4_cohort_customers_model.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": " cohort_customers_model.ipynb",
      "provenance": [],
      "toc_visible": true,
      "authorship_tag": "ABX9TyN3eSmz1jW8gfLhrjT+86yF",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/dlskawns/cp1/blob/main/4_cohort_customers_model.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# 4. 문맥을 통한 고객 집단 분류 모델  \n"
      ],
      "metadata": {
        "id": "9lrjEh21-OD7"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Transformer - Bert(bert-base-uncased)를 이용 \n",
        "반려동물용 상품을 판매하는 것이 파악되어 반려동물 유무(0 - 없음, 1 - 강아지, 2 - 고양이, 3 - 강아지와 고양이)로 고객 세분화 진행"
      ],
      "metadata": {
        "id": "fhvULUN8hIeh"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "### * 선정 이유:  \n",
        "label(반려동물 유무 여부)을 설정한 뒤, 전체 문맥에 따른 특징적인 패턴을 찾아 전체 데이터셋에 자동으로 label을 지정하기 위함.  \n",
        "ㄴ> 이를 통해 궁극적으로 고객의 집단은 분류하는 feature를 생성.\n"
      ],
      "metadata": {
        "id": "GGVKFIPIgmDj"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "### * 가설:  \n",
        "BERT를 이용해 문맥데이터와 label을 학습하면 문장들의 패턴을 파악해 반려동물 유무에 대한 고객 세분화가 가능할 것이다.\n"
      ],
      "metadata": {
        "id": "pubBJ8glgors"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "### * 방법:  \n",
        "  * 리뷰 내에서 dog, cat을 파악할 수 있는 키워드를 통해 일차적으로 label 생성\n",
        "  * label 분포 파악 후 imbalanced 할 경우, under sampling 및 SMOTE 진행 후 성능 파악\n",
        "  * 샘플링 방법 선정 후(최종-under sampling) BERT 다중분류 모델 학습 진행 \n",
        "  * 학습 완료된 모델을 통해 자동 라벨링 진행\n",
        "\n",
        "  <br>\n",
        "\n",
        "  ---"
      ],
      "metadata": {
        "id": "8kTcPBOzgru8"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "### 반려동물(dog, cat)을 키우는 고객 여부 파악. 1차 키워드 분류\n",
        "\n",
        "dog과 cat의 여부를 파악할 수 있는 words를 포함하고 있다면, 각각의 label을 부여한다.  \n",
        "(No: 0, dog: 1, cat: 2, both: 3)"
      ],
      "metadata": {
        "id": "p2rt9ibo7u82"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# Text에서 고객 파악\n",
        "def sorting_labels(df):\n",
        "  labels = []\n",
        "  for i in df['Text']:\n",
        "    i = i.lower()\n",
        "    if ('my dog' in i) and ('my cat' in i):\n",
        "      labels.append(3)\n",
        "    elif ('my dog' in i):\n",
        "      labels.append(1)\n",
        "    elif ('my cat' in i):\n",
        "      labels.append(2)\n",
        "    elif (('tasty' in i) or  ('taste good' in i) or ('yummy' in i)) and (('dog' not in i) or ('cat' not in i) or ('puppy' not in i)):\n",
        "      labels.append(0)\n",
        "    else:\n",
        "      labels.append(None)\n",
        "  df['labels'] = labels\n",
        "  return df"
      ],
      "metadata": {
        "id": "VC-5SHSwsD5i"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# summary에서 고객파악\n",
        "def sorting_hints(df):\n",
        "  labels = []\n",
        "  for i, k in zip(df['Summary'], range(len(df['labels']))):\n",
        "    i = i.lower()\n",
        "    if pd.isna(df['labels'][k])==True:\n",
        "      if ('dog' in i) and ('cat' in i):\n",
        "        df['labels'][k] = 3\n",
        "      elif ('dog' in i ):\n",
        "        df['labels'][k] = 1\n",
        "      elif ('cat' in i ):\n",
        "        df['labels'][k] = 2\n",
        "      elif (('tasty' in i ) or  ('taste good' in i ) or ('yummy' in i )) and (('dog' not in i ) or ('cat' not in i ) or ('puppy' not in i )):\n",
        "        df['labels'][k] = 0\n",
        "    else:\n",
        "      continue\n",
        "  return df"
      ],
      "metadata": {
        "id": "VlqCD_XOsD3P"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "fxGfch2_dVUq",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "baf88eaa-5129-426d-b5c6-28353e16dde2"
      },
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "\n",
              "  <div id=\"df-685a80db-41fb-4692-ae8e-b0d4c8b3ede2\">\n",
              "    <div class=\"colab-df-container\">\n",
              "      <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>Id</th>\n",
              "      <th>ProductId</th>\n",
              "      <th>UserId</th>\n",
              "      <th>ProfileName</th>\n",
              "      <th>HelpfulnessNumerator</th>\n",
              "      <th>HelpfulnessDenominator</th>\n",
              "      <th>Score</th>\n",
              "      <th>Time</th>\n",
              "      <th>Summary</th>\n",
              "      <th>Text</th>\n",
              "      <th>date</th>\n",
              "      <th>dayofweek</th>\n",
              "      <th>r_counts</th>\n",
              "      <th>keyword</th>\n",
              "      <th>labels</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>1</td>\n",
              "      <td>B001E4KFG0</td>\n",
              "      <td>A3SGXH7AUHU8GW</td>\n",
              "      <td>delmartian</td>\n",
              "      <td>1</td>\n",
              "      <td>1</td>\n",
              "      <td>5</td>\n",
              "      <td>1303862400</td>\n",
              "      <td>Good Quality Dog Food</td>\n",
              "      <td>I have bought several of the Vitality canned d...</td>\n",
              "      <td>2011-04-27</td>\n",
              "      <td>Wednesday</td>\n",
              "      <td>1</td>\n",
              "      <td>['vitality', 'finicky', 'smells', 'labrador', ...</td>\n",
              "      <td>1.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>2</td>\n",
              "      <td>B00813GRG4</td>\n",
              "      <td>A1D87F6ZCVE5NK</td>\n",
              "      <td>dll pa</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>1</td>\n",
              "      <td>1346976000</td>\n",
              "      <td>Not as Advertised</td>\n",
              "      <td>Product arrived labeled as Jumbo Salted Peanut...</td>\n",
              "      <td>2012-09-07</td>\n",
              "      <td>Friday</td>\n",
              "      <td>1</td>\n",
              "      <td>['jumbo', 'unsalted', 'small', 'sized', 'vendo...</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>3</td>\n",
              "      <td>B000LQOCH0</td>\n",
              "      <td>ABXLMWJIXXAIN</td>\n",
              "      <td>Natalia Corres \"Natalia Corres\"</td>\n",
              "      <td>1</td>\n",
              "      <td>1</td>\n",
              "      <td>4</td>\n",
              "      <td>1219017600</td>\n",
              "      <td>\"Delight\" says it all</td>\n",
              "      <td>This is a confection that has been around a fe...</td>\n",
              "      <td>2008-08-18</td>\n",
              "      <td>Monday</td>\n",
              "      <td>3</td>\n",
              "      <td>['gelatin', 'pillowy', 'mouthful', 'coated', '...</td>\n",
              "      <td>0.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>4</td>\n",
              "      <td>B000UA0QIQ</td>\n",
              "      <td>A395BORC6FGVXV</td>\n",
              "      <td>Karl</td>\n",
              "      <td>3</td>\n",
              "      <td>3</td>\n",
              "      <td>2</td>\n",
              "      <td>1307923200</td>\n",
              "      <td>Cough Medicine</td>\n",
              "      <td>If you are looking for the secret ingredient i...</td>\n",
              "      <td>2011-06-13</td>\n",
              "      <td>Monday</td>\n",
              "      <td>3</td>\n",
              "      <td>['medicinal', 'robitussin', 'cherry', 'ingredi...</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>5</td>\n",
              "      <td>B006K2ZZ7K</td>\n",
              "      <td>A1UQRSCLF8GW1T</td>\n",
              "      <td>Michael D. Bigham \"M. Wassir\"</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>5</td>\n",
              "      <td>1350777600</td>\n",
              "      <td>Great taffy</td>\n",
              "      <td>Great taffy at a great price.  There was a wid...</td>\n",
              "      <td>2012-10-21</td>\n",
              "      <td>Sunday</td>\n",
              "      <td>2</td>\n",
              "      <td>['quick', 'wide', 'delivery', 'deal', 'lover']</td>\n",
              "      <td>0.0</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>568449</th>\n",
              "      <td>568450</td>\n",
              "      <td>B001EO7N10</td>\n",
              "      <td>A28KG5XORO54AY</td>\n",
              "      <td>Lettie D. Carter</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>5</td>\n",
              "      <td>1299628800</td>\n",
              "      <td>Will not do without</td>\n",
              "      <td>Great for sesame chicken..this is a good if no...</td>\n",
              "      <td>2011-03-09</td>\n",
              "      <td>Wednesday</td>\n",
              "      <td>1</td>\n",
              "      <td>['sesame', 'chicken', 'husband', 'recipes', 'r...</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>568450</th>\n",
              "      <td>568451</td>\n",
              "      <td>B003S1WTCU</td>\n",
              "      <td>A3I8AFVPEE8KI5</td>\n",
              "      <td>R. Sawyer</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>2</td>\n",
              "      <td>1331251200</td>\n",
              "      <td>disappointed</td>\n",
              "      <td>I'm disappointed with the flavor. The chocolat...</td>\n",
              "      <td>2012-03-09</td>\n",
              "      <td>Friday</td>\n",
              "      <td>9</td>\n",
              "      <td>['notes', 'small', 'thanks', 'weak', 'disappoi...</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>568451</th>\n",
              "      <td>568452</td>\n",
              "      <td>B004I613EE</td>\n",
              "      <td>A121AA1GQV751Z</td>\n",
              "      <td>pksd \"pk_007\"</td>\n",
              "      <td>2</td>\n",
              "      <td>2</td>\n",
              "      <td>5</td>\n",
              "      <td>1329782400</td>\n",
              "      <td>Perfect for our maltipoo</td>\n",
              "      <td>These stars are small, so you can give 10-15 o...</td>\n",
              "      <td>2012-02-21</td>\n",
              "      <td>Tuesday</td>\n",
              "      <td>1</td>\n",
              "      <td>['compare', 'train', 'smell', 'preservatives',...</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>568452</th>\n",
              "      <td>568453</td>\n",
              "      <td>B004I613EE</td>\n",
              "      <td>A3IBEVCTXKNOH</td>\n",
              "      <td>Kathy A. Welch \"katwel\"</td>\n",
              "      <td>1</td>\n",
              "      <td>1</td>\n",
              "      <td>5</td>\n",
              "      <td>1331596800</td>\n",
              "      <td>Favorite Training and reward treat</td>\n",
              "      <td>These are the BEST treats for training and rew...</td>\n",
              "      <td>2012-03-13</td>\n",
              "      <td>Tuesday</td>\n",
              "      <td>2</td>\n",
              "      <td>['rewarding', 'grooming', 'sweet', 'wet', 'tre...</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>568453</th>\n",
              "      <td>568454</td>\n",
              "      <td>B001LR2CU2</td>\n",
              "      <td>A3LGQPJCZVL9UC</td>\n",
              "      <td>srfell17</td>\n",
              "      <td>0</td>\n",
              "      <td>0</td>\n",
              "      <td>5</td>\n",
              "      <td>1338422400</td>\n",
              "      <td>Great Honey</td>\n",
              "      <td>I am very satisfied ,product is as advertised,...</td>\n",
              "      <td>2012-05-31</td>\n",
              "      <td>Thursday</td>\n",
              "      <td>1</td>\n",
              "      <td>['general', 'cereal', 'sweetner', 'vinegar']</td>\n",
              "      <td>NaN</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>568454 rows × 15 columns</p>\n",
              "</div>\n",
              "      <button class=\"colab-df-convert\" onclick=\"convertToInteractive('df-685a80db-41fb-4692-ae8e-b0d4c8b3ede2')\"\n",
              "              title=\"Convert this dataframe to an interactive table.\"\n",
              "              style=\"display:none;\">\n",
              "        \n",
              "  <svg xmlns=\"http://www.w3.org/2000/svg\" height=\"24px\"viewBox=\"0 0 24 24\"\n",
              "       width=\"24px\">\n",
              "    <path d=\"M0 0h24v24H0V0z\" fill=\"none\"/>\n",
              "    <path d=\"M18.56 5.44l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94zm-11 1L8.5 8.5l.94-2.06 2.06-.94-2.06-.94L8.5 2.5l-.94 2.06-2.06.94zm10 10l.94 2.06.94-2.06 2.06-.94-2.06-.94-.94-2.06-.94 2.06-2.06.94z\"/><path d=\"M17.41 7.96l-1.37-1.37c-.4-.4-.92-.59-1.43-.59-.52 0-1.04.2-1.43.59L10.3 9.45l-7.72 7.72c-.78.78-.78 2.05 0 2.83L4 21.41c.39.39.9.59 1.41.59.51 0 1.02-.2 1.41-.59l7.78-7.78 2.81-2.81c.8-.78.8-2.07 0-2.86zM5.41 20L4 18.59l7.72-7.72 1.47 1.35L5.41 20z\"/>\n",
              "  </svg>\n",
              "      </button>\n",
              "      \n",
              "  <style>\n",
              "    .colab-df-container {\n",
              "      display:flex;\n",
              "      flex-wrap:wrap;\n",
              "      gap: 12px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert {\n",
              "      background-color: #E8F0FE;\n",
              "      border: none;\n",
              "      border-radius: 50%;\n",
              "      cursor: pointer;\n",
              "      display: none;\n",
              "      fill: #1967D2;\n",
              "      height: 32px;\n",
              "      padding: 0 0 0 0;\n",
              "      width: 32px;\n",
              "    }\n",
              "\n",
              "    .colab-df-convert:hover {\n",
              "      background-color: #E2EBFA;\n",
              "      box-shadow: 0px 1px 2px rgba(60, 64, 67, 0.3), 0px 1px 3px 1px rgba(60, 64, 67, 0.15);\n",
              "      fill: #174EA6;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert {\n",
              "      background-color: #3B4455;\n",
              "      fill: #D2E3FC;\n",
              "    }\n",
              "\n",
              "    [theme=dark] .colab-df-convert:hover {\n",
              "      background-color: #434B5C;\n",
              "      box-shadow: 0px 1px 3px 1px rgba(0, 0, 0, 0.15);\n",
              "      filter: drop-shadow(0px 1px 2px rgba(0, 0, 0, 0.3));\n",
              "      fill: #FFFFFF;\n",
              "    }\n",
              "  </style>\n",
              "\n",
              "      <script>\n",
              "        const buttonEl =\n",
              "          document.querySelector('#df-685a80db-41fb-4692-ae8e-b0d4c8b3ede2 button.colab-df-convert');\n",
              "        buttonEl.style.display =\n",
              "          google.colab.kernel.accessAllowed ? 'block' : 'none';\n",
              "\n",
              "        async function convertToInteractive(key) {\n",
              "          const element = document.querySelector('#df-685a80db-41fb-4692-ae8e-b0d4c8b3ede2');\n",
              "          const dataTable =\n",
              "            await google.colab.kernel.invokeFunction('convertToInteractive',\n",
              "                                                     [key], {});\n",
              "          if (!dataTable) return;\n",
              "\n",
              "          const docLinkHtml = 'Like what you see? Visit the ' +\n",
              "            '<a target=\"_blank\" href=https://colab.research.google.com/notebooks/data_table.ipynb>data table notebook</a>'\n",
              "            + ' to learn more about interactive tables.';\n",
              "          element.innerHTML = '';\n",
              "          dataTable['output_type'] = 'display_data';\n",
              "          await google.colab.output.renderOutput(dataTable, element);\n",
              "          const docLink = document.createElement('div');\n",
              "          docLink.innerHTML = docLinkHtml;\n",
              "          element.appendChild(docLink);\n",
              "        }\n",
              "      </script>\n",
              "    </div>\n",
              "  </div>\n",
              "  "
            ],
            "text/plain": [
              "            Id  ... labels\n",
              "0            1  ...    1.0\n",
              "1            2  ...    NaN\n",
              "2            3  ...    0.0\n",
              "3            4  ...    NaN\n",
              "4            5  ...    0.0\n",
              "...        ...  ...    ...\n",
              "568449  568450  ...    NaN\n",
              "568450  568451  ...    NaN\n",
              "568451  568452  ...    NaN\n",
              "568452  568453  ...    NaN\n",
              "568453  568454  ...    NaN\n",
              "\n",
              "[568454 rows x 15 columns]"
            ]
          },
          "metadata": {},
          "execution_count": 58
        }
      ],
      "source": [
        "sorting_labels(df)\n",
        "sorting_hints(df)"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# 라벨 분포 => 3 (둘 다 있는 경우의 라벨이 매우 적다.)\n",
        "df['labels'].value_counts()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "_pcNXbkJiwlo",
        "outputId": "03b789bb-ecaf-4210-edd8-4e617f81c279"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.0    52211\n",
              "1.0    35806\n",
              "2.0    17439\n",
              "3.0      430\n",
              "Name: labels, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 48
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#######\n",
        "######## 텍스트에 dog or cat이 없더라도 Summary에 있는 경우 캐치한다.\n",
        "\n",
        "df_bert= df[(((df['Summary'].str.contains('dog')) |\n",
        "    (df['Summary'].str.contains('Dog')) |\n",
        "    (df['Summary'].str.contains('DOG'))) &\n",
        "    ~((df['Text'].str.contains('dog')) |\n",
        "    (df['Text'].str.contains('Dog')) |\n",
        "    (df['Text'].str.contains('DOG')))) |\n",
        "\n",
        "    (((df['Summary'].str.contains('cat')) |\n",
        "    (df['Summary'].str.contains('Cat')) |\n",
        "    (df['Summary'].str.contains('CAT'))) &\n",
        "    ~((df['Text'].str.contains('cat')) |\n",
        "    (df['Text'].str.contains('Cat')) |\n",
        "    (df['Text'].str.contains('CAT')))) |\n",
        "\n",
        "    ((df['labels']== 0) &\n",
        "    ~((df['Text'].str.contains('cat')) |\n",
        "    (df['Text'].str.contains('Cat')) |\n",
        "    (df['Text'].str.contains('CAT')) |\n",
        "    (df['Text'].str.contains('dog')) |\n",
        "    (df['Text'].str.contains('Dog')) |\n",
        "    (df['Text'].str.contains('DOG')))) |\n",
        "    \n",
        "    (df['labels']==3)]\n"
      ],
      "metadata": {
        "id": "NDiENU8JdVUr"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df['labels'].value_counts()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "T6MrTKLbFSNP",
        "outputId": "acfec0a6-4f6b-4f79-b2ea-b2feef6bab48"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.0    39068\n",
              "1.0    35785\n",
              "2.0    17465\n",
              "3.0      420\n",
              "Name: labels, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert['labels'].value_counts()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "53491412-69fe-42f4-d28f-f74bf2d44d3a",
        "id": "EtFvk3g5dVUs"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.0    37109\n",
              "1.0     2811\n",
              "2.0      962\n",
              "3.0      420\n",
              "Name: labels, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 10
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "### 반려동물(dog, cat)을 키우는 고객 여부 파악. 2차 BERT를 이용한 분류\n",
        "\n",
        "dog과 cat의 여부를 파악할 수 있는 words를 포함하고 있다면, 각각의 label을 부여한다.  \n",
        "(No: 0, dog: 1, cat: 2, both: 3)"
      ],
      "metadata": {
        "id": "yDbpy2S87nLx"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### Tokenizer 작성 \n",
        "\n",
        "모델 입력에 이용할 BertTokenizer를 작성합니다. "
      ],
      "metadata": {
        "id": "8aSGudMb2eOT"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "\n",
        "# Config 설적을 통해 Bert base 모델 가져오기\n",
        "config = BertConfig.from_pretrained('bert-base-uncased')\n",
        "config.output_hidden_states = False\n",
        "\n",
        "# 토크나이저 정의 - Config를 일치시켜 같은 모델에 대한 토크나이저를 가져오기\n",
        "tokenizer = BertTokenizer.from_pretrained(pretrained_model_name_or_path = 'bert-base-uncased', config = config)\n",
        "\n",
        "# Bert 모델 가져오기\n",
        "transformer_model = TFBertModel.from_pretrained('bert-base-uncased', config = config)\n"
      ],
      "metadata": {
        "id": "NJN71P_q7EAE",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "referenced_widgets": [
            "ee7ba6b4edd240a7b445361cd247cf67",
            "956db3d636f140329540739490d6c7f0",
            "3c1f42928693403498d8a9a751294d1d",
            "06ebd9bd0ad445d8be335772043ac15c",
            "47a1cba1b4404be39d0296a4a4ad05c6",
            "17a2c98e930a4288995684ca90655250",
            "c58e41f6ef69457184691af00d7dbba6",
            "6f9e8ab5028042d5b02f8b857c554150",
            "83608cc497f24f6980d064ce3f202674",
            "003345b1297d4f5fa29fee56786eddec",
            "b48bf9855ceb4406806f8de06eabff48",
            "91831e18881945bb9f181d17a38135b0",
            "8cd717ce7c554817949c0a040e7b6009",
            "42905d608e224626b41ab1a79803eaa9",
            "7997fbd2b9164995bc96e07b4fd1be2d",
            "a68b4daabaf14a79afb3471b39dae5cc",
            "f00e73fc84a9411286ed10a7a4face25",
            "ab687cca1b8b4c6e82b667290b01b75f",
            "932117b02c584f2fa28292365a9a4d0e",
            "441033095bad4ba5ba50735b9726e999",
            "f0434fe523b742189ec1e443c58c05ef",
            "2bd93abf2edd4f65b660341bb4e52cf9",
            "753d1e8bfbb0422697e7b69545640ece",
            "11bbb9129bde45dfb5a06d9a92ffeee3",
            "68169b6777b84d5f9739e54713676e56",
            "61240cef01044bb69adbfca02f81e81d",
            "26d919a496fe44a3b3a0729c78da8c37",
            "dcad205fff5a4fc3a112cb1e062aba75",
            "e341598cc2b341ca94a6250898f03ef4",
            "ced46c79f42f43849e978dca2bd0c127",
            "58ca58deb2e549a284bb960eff195335",
            "948ab2240733474aa5c293b2ebde0320",
            "e0d60cf485864f37b29c4909ea7dd422",
            "fac7c174c64641b7bdad8a5d5fc17062",
            "fe158422b282423db8feca1f47962e3b",
            "154d81e893de462bbc915273f2d168d5",
            "555f7bd14c3c4269b1682f158dcc2828",
            "7dac593cd0604122af6566c3216cb188",
            "ee3e9ec863c44c8396c38edae80d9361",
            "59a4ab7761ce4d8e9004f3fa617cd90f",
            "e2249acb94bc42ff83eedbe20a61b5b2",
            "dcf657c5a62348d99db498411cb6b8ce",
            "65d8075e2c6d4726919d76ede90aaab9",
            "8b3cf1ac62e34582b9aae0a5de1186c4",
            "8f48177e535d46f18155548db339b259",
            "d29e921b7d7a41458ab70ecd96333164",
            "4e92415c720b4aa8b7779a7db9a39e40",
            "f5bd69de0f77461fa78ad8b37b677d80",
            "b7b9d921373047f2bafddd541875c7b3",
            "3fc2ff3725b34cf6865c11c1f106b3a0",
            "0133f671023a4831b62511a666b98da2",
            "ab759ad121d049a490e923797a2c6879",
            "f7b98a705cfa4845b74c9afeffce1e70",
            "28ec73b4c2854985ad6b89fb53cfd325",
            "86bc2546c55f46918878f3ac8158f6f9"
          ],
          "height": 268
        },
        "outputId": "0cccc09f-41b8-4b27-848b-978f35941167"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "ee7ba6b4edd240a7b445361cd247cf67",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "Downloading:   0%|          | 0.00/570 [00:00<?, ?B/s]"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "91831e18881945bb9f181d17a38135b0",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "Downloading:   0%|          | 0.00/28.0 [00:00<?, ?B/s]"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "753d1e8bfbb0422697e7b69545640ece",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "Downloading:   0%|          | 0.00/226k [00:00<?, ?B/s]"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "fac7c174c64641b7bdad8a5d5fc17062",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "Downloading:   0%|          | 0.00/455k [00:00<?, ?B/s]"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "application/vnd.jupyter.widget-view+json": {
              "model_id": "8f48177e535d46f18155548db339b259",
              "version_minor": 0,
              "version_major": 2
            },
            "text/plain": [
              "Downloading:   0%|          | 0.00/511M [00:00<?, ?B/s]"
            ]
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "Some layers from the model checkpoint at bert-base-uncased were not used when initializing TFBertModel: ['nsp___cls', 'mlm___cls']\n",
            "- This IS expected if you are initializing TFBertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
            "- This IS NOT expected if you are initializing TFBertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n",
            "All the layers of TFBertModel were initialized from the model checkpoint at bert-base-uncased.\n",
            "If your task is similar to the task the model of the checkpoint was trained on, you can already use TFBertModel for predictions without further training.\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### Model Architecture 작성\n"
      ],
      "metadata": {
        "id": "F3nLKCkD3I2U"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "bert = transformer_model.layers[0]\n",
        "input_ids = tf.keras.Input(shape = (128, ), name = 'input_ids', dtype = 'int32')\n",
        "inputs = {'input':input_ids}\n",
        "bert_model = bert(input_ids)[1]\n",
        "dropout = Dropout(config.hidden_dropout_prob, name = 'pooled_output')\n",
        "pooled_output = dropout(bert_model, training=False)\n",
        "outputs = tf.keras.layers.Dense(4, activation = 'softmax', kernel_initializer = tf.keras.initializers.TruncatedNormal(stddev=config.initializer_range), name = 'output')(pooled_output)\n",
        "model = tf.keras.Model(inputs = inputs, outputs = outputs, name = 'Bert_Classifier')\n",
        "model.summary()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "b083f2bd-2af7-4ead-a793-dd52491935a7",
        "id": "v2XeISD6Z9tF"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"Bert_Classifier\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " input_ids (InputLayer)      [(None, 128)]             0         \n",
            "                                                                 \n",
            " bert (TFBertMainLayer)      TFBaseModelOutputWithPoo  109482240 \n",
            "                             lingAndCrossAttentions(l            \n",
            "                             ast_hidden_state=(None,             \n",
            "                             128, 768),                          \n",
            "                              pooler_output=(None, 76            \n",
            "                             8),                                 \n",
            "                              past_key_values=None, h            \n",
            "                             idden_states=None, atten            \n",
            "                             tions=None, cross_attent            \n",
            "                             ions=None)                          \n",
            "                                                                 \n",
            " pooled_output (Dropout)     (None, 768)               0         \n",
            "                                                                 \n",
            " output (Dense)              (None, 4)                 3076      \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 109,485,316\n",
            "Trainable params: 109,485,316\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### Under Sampling\n",
        "\n",
        "고양이와 강아지 모두 키우는 사람이 수가 적으므로 해당 수에 맞춰서 학습시킨다."
      ],
      "metadata": {
        "id": "jNF9v6cxrJP6"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_1 = df_bert[df_bert['labels']==1][:300]\n",
        "df_bert_2 = df_bert[df_bert['labels']==2][:300]\n",
        "df_bert_0 = df_bert[df_bert['labels']==0][:300]\n",
        "df_bert_3 = df_bert[df_bert['labels']==3][:300]\n",
        "df_bert_train = pd.concat([df_bert_1,df_bert_2,df_bert_3,df_bert_0])\n",
        "df_bert_train = df_bert_train.sample(frac=1)"
      ],
      "metadata": {
        "id": "BytDlUEHdVUs"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_11 = df_bert[df_bert['labels']==1][300:420]\n",
        "df_bert_22 = df_bert[df_bert['labels']==2][300:420]\n",
        "df_bert_00 = df_bert[df_bert['labels']==0][300:420]\n",
        "df_bert_33 = df_bert[df_bert['labels']==3][300:420]\n",
        "df_bert_val = pd.concat([df_bert_11,df_bert_22,df_bert_33,df_bert_00])\n",
        "df_bert_val = df_bert_val.sample(frac=1)"
      ],
      "metadata": {
        "id": "W1bew35hdVUs"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_111 = df_bert[df_bert['labels']==1][420:5420]\n",
        "df_bert_222 = df_bert[df_bert['labels']==2][420:5420]\n",
        "df_bert_000 = df_bert[df_bert['labels']==0][420:5420]\n",
        "# df_bert_33 = df_bert[df_bert['labels']==3][420:5420]\n",
        "df_bert_test = pd.concat([df_bert_111,df_bert_222,df_bert_000])\n",
        "df_bert_test = df_bert_test.sample(frac=1)"
      ],
      "metadata": {
        "id": "aXhj-95zf-Iu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df[df['labels'].notnull()]['labels'].value_counts()"
      ],
      "metadata": {
        "outputId": "e4f871c6-f0de-46ba-d236-66a432b4e4e3",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "w8qJewpmdVUt"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.0    39068\n",
              "1.0    35785\n",
              "2.0    17465\n",
              "3.0      420\n",
              "Name: labels, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "%%time\n",
        "\n",
        "optimizer = tf.keras.optimizers.Adam(learning_rate = 5e-05,\n",
        "                                     epsilon=1e-08,\n",
        "                                     decay=0.01,\n",
        "                                     clipnorm=1.0)\n",
        "loss = 'sparse_categorical_crossentropy'\n",
        "metric = ['accuracy']\n",
        "\n",
        "model.compile(optimizer = optimizer, loss = loss, metrics = metric)\n",
        "\n",
        "x= tokenizer(\n",
        "    text = df_bert_train['Text'].to_list(),\n",
        "    add_special_tokens=True,\n",
        "    max_length = 128,\n",
        "    truncation = True,\n",
        "    padding = True,\n",
        "    return_tensors='tf',\n",
        "    return_token_type_ids=False,\n",
        "    # returen_attention_mask=False,\n",
        "    verbose = True\n",
        "     \n",
        ")\n",
        "\n",
        "early = tf.keras.callbacks.EarlyStopping(monitor = 'val_loss', patience=3, restore_best_weights= True)\n",
        "\n",
        "\n",
        "history = model.fit(x['input_ids'],df_bert_train['labels'], validation_split=0.2, batch_size = 64, epochs=10, callbacks=[early])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "QD4qJkcEIPWf",
        "outputId": "b811c537-810d-4f7b-8988-5bfd9ba3e859"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/10\n",
            "15/15 [==============================] - 32s 1s/step - loss: 1.3387 - accuracy: 0.3406 - val_loss: 1.3053 - val_accuracy: 0.3958\n",
            "Epoch 2/10\n",
            "15/15 [==============================] - 14s 972ms/step - loss: 1.2403 - accuracy: 0.4219 - val_loss: 1.1563 - val_accuracy: 0.4917\n",
            "Epoch 3/10\n",
            "15/15 [==============================] - 14s 972ms/step - loss: 0.9695 - accuracy: 0.5698 - val_loss: 0.7455 - val_accuracy: 0.6917\n",
            "Epoch 4/10\n",
            "15/15 [==============================] - 14s 973ms/step - loss: 0.6197 - accuracy: 0.7635 - val_loss: 0.4728 - val_accuracy: 0.8417\n",
            "Epoch 5/10\n",
            "15/15 [==============================] - 14s 973ms/step - loss: 0.3886 - accuracy: 0.8823 - val_loss: 0.4667 - val_accuracy: 0.8167\n",
            "Epoch 6/10\n",
            "15/15 [==============================] - 14s 973ms/step - loss: 0.2298 - accuracy: 0.9406 - val_loss: 0.4369 - val_accuracy: 0.8458\n",
            "Epoch 7/10\n",
            "15/15 [==============================] - 14s 953ms/step - loss: 0.1198 - accuracy: 0.9792 - val_loss: 0.4439 - val_accuracy: 0.8458\n",
            "Epoch 8/10\n",
            "15/15 [==============================] - 14s 951ms/step - loss: 0.0728 - accuracy: 0.9917 - val_loss: 0.4447 - val_accuracy: 0.8500\n",
            "Epoch 9/10\n",
            "15/15 [==============================] - 14s 964ms/step - loss: 0.0508 - accuracy: 0.9917 - val_loss: 0.5065 - val_accuracy: 0.8583\n",
            "CPU times: user 1min 35s, sys: 47.6 s, total: 2min 22s\n",
            "Wall time: 2min 30s\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.metrics import accuracy_score, confusion_matrix, classification_report\n",
        "x1= tokenizer(\n",
        "    text = df_bert_val['Text'].to_list(),\n",
        "    add_special_tokens=True,\n",
        "    max_length = 128,\n",
        "    truncation = True,\n",
        "    padding = True,\n",
        "    return_tensors='tf',\n",
        "    return_token_type_ids=False,\n",
        "    # returen_attention_mask=False,\n",
        "    verbose = True\n",
        ")\n",
        "\n",
        "pred = model.predict(x1['input_ids'])\n",
        "# pred_opt = pred >= 0.5\n",
        "\n",
        "pred_result = [np.argmax(pred[i]) for i in range(len(pred))]\n",
        "len(pred_result)\n",
        "\n",
        "\n",
        "print(\"accuracy: {:.3f}\".format(accuracy_score(df_bert_val['labels'],pred_result)))\n",
        "# print(\"f1_score \\n\",f1_score(df_bert_val['labels'],pred_result))\n",
        "print(\"confusion matrix \\n\",confusion_matrix(df_bert_val['labels'],pred_result))\n",
        "print('classification report \\n', classification_report(df_bert_val['labels'], pred_result, target_names = ['0 없음','1 강아지','2 고양이','3 강아지 + 고양이']))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "UvaAAbtxZrKE",
        "outputId": "fea586cd-8ad3-4bd5-d537-c5b4b50e30aa"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "accuracy: 0.890\n",
            "confusion matrix \n",
            " [[111   0   9   0]\n",
            " [  1 108   9   2]\n",
            " [ 13  13  90   4]\n",
            " [  0   1   1 118]]\n",
            "classification report \n",
            "               precision    recall  f1-score   support\n",
            "\n",
            "        0 없음       0.89      0.93      0.91       120\n",
            "       1 강아지       0.89      0.90      0.89       120\n",
            "       2 고양이       0.83      0.75      0.79       120\n",
            " 3 강아지 + 고양이       0.95      0.98      0.97       120\n",
            "\n",
            "    accuracy                           0.89       480\n",
            "   macro avg       0.89      0.89      0.89       480\n",
            "weighted avg       0.89      0.89      0.89       480\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        ""
      ],
      "metadata": {
        "id": "aXfexs_VKu4a"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### 재학습"
      ],
      "metadata": {
        "id": "aKzaJhvpJ9w0"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_1111 = df[df['labels']==1][:10000]\n",
        "df_bert_2222 = df[df['labels']==2][:10000]\n",
        "df_bert_0000 = df[df['labels']==0][:10000]\n",
        "# df_bert_3 = df[df['labels']==3][:300]\n",
        "df_bert_train2 = pd.concat([df_bert_1111,df_bert_2222,df_bert_0000])\n",
        "df_bert_train2 = df_bert_train2.sample(frac=1)"
      ],
      "metadata": {
        "id": "DN1ffNxHJXjE"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_11111 = df[df['labels']==1][10000:]\n",
        "df_bert_22222 = df[df['labels']==2][10000:]\n",
        "df_bert_00000 = df[df['labels']==0][10000:]\n",
        "df_bert_33333 = df[df['labels']==3]\n",
        "df_bert_test2 = pd.concat([df_bert_11111,df_bert_22222,df_bert_33333,df_bert_00000])\n",
        "df_bert_test2 = df_bert_test2.sample(frac=1)"
      ],
      "metadata": {
        "id": "91HrGVXMKelV"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "%%time\n",
        "\n",
        "optimizer = tf.keras.optimizers.Adam(learning_rate = 5e-05,\n",
        "                                     epsilon=1e-08,\n",
        "                                     decay=0.01,\n",
        "                                     clipnorm=1.0)\n",
        "loss = 'sparse_categorical_crossentropy'\n",
        "metric = ['accuracy']\n",
        "\n",
        "model.compile(optimizer = optimizer, loss = loss, metrics = metric)\n",
        "\n",
        "x2= tokenizer(\n",
        "    text = df_bert_train2['Text'].to_list(),\n",
        "    add_special_tokens=True,\n",
        "    max_length = 128,\n",
        "    truncation = True,\n",
        "    padding = True,\n",
        "    return_tensors='tf',\n",
        "    return_token_type_ids=False,\n",
        "    # returen_attention_mask=False,\n",
        "    verbose = True\n",
        "     \n",
        ")\n",
        "\n",
        "early = tf.keras.callbacks.EarlyStopping(monitor = 'val_loss', patience=3, restore_best_weights= True)\n",
        "\n",
        "\n",
        "history = model.fit(x2['input_ids'],df_bert_train2['labels'], validation_split=0.2, batch_size = 64, initial_epoch= history.epoch[-1], epochs=20, callbacks=[early])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "75IS3ZARK7e_",
        "outputId": "0a01fd23-36b2-4229-9c2f-46e4b610e751"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 9/20\n",
            "375/375 [==============================] - 371s 953ms/step - loss: 0.1222 - accuracy: 0.9625 - val_loss: 0.0831 - val_accuracy: 0.9780\n",
            "Epoch 10/20\n",
            "375/375 [==============================] - 354s 944ms/step - loss: 0.0569 - accuracy: 0.9841 - val_loss: 0.0732 - val_accuracy: 0.9810\n",
            "Epoch 11/20\n",
            "375/375 [==============================] - 353s 943ms/step - loss: 0.0396 - accuracy: 0.9893 - val_loss: 0.0742 - val_accuracy: 0.9813\n",
            "Epoch 12/20\n",
            "375/375 [==============================] - 354s 943ms/step - loss: 0.0293 - accuracy: 0.9934 - val_loss: 0.0708 - val_accuracy: 0.9817\n",
            "Epoch 13/20\n",
            "375/375 [==============================] - 353s 942ms/step - loss: 0.0227 - accuracy: 0.9947 - val_loss: 0.0725 - val_accuracy: 0.9827\n",
            "Epoch 14/20\n",
            "375/375 [==============================] - 353s 943ms/step - loss: 0.0186 - accuracy: 0.9959 - val_loss: 0.0774 - val_accuracy: 0.9827\n",
            "Epoch 15/20\n",
            "375/375 [==============================] - 354s 943ms/step - loss: 0.0153 - accuracy: 0.9968 - val_loss: 0.0761 - val_accuracy: 0.9825\n",
            "CPU times: user 25min 39s, sys: 14min 50s, total: 40min 29s\n",
            "Wall time: 42min 47s\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### 과적합일 수 있으므로 불균형 대용량 테스트셋 생성"
      ],
      "metadata": {
        "id": "QAwAYnWtaZY1"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.metrics import accuracy_score, confusion_matrix, classification_report, plot_confusion_matrix\n",
        "X= tokenizer(\n",
        "    text = df_bert_test2['Text'].to_list(),\n",
        "    add_special_tokens=True,\n",
        "    max_length = 128,\n",
        "    truncation = True,\n",
        "    padding = True,\n",
        "    return_tensors='tf',\n",
        "    return_token_type_ids=False,\n",
        "    # returen_attention_mask=False,\n",
        "    verbose = True\n",
        ")\n",
        "\n",
        "pred = model.predict(X['input_ids'])\n",
        "# pred_opt = pred >= 0.5\n",
        "\n",
        "pred_result = [np.argmax(pred[i]) for i in range(len(pred))]\n",
        "len(pred_result)\n",
        "\n",
        "\n",
        "print(\"accuracy: {:.3f}\".format(accuracy_score(df_bert_test2['labels'],pred_result)))\n",
        "# print(\"f1_score \\n\",f1_score(df_bert_test2['labels'],pred_result))\n",
        "print(\"confusion matrix \\n\",confusion_matrix(df_bert_test2['labels'],pred_result))\n",
        "print('classification report \\n', classification_report(df_bert_test2['labels'], pred_result, target_names = ['0 없음','1 강아지','2 고양이','3 강아지 + 고양이']))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "J5EVJAU-a_tV",
        "outputId": "fc444eca-341d-4aa2-a54f-eab1892427fc"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "accuracy: 0.974\n",
            "confusion matrix \n",
            " [[28485   174   409     0]\n",
            " [  215 25408   162     0]\n",
            " [  178    54  7233     0]\n",
            " [    1   179   240     0]]\n",
            "classification report \n",
            "               precision    recall  f1-score   support\n",
            "\n",
            "        0 없음       0.99      0.98      0.98     29068\n",
            "       1 강아지       0.98      0.99      0.98     25785\n",
            "       2 고양이       0.90      0.97      0.93      7465\n",
            " 3 강아지 + 고양이       0.00      0.00      0.00       420\n",
            "\n",
            "    accuracy                           0.97     62738\n",
            "   macro avg       0.72      0.73      0.73     62738\n",
            "weighted avg       0.97      0.97      0.97     62738\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### SMOTE를 통한 Over Sampling\n",
        "\n",
        "K-MEANS 군집화를 이용해 군집 내 랜덤 샘플 생성"
      ],
      "metadata": {
        "id": "hSnrXRaasyX-"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert['labels'].value_counts()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "v5U5iA5SFgL3",
        "outputId": "597ffc73-5921-40ad-84c4-9bacc5a9d351"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.0    37109\n",
              "1.0     2811\n",
              "2.0      962\n",
              "3.0      420\n",
              "Name: labels, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 40
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_1 = df_bert[df_bert['labels']==1][:800]\n",
        "df_bert_2 = df_bert[df_bert['labels']==2][:800]\n",
        "df_bert_0 = df_bert[df_bert['labels']==0][:800]\n",
        "df_bert_3 = df_bert[df_bert['labels']==3][:300]\n",
        "df_bert_train1 = pd.concat([df_bert_1,df_bert_2,df_bert_3,df_bert_0])\n",
        "df_bert_train1 = df_bert_train1.sample(frac=1)"
      ],
      "metadata": {
        "id": "pPolzCMgFdoR"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_11 = df_bert[df_bert['labels']==1][800:920]\n",
        "df_bert_22 = df_bert[df_bert['labels']==2][800:920]\n",
        "df_bert_00 = df_bert[df_bert['labels']==0][800:920]\n",
        "df_bert_33 = df_bert[df_bert['labels']==3][300:420]\n",
        "df_bert_val1 = pd.concat([df_bert_11,df_bert_22,df_bert_33,df_bert_00])\n",
        "df_bert_val1 = df_bert_val1.sample(frac=1)"
      ],
      "metadata": {
        "id": "0jQuYutJFdoX"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_train1.drop(columns=['ProfileName', 'HelpfulnessNumerator','HelpfulnessDenominator','date', 'keyword'], inplace = True)\n",
        "df_bert_val1.drop(columns=['ProfileName', 'HelpfulnessNumerator','HelpfulnessDenominator','date', 'keyword'], inplace = True)"
      ],
      "metadata": {
        "id": "VWj-Cf7TVYYQ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "from imblearn.over_sampling import BorderlineSMOTE\n",
        "import numpy as np\n",
        "from sklearn.cluster import KMeans\n",
        "from sklearn.decomposition import PCA \n",
        "from sklearn.preprocessing import StandardScaler\n",
        "from category_encoders import OrdinalEncoder\n",
        "\n",
        "def berttok(df, col):\n",
        "\n",
        "  df[col]= np.array(tokenizer(\n",
        "      text = df[col].to_list(),\n",
        "      add_special_tokens=True,\n",
        "      max_length = 128,\n",
        "      truncation = True,\n",
        "      padding = True,\n",
        "      # return_tensors='tf',\n",
        "      return_token_type_ids=False,\n",
        "      # returen_attention_mask=False,\n",
        "      verbose = True)['input_ids'])\n",
        "\n",
        "text = 'Text'\n",
        "summary = 'Summary'\n",
        "berttok(df_bert_train1, text)\n",
        "berttok(df_bert_train1, summary)\n",
        "berttok(df_bert_val1, text)\n",
        "berttok(df_bert_val1, summary)\n",
        "\n",
        "encoder = OrdinalEncoder(cols=['ProductId','UserId','dayofweek'])\n",
        "bert_train_encoded = encoder.fit_transform(df_bert_train1)\n",
        "bert_val_encoded = encoder.transform(df_bert_val1)\n",
        "\n",
        "scaler = StandardScaler()\n",
        "bert_train_scaled = scaler.fit_transform(bert_train_encoded)\n",
        "bert_val_scaled = scaler.transform(bert_val_encoded)\n",
        "\n",
        "\n",
        "pca = PCA(n_components = 2)           # 2차원으로 시각화를 진행할 것이므로 2개로 설정한다.\n",
        "pca.fit(bert_train_scaled)             \n",
        "df_p = pca.transform(bert_train_scaled)\n",
        "df_p = pd.DataFrame(df_p, columns = ['PC1','PC2']) #PCA진행 한 두 개의 값을 column으로 데이터프레임화 시킨다.\n",
        "df_p"
      ],
      "metadata": {
        "id": "at5q3PB8F6HQ"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# 군집 시각화\n",
        "sns.scatterplot(data = df_p, x = 'PC1',y = 'PC2', hue = 'label'); "
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 279
        },
        "id": "1C3pDw9OZogQ",
        "outputId": "39f19f1e-ee4d-428a-e776-0543189c241e"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# SMOTE 진행\n",
        "smote = BorderlineSMOTE()\n",
        "data,label = smote.fit_resample(df_p, labels)"
      ],
      "metadata": {
        "id": "WIJTmlIzGcjA"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# %%time\n",
        "\n",
        "bert = transformer_model.layers[0]\n",
        "input_ids = tf.keras.Input(shape = (10, ), name = 'input_ids', dtype = 'int32')\n",
        "inputs = {'input':input_ids}\n",
        "bert_model = bert(input_ids)[1]\n",
        "dropout = Dropout(config.hidden_dropout_prob, name = 'pooled_output')\n",
        "pooled_output = dropout(bert_model, training=False)\n",
        "outputs = tf.keras.layers.Dense(4, activation = 'softmax', kernel_initializer = tf.keras.initializers.TruncatedNormal(stddev=config.initializer_range), name = 'output')(pooled_output)\n",
        "model1= tf.keras.Model(inputs = inputs, outputs = outputs, name = 'Bert_Classifier')\n",
        "model1.summary()\n",
        "\n",
        "\n",
        "optimizer = tf.keras.optimizers.Adam(learning_rate = 5e-05,\n",
        "                                     epsilon=1e-08,\n",
        "                                     decay=0.01,\n",
        "                                     clipnorm=1.0)\n",
        "loss = 'sparse_categorical_crossentropy'\n",
        "metric = ['accuracy']\n",
        "\n",
        "model1.compile(optimizer = optimizer, loss = loss, metrics = metric)\n",
        "\n",
        "early = tf.keras.callbacks.EarlyStopping(monitor = 'val_loss', patience=3, restore_best_weights= True)\n",
        "\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "L4wpAhbbIMmg",
        "outputId": "86c7fd33-6e3f-4f69-b439-79cfc863547b"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"Bert_Classifier\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " input_ids (InputLayer)      [(None, 128)]             0         \n",
            "                                                                 \n",
            " bert (TFBertMainLayer)      multiple                  109482240 \n",
            "                                                                 \n",
            " pooled_output (Dropout)     (None, 768)               0         \n",
            "                                                                 \n",
            " output (Dense)              (None, 4)                 3076      \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 109,485,316\n",
            "Trainable params: 109,485,316\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "history = model1.fit(data,label, validation_split=0.2, batch_size = 64, epochs=10, callbacks=[early])"
      ],
      "metadata": {
        "id": "OthMcFWabMxn"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### Scaling 하지 않고 SMOTE 진행"
      ],
      "metadata": {
        "id": "N2qTVfejkDvu"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "x3= tokenizer(\n",
        "  text = df_bert_train1['Text'].to_list(),\n",
        "  add_special_tokens=True,\n",
        "  max_length = 128,\n",
        "  truncation = True,\n",
        "  padding = True,\n",
        "  return_tensors='tf',\n",
        "  return_token_type_ids=False,\n",
        "  # returen_attention_mask=False,\n",
        "  verbose = True)['input_ids']\n"
      ],
      "metadata": {
        "id": "hNg3cW2ReUCt"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "kmeans = KMeans().fit(x3)\n",
        "labels = kmeans.predict(x3)\n",
        "smote = BorderlineSMOTE()\n",
        "data,label = smote.fit_resample(x3,df_bert_train1['labels'])"
      ],
      "metadata": {
        "id": "Hr-yZD9ve1Nu"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "bert = transformer_model.layers[0]\n",
        "input_ids = tf.keras.Input(shape = (128, ), name = 'input_ids', dtype = 'int32')\n",
        "inputs = {'input':input_ids}\n",
        "bert_model = bert(input_ids)[1]\n",
        "dropout = Dropout(config.hidden_dropout_prob, name = 'pooled_output')\n",
        "pooled_output = dropout(bert_model, training=False)\n",
        "outputs = tf.keras.layers.Dense(4, activation = 'softmax', kernel_initializer = tf.keras.initializers.TruncatedNormal(stddev=config.initializer_range), name = 'output')(pooled_output)\n",
        "model2= tf.keras.Model(inputs = inputs, outputs = outputs, name = 'Bert_Classifier')\n",
        "model2.summary()\n",
        "\n",
        "\n",
        "optimizer = tf.keras.optimizers.Adam(learning_rate = 5e-06,\n",
        "                                     epsilon=1e-08,\n",
        "                                     decay=0.01,\n",
        "                                     clipnorm=1.0)\n",
        "loss = 'sparse_categorical_crossentropy'\n",
        "metric = ['accuracy']\n",
        "\n",
        "early = tf.keras.callbacks.EarlyStopping(monitor= 'val_loss', patience= 3)\n",
        "\n",
        "model2.compile(optimizer = optimizer, loss = loss, metrics = metric)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "CZMRRTDfe-il",
        "outputId": "ff1133e8-dd56-4b53-c20b-adaa01e33b0a"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"Bert_Classifier\"\n",
            "_________________________________________________________________\n",
            " Layer (type)                Output Shape              Param #   \n",
            "=================================================================\n",
            " input_ids (InputLayer)      [(None, 128)]             0         \n",
            "                                                                 \n",
            " bert (TFBertMainLayer)      TFBaseModelOutputWithPoo  109482240 \n",
            "                             lingAndCrossAttentions(l            \n",
            "                             ast_hidden_state=(None,             \n",
            "                             128, 768),                          \n",
            "                              pooler_output=(None, 76            \n",
            "                             8),                                 \n",
            "                              past_key_values=None, h            \n",
            "                             idden_states=None, atten            \n",
            "                             tions=None, cross_attent            \n",
            "                             ions=None)                          \n",
            "                                                                 \n",
            " pooled_output (Dropout)     (None, 768)               0         \n",
            "                                                                 \n",
            " output (Dense)              (None, 4)                 3076      \n",
            "                                                                 \n",
            "=================================================================\n",
            "Total params: 109,485,316\n",
            "Trainable params: 109,485,316\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "model2.fit(data,label, validation_split=0.2, batch_size = 64, epochs=10, callbacks=[early])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ViLgrnt2k8GK",
        "outputId": "03de723d-ae42-497e-9b26-7abae61c95ca"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/10\n",
            "40/40 [==============================] - 58s 991ms/step - loss: 1.1316 - accuracy: 0.6559 - val_loss: 1.5027 - val_accuracy: 0.4766\n",
            "Epoch 2/10\n",
            "40/40 [==============================] - 37s 936ms/step - loss: 0.6938 - accuracy: 0.9219 - val_loss: 1.5424 - val_accuracy: 0.4703\n",
            "Epoch 3/10\n",
            "40/40 [==============================] - 37s 936ms/step - loss: 0.4960 - accuracy: 0.9324 - val_loss: 1.5888 - val_accuracy: 0.4766\n",
            "Epoch 4/10\n",
            "40/40 [==============================] - 40s 1000ms/step - loss: 0.3835 - accuracy: 0.9461 - val_loss: 1.6445 - val_accuracy: 0.4766\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.History at 0x7f5bc8639d10>"
            ]
          },
          "metadata": {},
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "x4= tokenizer(\n",
        "text = df_bert_val1['Text'].to_list(),\n",
        "add_special_tokens=True,\n",
        "max_length = 128,\n",
        "truncation = True,\n",
        "padding = True,\n",
        "return_tensors='tf',\n",
        "return_token_type_ids=False,\n",
        "# returen_attention_mask=False,\n",
        "verbose = True)['input_ids']\n",
        "\n",
        "kmeans = KMeans().fit(x4)\n",
        "labels = kmeans.predict(x4)\n",
        "smote = BorderlineSMOTE()\n",
        "data,label = smote.fit_resample(x4, df_bert_val1['labels'])\n",
        "\n",
        "model2.evaluate(x4, df_bert_val1)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-WhZ9Gxaldzd",
        "outputId": "aa1c9f43-3cae-4868-e1b1-c1f57843c253"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "15/15 [==============================] - 4s 146ms/step - loss: 0.4508 - accuracy: 0.8896\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0.4508122503757477, 0.8895833492279053]"
            ]
          },
          "metadata": {},
          "execution_count": 21
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.metrics import accuracy_score, confusion_matrix, classification_report\n",
        "\n",
        "pred = model2.predict(data)\n",
        "# pred_opt = pred >= 0.5\n",
        "\n",
        "pred_result = [np.argmax(pred[i]) for i in range(len(pred))]\n",
        "len(pred_result)\n",
        "\n",
        "\n",
        "print(\"accuracy: {:.3f}\".format(accuracy_score(label,pred_result)))\n",
        "# print(\"f1_score \\n\",f1_score(label,pred_result))\n",
        "print(\"confusion matrix \\n\",confusion_matrix(label,pred_result))\n",
        "print('classification report \\n', classification_report(label, pred_result, target_names = ['0 없음','1 강아지','2 고양이','3 강아지 + 고양이']))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lrAikdQkli0l",
        "outputId": "ce1e6717-4eea-4f2f-802f-090997235f6d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "accuracy: 0.890\n",
            "confusion matrix \n",
            " [[100   3  17   0]\n",
            " [  1 108   9   2]\n",
            " [  8   5 102   5]\n",
            " [  1   1   1 117]]\n",
            "classification report \n",
            "               precision    recall  f1-score   support\n",
            "\n",
            "        0 없음       0.91      0.83      0.87       120\n",
            "       1 강아지       0.92      0.90      0.91       120\n",
            "       2 고양이       0.79      0.85      0.82       120\n",
            " 3 강아지 + 고양이       0.94      0.97      0.96       120\n",
            "\n",
            "    accuracy                           0.89       480\n",
            "   macro avg       0.89      0.89      0.89       480\n",
            "weighted avg       0.89      0.89      0.89       480\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from sklearn.metrics import accuracy_score, confusion_matrix, classification_report\n",
        "\n",
        "pred = model2.predict(x4)\n",
        "# pred_opt = pred >= 0.5\n",
        "\n",
        "pred_result = [np.argmax(pred[i]) for i in range(len(pred))]\n",
        "len(pred_result)\n",
        "\n",
        "\n",
        "print(\"accuracy: {:.3f}\".format(accuracy_score(df_bert_val1['labels'],pred_result)))\n",
        "# print(\"f1_score \\n\",f1_score(df_bert_val1['labels'],pred_result))\n",
        "print(\"confusion matrix \\n\",confusion_matrix(df_bert_val1['labels'],pred_result))\n",
        "print('classification report \\n', classification_report(df_bert_val1['labels'], pred_result, target_names = ['0 없음','1 강아지','2 고양이','3 강아지 + 고양이']))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "y7T7q3eqU_aZ",
        "outputId": "677fc8b6-6498-416f-9be8-c6679b84e1b7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "accuracy: 0.890\n",
            "confusion matrix \n",
            " [[100   3  17   0]\n",
            " [  1 108   9   2]\n",
            " [  8   5 102   5]\n",
            " [  1   1   1 117]]\n",
            "classification report \n",
            "               precision    recall  f1-score   support\n",
            "\n",
            "        0 없음       0.91      0.83      0.87       120\n",
            "       1 강아지       0.92      0.90      0.91       120\n",
            "       2 고양이       0.79      0.85      0.82       120\n",
            " 3 강아지 + 고양이       0.94      0.97      0.96       120\n",
            "\n",
            "    accuracy                           0.89       480\n",
            "   macro avg       0.89      0.89      0.89       480\n",
            "weighted avg       0.89      0.89      0.89       480\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "#### 최종 SMOTE test셋 검증"
      ],
      "metadata": {
        "id": "_MzcoiokSNrw"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_test = df.drop(index= df_bert['Id'].index)"
      ],
      "metadata": {
        "id": "nXIoEhpuSW09"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "df['labels'].value_counts()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "pwnLvn_CULTL",
        "outputId": "7e1e82be-413e-49af-9675-540c21c53f69"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.0    39068\n",
              "1.0    35785\n",
              "2.0    17465\n",
              "3.0      420\n",
              "Name: labels, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 48
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df_bert_111 = df_bert_test[df_bert_test['labels']==1][:10000]\n",
        "df_bert_222 = df_bert_test[df_bert_test['labels']==2][:10000]\n",
        "df_bert_000 = df_bert_test[df_bert_test['labels']==0][:10000]\n",
        "# df_bert_33 = df_bert_test[df_bert_test['labels']==3][300:420]\n",
        "df_bert_test1 = pd.concat([df_bert_111,df_bert_222,df_bert_000])\n",
        "df_bert_test1 = df_bert_test1.sample(frac=1)"
      ],
      "metadata": {
        "id": "mF2Q6jjWRZMR"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x5= tokenizer(\n",
        "text = df_bert_test1['Text'].to_list(),\n",
        "add_special_tokens=True,\n",
        "max_length = 128,\n",
        "truncation = True,\n",
        "padding = True,\n",
        "return_tensors='tf',\n",
        "return_token_type_ids=False,\n",
        "# returen_attention_mask=False,\n",
        "verbose = True)['input_ids']\n",
        "\n",
        "\n",
        "model2.evaluate(x5, df_bert_test1['labels'])"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "qn9XHIzWUj0f",
        "outputId": "7e4819c2-50ac-4025-ff4f-923e87ba54ef"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "687/687 [==============================] - 119s 169ms/step - loss: 1.9562 - accuracy: 0.1402\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[1.9561833143234253, 0.14021585881710052]"
            ]
          },
          "metadata": {},
          "execution_count": 52
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "pred = model2.predict(x5)\n",
        "# pred_opt = pred >= 0.5\n",
        "\n",
        "pred_result = [np.argmax(pred[i]) for i in range(len(pred))]\n",
        "len(pred_result)\n",
        "\n",
        "\n",
        "print(\"accuracy: {:.3f}\".format(accuracy_score(df_bert_test1['labels'],pred_result)))\n",
        "# print(\"f1_score \\n\",f1_score(df_bert_test1['labels'],pred_result))\n",
        "print(\"confusion matrix \\n\",confusion_matrix(df_bert_test1['labels'],pred_result))\n",
        "print('classification report \\n', classification_report(df_bert_test1['labels'], pred_result, target_names = ['0 없음','1 강아지','2 고양이','3 강아지 + 고양이']))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ieLEAOIIVRg3",
        "outputId": "4e48ca11-c210-4b5f-d80c-42a6e3c8769d"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "accuracy: 0.140\n",
            "confusion matrix \n",
            " [[1376   20  285  278]\n",
            " [ 117  223   96 9564]\n",
            " [ 195   42 1480 8283]\n",
            " [   0    0    0    0]]\n",
            "classification report \n",
            "               precision    recall  f1-score   support\n",
            "\n",
            "        0 없음       0.82      0.70      0.75      1959\n",
            "       1 강아지       0.78      0.02      0.04     10000\n",
            "       2 고양이       0.80      0.15      0.25     10000\n",
            " 3 강아지 + 고양이       0.00      0.00      0.00         0\n",
            "\n",
            "    accuracy                           0.14     21959\n",
            "   macro avg       0.60      0.22      0.26     21959\n",
            "weighted avg       0.79      0.14      0.20     21959\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "### 최종 카테고리 분류 진행\n",
        "\n",
        "위에서 학습한 내용으로 예측을 하여 자동으로 반려동물 유무 카테고리 라벨링을 한다."
      ],
      "metadata": {
        "id": "aIMUOmgChh4C"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "X_df= tokenizer(\n",
        "    text = df['Text'].to_list(),\n",
        "    add_special_tokens=True,\n",
        "    max_length = 128,\n",
        "    truncation = True,\n",
        "    padding = True,\n",
        "    return_tensors='tf',\n",
        "    return_token_type_ids=False,\n",
        "    # returen_attention_mask=False,\n",
        "    verbose = True\n",
        ")\n",
        "pred_df = model.predict(X_df['input_ids'])\n",
        "pred_result_df = [np.argmax(pred_df[i]) for i in range(len(pred_df))]\n"
      ],
      "metadata": {
        "id": "L7Rja_zjhhfH"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "plt.bar(df['pet'].value_counts().index, df['pet'].value_counts().values)\n",
        "plt.xticks(df['pet'].value_counts().index, ['No pet','Cat','Dog'])\n",
        "plt.show();\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        },
        "id": "ied3APM8Q7XT",
        "outputId": "b7e1256c-e502-47c3-e417-06ed24bc8ffd"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "df['pet'].value_counts()"
      ],
      "metadata": {
        "id": "0n8RdRvhy-8x",
        "outputId": "466b569f-0684-4a08-eac1-8de6e65fc78a",
        "colab": {
          "base_uri": "https://localhost:8080/"
        }
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0    415653\n",
              "2     87166\n",
              "1     65635\n",
              "Name: pet, dtype: int64"
            ]
          },
          "metadata": {},
          "execution_count": 79
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "### * 문맥을 통한 고객 집단 분류모델 학습 결과:  \n",
        "  * 학습시 under sampling이 SMOTE보다 성능이 좋으며, accuracy 90% 이상의 준수한 성능을 보임\n",
        "  * 모델 predict를 통해 자동으로 라벨링 작업을 진행할 수 있음 \n",
        "\n"
      ],
      "metadata": {
        "id": "yCnEsEtAy2OU"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "\n",
        "### * 문제 및 개선점:\n",
        "  * 1차 학습시 86%, 2차 학습시 95% 성능을 보였지만, 강아지와 고양이 둘 다 키우는 고객을 분류하는 작업은 accuracy 0으로 전혀 분류진행이 되지 않음. 초기 학습때 undersampling을 했지만, 1(dog)과 2(cat)에 대한 패턴과 상당히 유사해 그 차이를 파악하지 못한것으로 파악.\n",
        "    - 3번 카테고리를 없애거나 카테고리를 통합해 지도학습 진행 시 학습량 자체를 높여야 한다.\n",
        "    \n",
        "  * 1차 키워드 추출시에는 고양이관련 고객이 강아지관련 유저보다 적었지만, 2차 추출 모델링 이후엔 고양이 관련 유저가 급격히 늘게 됨  \n",
        "  * 고양이를 키우지 않는 많은 고객들에게 추천할 확률이 높으므로 개선 필요:\n",
        "\n",
        "    - 텍스트 전처리를 더 정교하게 진행하여 좀 더 성능을 높여본다.\n"
      ],
      "metadata": {
        "id": "6mB0XX4fidh0"
      }
    }
  ]
}