqTJs0aZIDKqLmYNhcxKRJkyAiNuN0Oh3GjBnjoIqoqRg2FzFmzBibrwDc3d3x+OOPw8fHx4FVUVMwbC7Cy8sLI0aMgF6vB3DzxEhiYqKDq6KmYNhcyMSJE1FdXQ0AMBgMGDFihIMroqZg2FzIE088AZPJBACIjY2F0Wh0cEXUFHV+B3Tu3Dns37/fEbVQIwwcOBC7du1C165dkZWV5ehyqAH1fl0j/yQzM1MAcODAoQVDPbIaPIwUEQ5OOFRXV+P11193eB0c6h8yMzMbihQ/s7kaNzc3zJ8/39FlUDMwbC6oMX+5IefDsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESni1GEbOHAg3Nzc0K9fP7u3/fTTT8PLywuapuG7775r8nxbtmyBxWLBpk2b7F5bU2RnZyM0NBSapjU4dOvWzS59cXu0jFOH7dChQ3j44Ydbpe0PP/wQf/nLX5o9n4i0RllNFhsbi9OnTyMsLAwWi8Xmf2/l5eW4ePGi9VIKLcXt0TIu8V+Nhi697UjDhw9HSUmJo8tokJubG4xGI4xGI3r06GHXtrk9msep92y1ai/fZm+NfdGoeHGJCNauXYsPPvjA7m1v2LDBru1xezSPXcJWU1OD1157DcHBwTAajejTp4/17+FpaWkwm83Q6XQYMGAAAgMDodfrYTab0b9/fwwePBhdu3aFwWCAj48PkpKS6rR/8uRJREREwGw2w2g0YvDgwfjqq68aXQNwc+UtWbIEd999Nzw9PWGxWPDiiy/W6asx83311VcIDg6Gpml47733AAArV66E2WyGyWTCxo0b8fjjj8Pb2xtdunRBenp6nVrffPNN3H333TAajfD390f37t3x5ptv2lwoZtu2bfD29kZKSkoTt0jDuD2avz1aTP5J7QV/muKFF14QT09PWbdunRQXF8vLL78sOp1ODh06JCIif/rTnwSAfP3111JWVia//fabDBs2TADI3/72N7l06ZKUlZXJnDlzBIB899131rYfeeQRCQ0NlTNnzkhVVZX88MMPcv/994vBYJCffvqp0TW88sorommaLFu2TIqLi6W8vFzef/99ASA5OTnWdho73y+//CIAZMWKFTbPBSBffvmllJSUSGFhoQwePFjMZrNUVlZa50tJSRE3NzfZuHGjlJeXy+HDhyUwMFAeeughm/W6efNm8fLyktdff/222yAsLEwsFovNuGeffVa+//77OvNyezRvezTGLfKT1eKwVVRUiMlkkoSEBOu48vJy8fT0lNmzZ4vIPzbu1atXrfN8/PHHAsDmxfDNN98IAMnIyLCOe+SRR6Rv3742fR45ckQAyAsvvNCoGsrLy8VkMsmjjz5q0056errNRmvsfCK33rgVFRXWcbUvjJMnT1rHDRw4UO677z6bPv74xz+KTqeT69evS3OEhYXVe5WnW4WN2+Mme26PW4WtxYeRx48fR3l5Oe655x7rOKPRiKCgIOTl5TX4PA8PDwCwXuEX+Mdngd/fsL0+vXv3hsViwZEjRxpVw8mTJ1FeXo5HHnnklu02dr6mqF3O3y/TtWvX6pw9q6mpgV6vr/dONY31+7ORIoJnn322yXVye9xkj+3xz1octrKyMgDAggULbL7bOXv2LMrLy1tcYEP0er11hd2uhnPnzgG4eeulW2nsfC31xBNP4PDhw9i4cSMqKirwf//3f9iwYQNGjBhh142blpZm84JvTdwet9fisNWuiNTU1DrX0Dtw4ECLC6xPdXU1ioqKEBwc3KgaDAYDAOD69eu3bLex87XUwoULMXToUEydOhXe3t4YM2YMxo0b16jvmZwRt0fjtDhstWeubvWtv73t3LkTN27cQP/+/RtVwz333AOdTofdu3ffst3GztdSP/74I06dOoVLly6hqqoKP//8M1auXAlfX99W6e/8+fOYNm1aq7QNcHs0VovDZjAYMG3aNKSnp2PlypUoLS1FTU0Nzp07h/Pnz9ujRlRWVqKkpATV1dX49ttvMWfOHISEhGDq1KmNqiEgIACxsbFYt24d1qxZg9LSUhw5cqTOdyiNna+lnnnmGQQHB+Pvf//7LefbunVri079iwgqKiqQnZ0Nb2/vZrVRn/a6PVqsCWdTGnT9+nVJTk6W4OBgcXd3l4CAAImNjZUff/xR0tLSxGQyCQDp1q2b7N27VxYvXiwWi0UASGBgoHz66aeSkZEhgYGBAkB8fX0lPT1dREQ++ugjefjhh6Vjx47i7u4uHTp0kPHjx8vZs2cbXYOIyNWrV+Xpp5+WDh06yB133CExMTHy2muvCQDp0qWL5ObmNnq+FStWSFBQkAAQk8kko0aNkvfff9+6nHfddZecOnVKPvjgA/H29hYAEhISYj01vmPHDunQoYPNWUO9Xi89e/aU7Oxs6zJt2bJFvLy8ZNGiRQ2u+/Xr1zd4JvL3w4IFC0REuD1asD0ao1VP/VPTvf/++zJ37lybcdevX5fnnntOPD09pby83EGVtU/23B63CptL/DayLblw4QLmzJlT5/OMh4cHgoODUVVVhaqqKt57TRGV28MlfhvZlhiNRuj1eqxZswYXL15EVVUVCgoK8OGHH+K1115DQkKCXT9f0a2p3B4Mm2IWiwXbt2/HDz/8gB49esBoNCIyMhIfffQRFi9ejI8//tjRJbYrKrcHDyMdYPDgwfif//kfR5dB/5+q7cE9G5EiDBuRIgwbkSIMG5EiDBuRIgwbkSIMG5EiDBuRIgwbkSIMG5EiDBuRIgwbkSIMG5EiDf7qPysrS2UdRG3Cra4o12DY4uPjW6UYovZKE3GBG1uRVVZWFuLj413ifmRkYy0/sxEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnS4G1+yfEuXryI//qv/7IZd+TIEQDA22+/bTPe19cXf/zjH1WVRs3A2/w6serqagQGBqKkpATu7v94XxQRaJpmfXz9+nVMnz4dH3zwgSPKpMbhbX6dmbu7OxISEqDT6XD9+nXrUFlZafMYACZMmODgaul2GDYnN378eFRVVd1ynoCAAAwePFhRRdRcDJuTGzRoEDp16tTgdA8PD0yePBlubm4Kq6LmYNicnKZpSExMhF6vr3d6ZWUlxo8fr7gqag6GzQXc6lAyJCQEAwYMUFwRNQfD5gL69euHu+66q854Dw8PTJ06VX1B1CwMm4uYPHlynUPJyspKxMfHO6giaiqGzUWMHz8e1dXV1seapqFPnz7o2bOnA6uipmDYXERYWBj69esHne7mJnN3d8fkyZMdXBU1BcPmQiZPnmwNW3V1NQ8hXQzD5kLi4+Nx48YNAMADDzyALl26OLgiagqGzYXceeed1l+KTJkyxcHVUFO1mR8iZ2Vl8bCqDWojL08AWNvm/mKTmZnp6BJaVVlZGT744AM899xzji6lVR04cABpaWmOLsOu2lzYxo0b5+gSWt2jjz7aLj6vtbWw8TObC2oPQWuLGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRRg2IkUYNiJFGDYiRdpt2JYuXYqOHTtC0zSsXr3a0eU0KDs7G6GhodA0DZqmISgoCImJibd9Xm5uLhISEtC9e3d4enrC398fffv2xaJFi6zzJCQkWNu93bB58+Y6tbz66qu3rGH58uXQNA06nQ4RERHYs2dPi9eHK2u3YXvhhRewf/9+R5dxW7GxsTh9+jTCwsJgsVhw4cIF/PWvf73lc77//ntER0cjKCgIO3fuRElJCfbv349hw4Zh165dNvNu374dV65cQVVVFc6fPw8AGDVqFCorK1FWVobCwkJMnz69Ti0A8OGHHzZ4peaamhr8+c9/BgAMHToUeXl5GDJkSEtWhctrt2FrjoqKCkRHRzu6jNtaunQpfHx8kJaWhm7dusFgMKBHjx544403YDQarfNpmoZBgwbBYrHY3P9N0zTo9XqYTCYEBATUe3nzAQMG4MKFC9iwYUO9NWRnZ6Nz5872XzgXxrA1wZo1a1BYWOjoMm7r8uXLKCkpQVFRkc14Dw8PbNq0yfo4PT0dJpPptu3NmDEDI0aMsBk3e/ZsAMCqVavqfc7y5csxb968ppbepjFs/2T37t247777YDKZ4O3tjd69e6O0tBRz587FvHnzcOrUKWiahvDwcKSlpcFsNkOn02HAgAEIDAyEXq+H2WxG//79MXjwYHTt2hUGgwE+Pj5ISkqy6Wvbtm3w9vZGSkqKXZdh4MCBKCsrw9ChQ7Fv3z67tl1r6NCh6NmzJ3bu3Injx4/bTNu3bx/Ky8vx2GOPtUrfroph+52ysjKMGjUKY8eORVFREU6cOIEePXqgsrISaWlpGDlyJMLCwiAiOHnyJObOnYsXX3wRIoJVq1bhzJkzuHDhAoYMGYKcnBzMnz8fOTk5KCoqwpQpU7BkyRLk5uZa+6upqQEA67Ug7SUpKQn33nsvcnNzERMTg169euGdd96ps6drqZkzZwJAnRNMy5Ytw/PPP2/XvtoChu138vPzUVpail69esFgMCAwMBDZ2dnw9/e/7XMjIyNhMpnQoUMH6/3SgoOD4e/vD5PJZD2DmJeXZ33O8OHDUVpaetuzek1lNBqxf/9+vPvuu4iIiMDRo0eRnJyMnj17Yvfu3XbrZ8qUKTCbzfj4449RUVEBADh9+jQOHTrE2w7Xg2H7ndDQUHTs2BGJiYlYuHAh8vPzm9WOh4cHANjcCKP2DjS3u2Wvvej1esyZMwfHjh3DwYMH8dRTT6GwsBBxcXEoLi62Sx8WiwUTJkxAcXExMjIyAACpqamYPXu2dR3QPzBsv2M0GrFjxw7ExMQgJSUFoaGhSEhIsL5ru6r7778fn332GWbNmoVLly5h586ddmu79kTJ6tWrceXKFaxdu9Z6eEm2GLZ/0qtXL2zatAkFBQVITk5GZmYmli5d6uiybmnPnj1ITU21Po6NjbXZq9aaNGkSAKC8vNxufffr1w9RUVH45ptvMGPGDMTFxcHX19du7bclDNvvFBQU4OjRowCAgIAAvPXWW+jfv791nLM6fPgwzGaz9fH169frrbn2rGGfPn3s2n/t3m3dunVt/krNLcGw/U5BQQFmzpyJvLw8VFZWIicnB2fPnkVUVBQAwM/PDwUFBcjPz8fVq1db/Plr69atLTr1X1VVhYsXL2LXrl02YQOA0aNHIysrC1euXEFJSQk2btyIl156CU8++aTdwzZu3Dj4+/tj9OjRCA0NtWvbbYq0EZmZmdKUxVm2bJkEBgYKADGbzTJmzBjJz8+X6Oho8fX1FTc3N+nUqZO88sorUl1dLSIi3377rYSEhIjRaJSYmBiZP3++mEwmASDdunWTvXv3yuLFi8VisQgACQwMlE8//VQyMjKsffn6+kp6erqIiGzZskW8vLxk0aJFDda5fv16CQsLEwC3HNavX299zvbt2yU+Pl7CwsLE09NTPDw85O6775aFCxfKtWvX6vRRWloqQ4YMET8/PwEgOp1OwsPDJSUlpcFa/P395ZlnnrFOS0pKkv3791sfL1iwQIKCgqztRUZGyt69exu9fZq6PV1AVpu7i00bWZx2rw1uz7U8jCRShGEjUoRhI1KEYSNShGEjUoRhI1KEYSNShGEjUoRhI1KEYSNShGEjUoRhI1KEYSNShGEjUoRhI1KEYSNShGEjUsT99rO4Fk3THF0CUb3aTNiio6ORmZnp6DJa3YEDB5CWltYulrWtaTPXIGkv2uC1OdoLXoOESBWGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUgRho1IEYaNSBGGjUiRNnPn0baooqIC58+ftxl38eJFAMDp06dtxru5uSEkJERZbdR0vPOoE7t8+TKCgoJQXV1923mHDRuGrVu3KqiKmol3HnVmHTp0wKOPPgqd7tabSdM0JCQkKKqKmothc3KJiYm3vX+2u7s7nnrqKUUVUXMxbE7uySefhKenZ4PT3d3dMWrUKFgsFoVVUXMwbE7ObDbjySefhF6vr3d6TU0NJk6cqLgqag6GzQVMnDgRVVVV9U4zGo14/PHHFVdEzcGwuYBhw4bB29u7zni9Xo/4+HgYDAYHVEVNxbC5AL1ej3HjxtU5lKyqqsKECRMcVBU1FcPmIiZMmFDnULJDhw54+OGHHVQRNRXD5iIefPBBdOzY0frYw8MDiYmJcHNzc2BV1BQMm4vQ6XRITEyEh4cHAKCyshLjx493cFXUFAybCxk/fjwqKysBAF26dMF9993n4IqoKRg2F3Lvvfeie/fuAICpU6dC0zQHV0RN4fS/+j9w4ACWL1/u6DKchtFoBAB88803iIuLc3A1zmPt2rWOLuG2nH7P9ssvv2DdunWOLsNpdO3aFRaLpd7v3dqjc+fOuczrw+n3bLVc4Z1LlS+++AJ/+MMfHF2GU8jKykJ8fLyjy2gUp9+zUV0Mmmti2IgUYdiIFGHYiBRh2IgUYdiIFGHYiBRh2IgUYdiIFGHYiBRh2IgUYdiIFGHYiBRh2IgUaXNhW7p0KTp27AhN07B69WpHl9MoN27cQGpqKqKjo5vdRnZ2NkJDQ6FpGjRNQ1BQEBITE2/7vNzcXCQkJKB79+7w9PSEv78/+vbti0WLFlnnSUhIsLZ7u2Hz5s11ann11VdvWcPy5cuhaRp0Oh0iIiKwZ8+eZq8HpyZOLjMzU5pa5okTJwSArFq1qpWqsp+ffvpJBg0aJACkb9++LW4vLCxMLBZLo+Y9cuSImEwmefbZZ+XMmTNSUVEhx48fl6SkJHnkkUes88XHx8v27dvlypUrUlVVJefPnxcAMmrUKKmsrJSysjIpLCyU6dOny6ZNm2xqASBBQUFSWVlZbw3V1dUSEhIiAGz6bKzmvD4cJKvN7dmao6KiokV7lebKzc3FSy+9hFmzZqFfv37K+1+6dCl8fHyQlpaGbt26wWAwoEePHnjjjTesl18Abt6SatCgQbBYLHB3d7cZr9frYTKZEBAQgAEDBtTpY8CAAbhw4QI2bNhQbw3Z2dno3Lmz/RfOCTFsANasWYPCwkLl/fbt2xfZ2dmYOHHiLe9U01ouX76MkpISFBUV2Yz38PDApk2brI/T09NhMplu296MGTMwYsQIm3GzZwcGj/cAAA8OSURBVM8GAKxatare5yxfvhzz5s1raukuqd2Ebffu3bjvvvtgMpng7e2N3r17o7S0FHPnzsW8efNw6tQpaJqG8PBwpKWlwWw2Q6fTYcCAAQgMDIRer4fZbEb//v0xePBgdO3aFQaDAT4+PkhKSmrV2rdt2wZvb2+kpKTYtd2BAweirKwMQ4cOxb59++zadq2hQ4eiZ8+e2LlzJ44fP24zbd++fSgvL8djjz3WKn07m3YRtrKyMowaNQpjx45FUVERTpw4gR49eqCyshJpaWkYOXIkwsLCICI4efIk5s6dixdffBEiglWrVuHMmTO4cOEChgwZgpycHMyfPx85OTkoKirClClTsGTJEuTm5rZa/TU1NQBunkixp6SkJNx7773Izc1FTEwMevXqhXfeeafOnq6lZs6cCQB1TlgtW7YMzz//vF37cmbtImz5+fkoLS1Fr169YDAYEBgYiOzsbPj7+9/2uZGRkTCZTOjQoYP1CsTBwcHw9/eHyWSynvHLy8trtfqHDx+O0tLS257Vayqj0Yj9+/fj3XffRUREBI4ePYrk5GT07NkTu3fvtls/U6ZMgdlsxscff4yKigoAwOnTp3Ho0KF2dWOQdhG20NBQdOzYEYmJiVi4cCHy8/Ob1U7tpb9/f0P52jvLNHT/NGen1+sxZ84cHDt2DAcPHsRTTz2FwsJCxMXFobi42C59WCwWTJgwAcXFxcjIyAAApKamYvbs2dZ12h60i7AZjUbs2LEDMTExSElJQWhoKBISEqzvsnTT/fffj88++wyzZs3CpUuXsHPnTru1XXuiZPXq1bhy5QrWrl1rPbxsL9pF2ACgV69e2LRpEwoKCpCcnIzMzEwsXbrU0WUptWfPHqSmplofx8bG2uyla02aNAkAUF5ebre++/Xrh6ioKHzzzTeYMWMG4uLi4Ovra7f2XUG7CFtBQQGOHj0KAAgICMBbb72F/v37W8e1F4cPH4bZbLY+vn79er3roPasYZ8+fezaf+3ebd26dXjuuefs2rYraDdhmzlzJvLy8lBZWYmcnBycPXsWUVFRAAA/Pz8UFBQgPz8fV69edbrPX1u3bm3Rqf+qqipcvHgRu3btsgkbAIwePRpZWVm4cuUKSkpKsHHjRrz00kt48skn7R62cePGwd/fH6NHj0ZoaKhd23YJjv4Ny+009ec4y5Ytk8DAQAEgZrNZxowZI/n5+RIdHS2+vr7i5uYmnTp1kldeeUWqq6tFROTbb7+VkJAQMRqNEhMTI/PnzxeTySQApFu3brJ3715ZvHixWCwWASCBgYHy6aefSkZGhrUvX19fSU9Pb9KyHThwQAYNGiR33nmnALD+tCk6Olp2795tnW/Lli3i5eUlixYtarCt9evXW38edath/fr11uds375d4uPjJSwsTDw9PcXDw0PuvvtuWbhwoVy7dq1OH6WlpTJkyBDx8/MTAKLT6SQ8PFxSUlIarMXf31+eeeYZ67SkpCTZv3+/9fGCBQskKCjI2l5kZKTs3bu30evQlX6upYmIKM53k9Rey93JyyQHcaHXx9p2cRhJ5AwYNjvKy8tr1N9QEhISHF0qOYDL3DLKFURERLjC4Qw5CPdsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKuMxfbOLi4hxdAjmhc+fOObqERnP6PVvXrl0xduxYR5fhNAoKCvD55587ugyn0aVLF5d5fTj9NUjIlgtdc4Ns8RokRKowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEirjMPbXbo19//RUjR45EVVWVdVxZWRnuuOMO9O7d22befv364ZNPPlFdIjUBw+bEOnfujGvXruHYsWN1pv3www82j+Pj41WVRc3Ew0gnN3nyZLi73/49kWFzfgybk5swYQJqamoanK5pGvr374+77rpLYVXUHAybkwsODsbAgQOh09W/qdzc3DB58mTFVVFzMGwuYPLkydA0rd5pNTU1iIuLU1wRNQfD5gLGjRtX73g3Nzc8+OCD6NSpk+KKqDkYNhcQEBCAhx56CG5ubnWmTZo0yQEVUXMwbC5i0qRJEBGbcTqdDmPGjHFQRdRUDJuLGDNmjM1XAO7u7nj88cfh4+PjwKqoKRg2F+Hl5YURI0ZAr9cDuHliJDEx0cFVUVMwbC5k4sSJqK6uBgAYDAaMGDHCwRVRUzBsLuSJJ56AyWQCAMTGxsJoNDq4ImoKp/9t5Llz57B//35Hl+E0Bg4ciF27dqFr167IyspydDlOo6GvR5yJJv98isvJZGVl8Xd/dFtO/jIGgLUucxgpIhxEUF1djddff93hdTjLkJmZ6eiXZqO5TNjoJjc3N8yfP9/RZVAzMGwuqDF/uSHnw7ARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEpwrARKcKwESnCsBEp0ubCtnTpUnTs2BGapmH16tWOLueWXn/9dURGRsLb2xuenp4IDw9HUlIS/v73vze5rezsbISGhkLTNGiahqCgoEZdoyQ3NxcJCQno3r07PD094e/vj759+2LRokXWeRISEqzt3m7YvHlznVpeffXVW9awfPlyaJoGnU6HiIgI7Nmzp8nL7xLEyWVmZkpTyzxx4oQAkFWrVrVSVfbx4IMPyvvvvy+XL1+W0tJSyczMFL1eL8OGDWt2m2FhYWKxWBo175EjR8RkMsmzzz4rZ86ckYqKCjl+/LgkJSXJI488Yp0vPj5etm/fLleuXJGqqio5f/68AJBRo0ZJZWWllJWVSWFhoUyfPl02bdpkUwsACQoKksrKynprqK6ulpCQEAFg02djNef14SBZbW7P1hwVFRWIjo5W3u8dd9yBGTNmwM/PD15eXhg3bhxGjx6Nbdu24Zdffmn1/pcuXQofHx+kpaWhW7duMBgM6NGjB9544w2b65tomoZBgwbBYrHY/L1H0zTo9XqYTCYEBARgwIABdfoYMGAALly4gA0bNtRbQ3Z2Njp37mz/hXNCDBuANWvWoLCwUHm/mzdvrnOVY39/fwBAeXl5q/d/+fJllJSUoKioyGa8h4cHNm3aZH2cnp5uvdDQrcyYMaPOFb9mz54NAFi1alW9z1m+fDnmzZvX1NJdUrsJ2+7du3HffffBZDLB29sbvXv3RmlpKebOnYt58+bh1KlT0DQN4eHhSEtLg9lshk6nw4ABAxAYGAi9Xg+z2Yz+/ftj8ODB6Nq1KwwGA3x8fJCUlGS3On/99VcYjUZ0797dOm7btm3w9vZGSkqK3foBbl48qKysDEOHDsW+ffvs2natoUOHomfPnti5cyeOHz9uM23fvn0oLy/HY4891ip9O5t2EbaysjKMGjUKY8eORVFREU6cOIEePXqgsrISaWlpGDlyJMLCwiAiOHnyJObOnYsXX3wRIoJVq1bhzJkzuHDhAoYMGYKcnBzMnz8fOTk5KCoqwpQpU7BkyRLk5ua2uM7y8nLs2LED06dPh4eHh3V87f3Zbty40eI+fi8pKQn33nsvcnNzERMTg169euGdd96ps6drqZkzZwJAnRNWy5Ytw/PPP2/XvpxZuwhbfn4+SktL0atXLxgMBgQGBiI7O9t6yHYrkZGRMJlM6NChA8aPHw/g5j3T/P39YTKZrGf88vLyWlznm2++iTvvvNPmTCAADB8+HKWlpbc9q9dURqMR+/fvx7vvvouIiAgcPXoUycnJ6NmzJ3bv3m23fqZMmQKz2YyPP/4YFRUVAIDTp0/j0KFDmDBhgt36cXbtImyhoaHo2LEjEhMTsXDhQuTn5zerndq9Te1ViQFYLwf++5vMN8f69euRlZWFL774Al5eXi1qqyn0ej3mzJmDY8eO4eDBg3jqqadQWFiIuLg4FBcX26UPi8WCCRMmoLi4GBkZGQCA1NRUzJ4922YP3ta1i7AZjUbs2LEDMTExSElJQWhoKBISEqzvso6WkZGBxYsXY9euXejWrZvD6rj//vvx2WefYdasWbh06RJ27txpt7ZrT5SsXr0aV65cwdq1a62Hl+1FuwgbAPTq1QubNm1CQUEBkpOTkZmZiaVLlzq6LKxYsQJ//etfsWPHjla/qeGePXuQmppqfRwbG2uzl65Ve883e54R7devH6KiovDNN99gxowZiIuLg6+vr93adwXtImwFBQU4evQogJs3FnzrrbfQv39/6zhHEBEkJyfj+++/x4YNG3DHHXe0ep+HDx+G2Wy2Pr5+/Xq966D2rGGfPn3s2n/t3m3dunV47rnn7Nq2K2g3YZs5cyby8vJQWVmJnJwcnD17FlFRUQAAPz8/FBQUID8/H1evXm3x56/GOHr0KN555x385S9/gV6vr/Ozp9/vdbdu3dqiU/9VVVW4ePEidu3aZRM2ABg9ejSysrJw5coVlJSUYOPGjXjppZfw5JNP2j1s48aNg7+/P0aPHo3Q0FC7tu0SHPwTlttq6s9xli1bJoGBgQJAzGazjBkzRvLz8yU6Olp8fX3Fzc1NOnXqJK+88opUV1eLiMi3334rISEhYjQaJSYmRubPny8mk0kASLdu3WTv3r2yePFisVgsAkACAwPl008/lYyMDGtfvr6+kp6e3ug6v//+ewHQ4LBkyRLrvFu2bBEvLy9ZtGhRg+2tX7/e+vOoWw3r16+3Pmf79u0SHx8vYWFh4unpKR4eHnL33XfLwoUL5dq1a3X6KC0tlSFDhoifn58AEJ1OJ+Hh4ZKSktJgLf7+/vLMM89YpyUlJcn+/futjxcsWCBBQUHW9iIjI2Xv3r2NXo+u9HMtl7mxhpOXSQ7iQq8P17mxBpGrY9jsKC8vr1F/Q0lISHB0qeQAvEODHUVERLjC4Qw5CPdsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKMGxEijBsRIowbESKuMxfbLKyshxdAjmhAwcOOLqERnOZsMXHxzu6BKIWcfprkBC1EbwGCZEqDBuRIgwbkSIMG5Ei/w+o/WlO9b/YnQAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 24
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 661
        },
        "id": "t24eY1LA8EvD",
        "outputId": "aad25492-5067-4efb-cd7e-a9424acbe120"
      },
      "source": [
        "plot_model(gen_decoder)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 25
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "EcHYS9lQ8Fwu",
        "outputId": "df3e556d-dff4-489d-e741-4f162a2cf161"
      },
      "source": [
        "def generate_from_input(encoder_input):\n",
        "    values,h,c = gen_encoder.predict(encoder_input)\n",
        "    \n",
        "    decoder_seq = np.zeros((1,1))\n",
        "    decoder_seq[0,0] = a_tok.word_index['sos']\n",
        "    \n",
        "    generated_sent = ''\n",
        "    stop_condition= False\n",
        "    \n",
        "    while not stop_condition:\n",
        "        output,h_state,c_state = gen_decoder.predict([decoder_seq]+[values,h,c])\n",
        "        sampled_index = np.argmax(output[0,-1,:])\n",
        "        sampled_word = a_tok.index_word[sampled_index]\n",
        "        \n",
        "        if sampled_word != 'eos':\n",
        "            generated_sent = generated_sent + sampled_word + ' '\n",
        "        \n",
        "        if sampled_word == 'eos' or len(generated_sent) >= answer_sequence_size:\n",
        "            stop_condition=True\n",
        "        \n",
        "        decoder_seq = np.zeros((1,1))\n",
        "        decoder_seq[0,0] = sampled_index\n",
        "        h,c = h_state,c_state\n",
        "        \n",
        "    return generated_sent.strip()\n",
        " \n",
        "print(generate_from_input(train_encoder_input[24].reshape(1,question_sequence_size,-1)))\n",
        "print(generate_from_input(train_encoder_input[1021].reshape(1,question_sequence_size,-1)))\n",
        "print(generate_from_input(train_encoder_input[3001].reshape(1,question_sequence_size,-1)))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "mình học trường đại học tôn đức thắng eo ngành tớ thuộc khoá 15 eo eo eo eo cũng đặt trình hay chuyên trình chất lượng cao và truyền thông dữ liệu eo eo eo khác\n",
            "mình sinh ra ở thành phố hồ chí minh eo tài năng và trơn tru các vùng miền miền vin rẻ eo eo do bán trình ở bình edu eo do mình eo trình gì cũng có trình trung\n",
            "tui cũng không biết nữa nhưng thú thật đi chỗ nào cũng thấy eo ốc eo cái đã đi eo đỉnh eo mới bán eo cũng được thì thú lắm eo địa chỉ cần phải đọc hợp với chăn\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zTiUJS4YCHHG"
      },
      "source": [
        "def seq2question(encoder_input):\n",
        "    ret=[q_tok.index_word[idx] for idx in encoder_input if idx != 0]\n",
        "    return ' '.join(ret)\n",
        "\n",
        "def seq2answer(decoder_input):\n",
        "    ret = []\n",
        "    for idx in decoder_input:\n",
        "        if idx != 0:\n",
        "            if a_tok.index_word[idx] != 'sos' and a_tok.index_word[idx] != 'eos':\n",
        "                ret.append(a_tok.index_word[idx])\n",
        "                \n",
        "    return ' '.join(ret)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_HW8MUpLGcMr"
      },
      "source": [
        "## 3.Results on train dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "OnN1ND198KkV",
        "outputId": "213e7d4c-bf45-4a34-b980-51662a4d4795"
      },
      "source": [
        "sample_indices= [5,26,7,11,735,662,115,31,63]\n",
        "\n",
        "for idx in sample_indices:\n",
        "    generated_sent = generate_from_input(train_encoder_input[idx:idx+1])\n",
        "    print(f\"Question >> {seq2question(train_encoder_input[idx])}\")\n",
        "    print(f\"Answer(Generated) >> {generated_sent}\")\n",
        "    print(f\"Answer(Label) >> {seq2answer(train_decoder_input[idx])}\")\n",
        "    print(\"=\"*45)\n",
        "    print(\"\\n\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Question >> bạn nghĩ mình sẽ sống độc thân đến cuối đời\n",
            "Answer(Generated) >> không đâu mình nghĩ bạn sẽ tìm được ai đó thôi hiện tại không thể nhưng tương lai có thể eo eo một việc của bạn eo của người đó nếu muốn rất người yêu mình eo\n",
            "Answer(Label) >> không đâu mình nghĩ bạn sẽ tìm được ai đó thôi hiện tại không thể nhưng tương lai có thể\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> quê bạn có nhiều địa điểm du lịch không\n",
            "Answer(Generated) >> quê tớ có nhiều lắm ví dụ như là quảng trường hồ chí minh quê bác núi quyết eo eo eo có có đặc sản khác nhau là một số khác nhau là nơi có có chỗ mua lớn ở đây\n",
            "Answer(Label) >> quê tớ có nhiều lắm ví dụ như là quảng trường hồ chí minh quê bác núi quyết\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> chỗ cafe ưa thích mà bạn hay lui đến là ở đâu\n",
            "Answer(Generated) >> thường thì mình thích ngồi những quán lề đường để cảm nhận nhịp sống của sài gòn eo sài gòn eo sài gòn eo sài gòn eo ngắm bạn nếu rất dễ thương và sài gòn đã\n",
            "Answer(Label) >> thường thì mình thích ngồi những quán lề đường để cảm nhận nhịp sống của sài gòn\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> bạn sinh ngày nào\n",
            "Answer(Generated) >> mình sinh ngày 1 tháng 9 eo eo eo chi số gì eo eo chi eo eo eo eo eo eo mình eo eo du khách của du khách eo sinh trường eo truyền thông eo eo eo tỉnh thành eo\n",
            "Answer(Label) >> \n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> gu thời trang của bạn là gì vậy\n",
            "Answer(Generated) >> mình hay mặc đồ theo kiểu đường phố vì nó làm mình được thoải mái eo sợ tao eo eo đó hết eo eo eo nào eo eo gì eo eo eo hôm tới eo eo eo mới nữa eo eo gì eo\n",
            "Answer(Label) >> mình hay mặc đồ theo kiểu đường phố vì nó làm mình được thoải mái\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> khi nào cậu đi du lịch\n",
            "Answer(Generated) >> sau tết rồi đi eo này eo gần quán cà phê eo hay đi nữa eo tiếp tắm biển eo eo gần nhà mình eo eo về nhà khi đi du lịch về tắm đi eo eo hay sao eo giải tỏa eo\n",
            "Answer(Label) >> sau tết rồi đi\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> bạn thuê nhà như thế nào thế\n",
            "Answer(Generated) >> trải nghiệm của mình là mình ký hợp đồng với chủ nhà và đặt cọc nhớ chú ý các điều khoản trong hợp đồng tránh rắc rối nhé eo eo rộng rãi nhé eo eo luôn là con\n",
            "Answer(Label) >> trải nghiệm của mình là mình ký hợp đồng với chủ nhà và đặt cọc nhớ chú ý các điều khoản trong hợp đồng tránh rắc rối nhé\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> bây giờ em đặt ship đến đồng nai thì bao lâu đến ạ\n",
            "Answer(Generated) >> khoảng 2 3 ngày là tới nha eo eo khoảng 30 triệu eo eo khoảng khoảng 2 á eo eo eo khoảng khoảng 2 triệu eo eo khoảng khoảng 3 á eo eo eo khoảng khoảng 2 triệu\n",
            "Answer(Label) >> khoảng 2 3 ngày là tới nha\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> shop ơi mẫu laptop asus rog strix g 531 core i7 10300h còn không ạ\n",
            "Answer(Generated) >> chào bạn hiện tại dòng máy với option đó vẫn còn bạn nhé eo thì vẫn còn bạn lên cấp trên để còn thêm màn hình hạng thì em phải được cấp trên đó chọn hàng thì\n",
            "Answer(Label) >> chào bạn hiện tại dòng máy với option đó vẫn còn bạn nhé\n",
            "=============================================\n",
            "\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "MxRUgvHTGg1h"
      },
      "source": [
        "## 4.Results on test dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lDSa7bUn8LbS",
        "outputId": "49c270ff-b1ea-48bf-82b5-1932cd02376a"
      },
      "source": [
        "sample_indices= [159,50,62,31,57]\n",
        "\n",
        "for idx in sample_indices:\n",
        "    generated_sent = generate_from_input(test_encoder_input[idx:idx+1])\n",
        "    print(f\"Question >> {seq2question(test_encoder_input[idx])}\")\n",
        "    print(f\"Answer(Generated) >> {generated_sent}\")\n",
        "    print(f\"Answer(Label) >> {seq2answer(test_decoder_input[idx])}\")\n",
        "    print(\"=\"*45)\n",
        "    print(\"\\n\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Question >> bạn có thường xuyên đi chơi bên ngoài với bạn bè mình không\n",
            "Answer(Generated) >> có mình thường đến các show rap eo ngoài eo sạch là đi eo eo đi chơi eo đi chơi eo đi chơi eo đi học eo đi học về cntt eo eo mây mây mây nữa eo eo eo mây nữa\n",
            "Answer(Label) >> có mình thường xuyên đi chơi với bạn bè mình\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> mày tên gì\n",
            "Answer(Generated) >> quê mình ở tiền giang eo eo liên khai á eo eo eo số gì eo eo eo gì eo eo eo gì eo eo eo gì eo eo eo gì eo eo eo gì eo eo eo gì eo eo eo nữa eo eo eo nữa eo\n",
            "Answer(Label) >> vũ\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> hiện bạn cảm thấy như thế nào\n",
            "Answer(Generated) >> không bạn bè mình cực kỳ được yêu cùng lớp eo eo eo đi không eo hay đi eo eo đi học eo và đam mê của mình cũng rất vui tính eo rất vui với bạn eo eo không eo\n",
            "Answer(Label) >> bình thường\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> bạn có quần này màu không\n",
            "Answer(Generated) >> dạ cũng được con mới vào cũng không biết hay về sinh vào là noel eo eo eo eo sao không eo nữa eo eo eo sao eo eo eo sao eo eo eo sao eo eo eo sao eo eo eo sao\n",
            "Answer(Label) >> không mình chỉ có màu xanh thôi\n",
            "=============================================\n",
            "\n",
            "\n",
            "Question >> cậu thích đến nước nào\n",
            "Answer(Generated) >> tớ thích xem phim hành động và đặc biệt là như bánh canh quận 2 và yêu eo hàng eo eo này thì là một người con trai cùng nhau hò hét rất hiện đại eo eo và khi\n",
            "Answer(Label) >> mình thích nước anh vì có tháp\n",
            "=============================================\n",
            "\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}