Subtask B/BERT_B.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 2,
  "metadata": {
    "colab": {
      "name": "BERT_B.ipynb",
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3.9.6 64-bit"
    },
    "language_info": {
      "name": "python",
      "version": "3.9.6",
      "mimetype": "text/x-python",
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "pygments_lexer": "ipython3",
      "nbconvert_exporter": "python",
      "file_extension": ".py"
    },
    "interpreter": {
      "hash": "aee8b7b246df8f9039afb4144a1f6fd8d2ca17a180786b69acc140d282b71a49"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "execution_count": 5,
      "source": [
        "import pandas as pd\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "from keras.preprocessing.text import Tokenizer\n",
        "from sklearn.model_selection import train_test_split\n",
        "from nltk.stem import WordNetLemmatizer\n",
        "from nltk import word_tokenize\n",
        "import nltk\n",
        "import os\n",
        "import collections\n",
        "import re\n",
        "import string\n",
        "\n",
        "import preprocessor as p\n",
        "import ktrain\n",
        "from ktrain import text\n",
        "import nltk\n",
        "from sklearn.preprocessing import LabelEncoder\n"
      ],
      "outputs": [],
      "metadata": {
        "id": "86BDG_rup-c5"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 6,
      "source": [
        "df = pd.read_csv(\"preprocess_data.csv\",index_col=0)\n",
        "df= df.dropna()\n",
        "df.head()"
      ],
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   Unnamed: 0.1                       _id  \\\n",
              "0          4986  60c5d6bf5659ea5e55defa2c   \n",
              "1          3394  60c5d6bf5659ea5e55def461   \n",
              "2          1310  60c5d6bf5659ea5e55defaad   \n",
              "3          3390  60c5d6bf5659ea5e55def419   \n",
              "4          4626  60c5d6bf5659ea5e55def7fa   \n",
              "\n",
              "                                                text task_1 task_2  \\\n",
              "0  @wealth if you made it through this &amp;&amp;...    HOF   PRFN   \n",
              "1  Technically that's still turning back the cloc...    HOF   OFFN   \n",
              "2  @VMBJP @BJP4Bengal @BJP4India @narendramodi @J...    NOT   NONE   \n",
              "3  @krtoprak_yigit Soldier of Japan Who has dick ...    HOF   OFFN   \n",
              "4  @blueheartedly You'd be better off asking who ...    HOF   OFFN   \n",
              "\n",
              "                                          text_clean  \n",
              "0  made amp amp onli abl start make money sustain...  \n",
              "1            technic still turn back clock dick head  \n",
              "2  govt stop think world media liber gang ani opt...  \n",
              "3                            soldier japan dick head  \n",
              "4         would better ask think sleazi shitbag lmao  "
            ],
            "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>Unnamed: 0.1</th>\n",
              "      <th>_id</th>\n",
              "      <th>text</th>\n",
              "      <th>task_1</th>\n",
              "      <th>task_2</th>\n",
              "      <th>text_clean</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>4986</td>\n",
              "      <td>60c5d6bf5659ea5e55defa2c</td>\n",
              "      <td>@wealth if you made it through this &amp;amp;&amp;amp;...</td>\n",
              "      <td>HOF</td>\n",
              "      <td>PRFN</td>\n",
              "      <td>made amp amp onli abl start make money sustain...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>3394</td>\n",
              "      <td>60c5d6bf5659ea5e55def461</td>\n",
              "      <td>Technically that's still turning back the cloc...</td>\n",
              "      <td>HOF</td>\n",
              "      <td>OFFN</td>\n",
              "      <td>technic still turn back clock dick head</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1310</td>\n",
              "      <td>60c5d6bf5659ea5e55defaad</td>\n",
              "      <td>@VMBJP @BJP4Bengal @BJP4India @narendramodi @J...</td>\n",
              "      <td>NOT</td>\n",
              "      <td>NONE</td>\n",
              "      <td>govt stop think world media liber gang ani opt...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3390</td>\n",
              "      <td>60c5d6bf5659ea5e55def419</td>\n",
              "      <td>@krtoprak_yigit Soldier of Japan Who has dick ...</td>\n",
              "      <td>HOF</td>\n",
              "      <td>OFFN</td>\n",
              "      <td>soldier japan dick head</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4626</td>\n",
              "      <td>60c5d6bf5659ea5e55def7fa</td>\n",
              "      <td>@blueheartedly You'd be better off asking who ...</td>\n",
              "      <td>HOF</td>\n",
              "      <td>OFFN</td>\n",
              "      <td>would better ask think sleazi shitbag lmao</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ]
          },
          "metadata": {},
          "execution_count": 6
        }
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 310
        },
        "id": "XZdcBF-Uq3A-",
        "outputId": "839a32d8-30ea-4238-9b1a-a6bb824034b5"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 7,
      "source": [
        "LE = LabelEncoder()\n",
        "LE.fit(df['task_2'])\n",
        "classes= list(LE.classes_)\n",
        "classes"
      ],
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['HATE', 'NONE', 'OFFN', 'PRFN']"
            ]
          },
          "metadata": {},
          "execution_count": 7
        }
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "QAjZmDOPrYmP",
        "outputId": "e3a195d5-57d6-427a-ea51-1465f913f709"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 8,
      "source": [
        "df['task_2'] = LE.fit_transform(df['task_2'])\n",
        "task1_params = dict(zip(LE.classes_, LE.transform(LE.classes_)))\n",
        "df.head()"
      ],
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "   Unnamed: 0.1                       _id  \\\n",
              "0          4986  60c5d6bf5659ea5e55defa2c   \n",
              "1          3394  60c5d6bf5659ea5e55def461   \n",
              "2          1310  60c5d6bf5659ea5e55defaad   \n",
              "3          3390  60c5d6bf5659ea5e55def419   \n",
              "4          4626  60c5d6bf5659ea5e55def7fa   \n",
              "\n",
              "                                                text task_1  task_2  \\\n",
              "0  @wealth if you made it through this &amp;&amp;...    HOF       3   \n",
              "1  Technically that's still turning back the cloc...    HOF       2   \n",
              "2  @VMBJP @BJP4Bengal @BJP4India @narendramodi @J...    NOT       1   \n",
              "3  @krtoprak_yigit Soldier of Japan Who has dick ...    HOF       2   \n",
              "4  @blueheartedly You'd be better off asking who ...    HOF       2   \n",
              "\n",
              "                                          text_clean  \n",
              "0  made amp amp onli abl start make money sustain...  \n",
              "1            technic still turn back clock dick head  \n",
              "2  govt stop think world media liber gang ani opt...  \n",
              "3                            soldier japan dick head  \n",
              "4         would better ask think sleazi shitbag lmao  "
            ],
            "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>Unnamed: 0.1</th>\n",
              "      <th>_id</th>\n",
              "      <th>text</th>\n",
              "      <th>task_1</th>\n",
              "      <th>task_2</th>\n",
              "      <th>text_clean</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>4986</td>\n",
              "      <td>60c5d6bf5659ea5e55defa2c</td>\n",
              "      <td>@wealth if you made it through this &amp;amp;&amp;amp;...</td>\n",
              "      <td>HOF</td>\n",
              "      <td>3</td>\n",
              "      <td>made amp amp onli abl start make money sustain...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>3394</td>\n",
              "      <td>60c5d6bf5659ea5e55def461</td>\n",
              "      <td>Technically that's still turning back the cloc...</td>\n",
              "      <td>HOF</td>\n",
              "      <td>2</td>\n",
              "      <td>technic still turn back clock dick head</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>1310</td>\n",
              "      <td>60c5d6bf5659ea5e55defaad</td>\n",
              "      <td>@VMBJP @BJP4Bengal @BJP4India @narendramodi @J...</td>\n",
              "      <td>NOT</td>\n",
              "      <td>1</td>\n",
              "      <td>govt stop think world media liber gang ani opt...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>3390</td>\n",
              "      <td>60c5d6bf5659ea5e55def419</td>\n",
              "      <td>@krtoprak_yigit Soldier of Japan Who has dick ...</td>\n",
              "      <td>HOF</td>\n",
              "      <td>2</td>\n",
              "      <td>soldier japan dick head</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>4626</td>\n",
              "      <td>60c5d6bf5659ea5e55def7fa</td>\n",
              "      <td>@blueheartedly You'd be better off asking who ...</td>\n",
              "      <td>HOF</td>\n",
              "      <td>2</td>\n",
              "      <td>would better ask think sleazi shitbag lmao</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ]
          },
          "metadata": {},
          "execution_count": 8
        }
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 310
        },
        "id": "Jg0Gvva0rG78",
        "outputId": "7157f7a5-30ec-42ec-d621-cb3c456f7c41"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 9,
      "source": [
        "LE.fit(df['task_2'])\n",
        "classes= list(LE.classes_)\n",
        "classes"
      ],
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[0, 1, 2, 3]"
            ]
          },
          "metadata": {},
          "execution_count": 9
        }
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XCTInWPrrMlK",
        "outputId": "c4e75740-fa9f-4889-aeb8-6d7bcf5672bf"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 10,
      "source": [
        "# split train dataset into train, validation and test sets\n",
        "train_text, temp_text, train_labels, temp_labels = train_test_split(df['text_clean'], df['task_2'], \n",
        "                                                                    random_state=42, \n",
        "                                                                    test_size=0.1)"
      ],
      "outputs": [],
      "metadata": {
        "id": "ADXmoUjNreuP"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 11,
      "source": [
        "X_train = train_text.tolist()\n",
        "X_test = temp_text.tolist()\n",
        "\n",
        "y_train = train_labels.tolist()\n",
        "y_test = temp_labels.tolist()"
      ],
      "outputs": [],
      "metadata": {
        "id": "tp9RrOF6rjUN"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 12,
      "source": [
        "import ktrain\n",
        "from ktrain import text\n",
        "MODEL_NAME = 'distilbert-base-uncased'  # replace this with model of choice\n",
        "t = text.Transformer(MODEL_NAME, maxlen=500, class_names=classes)\n",
        "trn = t.preprocess_train(X_train, y_train)\n",
        "val = t.preprocess_test(X_test, y_test)\n",
        "model = t.get_classifier()\n",
        "learner = ktrain.get_learner(model, train_data=trn, val_data=val, batch_size=6)\n",
        "history = learner.fit_onecycle(5e-5, 2)\n",
        "learner.validate(class_names=classes) # class_names must be string values"
      ],
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "preprocessing train...\n",
            "language: en\n",
            "train sequence lengths:\n",
            "\tmean : 14\n",
            "\t95percentile : 27\n",
            "\t99percentile : 31\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ],
            "text/html": []
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Is Multi-Label? False\n",
            "preprocessing test...\n",
            "language: en\n",
            "test sequence lengths:\n",
            "\tmean : 14\n",
            "\t95percentile : 27\n",
            "\t99percentile : 31\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<IPython.core.display.HTML object>"
            ],
            "text/html": []
          },
          "metadata": {}
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "\n",
            "\n",
            "begin training using onecycle policy with max lr of 5e-05...\n",
            "Epoch 1/2\n",
            "576/576 [==============================] - 10238s 18s/step - loss: 1.0421 - accuracy: 0.5601 - val_loss: 0.9838 - val_accuracy: 0.5938\n",
            "Epoch 2/2\n",
            "576/576 [==============================] - 12057s 21s/step - loss: 0.8326 - accuracy: 0.6612 - val_loss: 0.9446 - val_accuracy: 0.6094\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.41      0.54      0.46        67\n",
            "           1       0.63      0.64      0.64       121\n",
            "           2       0.41      0.13      0.20        68\n",
            "           3       0.74      0.87      0.80       128\n",
            "\n",
            "    accuracy                           0.61       384\n",
            "   macro avg       0.55      0.55      0.52       384\n",
            "weighted avg       0.59      0.61      0.58       384\n",
            "\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[ 36,  19,   8,   4],\n",
              "       [ 25,  78,   4,  14],\n",
              "       [ 23,  14,   9,  22],\n",
              "       [  4,  12,   1, 111]])"
            ]
          },
          "metadata": {},
          "execution_count": 12
        }
      ],
      "metadata": {
        "colab": {
          "background_save": true,
          "base_uri": "https://localhost:8080/",
          "height": 384
        },
        "id": "BOrstaiHrqv7",
        "outputId": "09a91af4-de7d-4ee9-c8cc-b9bedd8c41a1"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 13,
      "source": [
        "# list all data in history\n",
        "print(history.history.keys())"
      ],
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "dict_keys(['loss', 'accuracy', 'val_loss', 'val_accuracy', 'lr', 'iterations', 'momentum'])\n"
          ]
        }
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "wbPoyO9Ttl7l",
        "outputId": "f0cfda00-6aa4-4c75-f43a-8820e230fbc5"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 14,
      "source": [
        "# summarize history for accuracy\n",
        "plt.plot(history.history['accuracy'])\n",
        "plt.plot(history.history['val_accuracy'])\n",
        "plt.title('model accuracy')\n",
        "plt.ylabel('accuracy')\n",
        "plt.xlabel('epoch')\n",
        "plt.legend(['train', 'test'], loc='upper left')\n",
        "plt.show()\n",
        "# summarize history for loss\n",
        "plt.plot(history.history['loss'])\n",
        "plt.plot(history.history['val_loss'])\n",
        "plt.title('model loss')\n",
        "plt.ylabel('loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.legend(['train', 'test'], loc='upper left')\n",
        "plt.show()"
      ],
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ],
      "metadata": {}
    },
    {
      "cell_type": "code",
      "execution_count": 15,
      "source": [
        "learner.view_top_losses(n=1, preproc=t)"
      ],
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "----------\n",
            "id:107 | loss:4.68 | true:0 | pred:3)\n",
            "\n"
          ]
        }
      ],
      "metadata": {}
    },
    {
      "cell_type": "code",
      "execution_count": 16,
      "source": [
        "predictor = ktrain.get_predictor(learner.model, preproc=t)"
      ],
      "outputs": [],
      "metadata": {}
    },
    {
      "cell_type": "code",
      "execution_count": 17,
      "source": [
        "predictor.save('bert_B')"
      ],
      "outputs": [],
      "metadata": {}
    },
    {
      "cell_type": "code",
      "execution_count": 18,
      "source": [
        "reloaded_predictor = ktrain.load_predictor('bert_B')"
      ],
      "outputs": [],
      "metadata": {}
    },
    {
      "cell_type": "code",
      "execution_count": 19,
      "source": [
        "df_ans = pd.read_csv(\"preprocess_test_data.csv\",index_col=0)\n",
        "df_ans['text_clean'] = df_ans['text_clean'].fillna('neutral')\n",
        "df_ans.head()"
      ],
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "                        _id  \\\n",
              "0  60c5d6bf5659ea5e55deffcb   \n",
              "1  60c5d6bf5659ea5e55df028c   \n",
              "2  60c5d6bf5659ea5e55def377   \n",
              "3  60c5d6bf5659ea5e55def4c7   \n",
              "4  60c5d6bf5659ea5e55df01a6   \n",
              "\n",
              "                                                text  \\\n",
              "0  Fewer people coming in for vaccinations. So sa...   \n",
              "1  @MattHancock This may all be true. But... What...   \n",
              "2          @Layla_EFC I’ve unfollowed him the wanker   \n",
              "3  You guys are losing it all over the world. The...   \n",
              "4  And thus death laughs... It is sad merriment, ...   \n",
              "\n",
              "                                          text_clean  \n",
              "0  fewer peopl come vaccin sad nurs covidvaccin vumc  \n",
              "1                              may true piss big dom  \n",
              "2                                    unfollow wanker  \n",
              "3             guy lose world jung label islamophobia  \n",
              "4  thus death laugh sad merriment still covid ind...  "
            ],
            "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>_id</th>\n",
              "      <th>text</th>\n",
              "      <th>text_clean</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>60c5d6bf5659ea5e55deffcb</td>\n",
              "      <td>Fewer people coming in for vaccinations. So sa...</td>\n",
              "      <td>fewer peopl come vaccin sad nurs covidvaccin vumc</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>60c5d6bf5659ea5e55df028c</td>\n",
              "      <td>@MattHancock This may all be true. But... What...</td>\n",
              "      <td>may true piss big dom</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>60c5d6bf5659ea5e55def377</td>\n",
              "      <td>@Layla_EFC I’ve unfollowed him the wanker</td>\n",
              "      <td>unfollow wanker</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>3</th>\n",
              "      <td>60c5d6bf5659ea5e55def4c7</td>\n",
              "      <td>You guys are losing it all over the world. The...</td>\n",
              "      <td>guy lose world jung label islamophobia</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>4</th>\n",
              "      <td>60c5d6bf5659ea5e55df01a6</td>\n",
              "      <td>And thus death laughs... It is sad merriment, ...</td>\n",
              "      <td>thus death laugh sad merriment still covid ind...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ]
          },
          "metadata": {},
          "execution_count": 19
        }
      ],
      "metadata": {}
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "source": [
        "label = []\n",
        "i = 0\n",
        "for tweet in df_ans['text_clean'].tolist():\n",
        "    ans = reloaded_predictor.predict(tweet)\n",
        "    i += 1\n",
        "    print(i,len(tweet))\n",
        "    if (ans == 0):\n",
        "        label.append('HATE')\n",
        "    elif (ans == 1):\n",
        "        label.append('NONE')\n",
        "    elif (ans == 2):\n",
        "        label.append('OFFN')\n",
        "    elif (ans == 3):\n",
        "        label.append('PRFN')"
      ],
      "outputs": [],
      "metadata": {}
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "source": [
        "df_ans['label'] = pd.Series(label)\n",
        "df_ans.head(20)"
      ],
      "outputs": [],
      "metadata": {}
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "source": [
        "df_ans = df_ans.drop(['text','text_clean'],axis=1)\n",
        "df_ans = df_ans.reset_index()\n",
        "df_ans = df_ans.drop(['index'],axis=1)\n",
        "df_ans.head()"
      ],
      "outputs": [],
      "metadata": {}
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "source": [
        "df_ans.to_csv('BERT_B.csv',index=False)"
      ],
      "outputs": [],
      "metadata": {}
    }
  ]
}