Text classification with XLNET.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "4ee9846c",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Since the GPL-licensed package `unidecode` is not installed, using Python's `unicodedata` package which yields worse results.\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from cleantext import clean\n",
    "import re\n",
    "from transformers import XLNetTokenizer, XLNetForSequenceClassification, TrainingArguments, Trainer, pipeline\n",
    "import torch\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "import datasets \n",
    "import evaluate\n",
    "import random"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7c921f26",
   "metadata": {},
   "source": [
    "## Preprocess our data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "1230ce0b",
   "metadata": {},
   "outputs": [],
   "source": [
    "data_train = pd.read_csv('./emotions_data/emotion-labels-train.csv') \n",
    "data_test = pd.read_csv('./emotions_data/emotion-labels-test.csv')\n",
    "data_val = pd.read_csv('./emotions_data/emotion-labels-val.csv')\n",
    "# data should be saved in a folder called 'emotions' which is saved in the same place as your notebook"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "aca9dc9d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Just got back from seeing @GaryDelaney in Burs...</td>\n",
       "      <td>joy</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Oh dear an evening of absolute hilarity I don'...</td>\n",
       "      <td>joy</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Been waiting all week for this game ❤️❤️❤️ #ch...</td>\n",
       "      <td>joy</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>@gardiner_love : Thank you so much, Gloria! Yo...</td>\n",
       "      <td>joy</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>I feel so blessed to work with the family that...</td>\n",
       "      <td>joy</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                text label\n",
       "0  Just got back from seeing @GaryDelaney in Burs...   joy\n",
       "1  Oh dear an evening of absolute hilarity I don'...   joy\n",
       "2  Been waiting all week for this game ❤️❤️❤️ #ch...   joy\n",
       "3  @gardiner_love : Thank you so much, Gloria! Yo...   joy\n",
       "4  I feel so blessed to work with the family that...   joy"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "8f409679",
   "metadata": {},
   "outputs": [],
   "source": [
    "data = pd.concat([data_train, data_test, data_val], ignore_index=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "a63d4c5b",
   "metadata": {},
   "outputs": [],
   "source": [
    "data['text_clean'] = data['text'].apply(lambda x: clean(x, no_emoji=True))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "62216557",
   "metadata": {},
   "outputs": [],
   "source": [
    "data['text_clean'] = data['text_clean'].apply(lambda x: re.sub('@[^\\s]+', '', x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "34e0bc62",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>label</th>\n",
       "      <th>text_clean</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Just got back from seeing @GaryDelaney in Burs...</td>\n",
       "      <td>joy</td>\n",
       "      <td>just got back from seeing  in burslem. amazing...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Oh dear an evening of absolute hilarity I don'...</td>\n",
       "      <td>joy</td>\n",
       "      <td>oh dear an evening of absolute hilarity i don'...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Been waiting all week for this game ❤️❤️❤️ #ch...</td>\n",
       "      <td>joy</td>\n",
       "      <td>been waiting all week for this game #cheer #fr...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>@gardiner_love : Thank you so much, Gloria! Yo...</td>\n",
       "      <td>joy</td>\n",
       "      <td>: thank you so much, gloria! you're so sweet,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>I feel so blessed to work with the family that...</td>\n",
       "      <td>joy</td>\n",
       "      <td>i feel so blessed to work with the family that...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>Today I reached 1000 subscribers on YT!! , #go...</td>\n",
       "      <td>joy</td>\n",
       "      <td>today i reached 1000 subscribers on yt!! , #go...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>@Singaholic121 Good morning, love! Happy first...</td>\n",
       "      <td>joy</td>\n",
       "      <td>good morning, love! happy first day of fall. ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>#BridgetJonesBaby is the best thing I've seen ...</td>\n",
       "      <td>joy</td>\n",
       "      <td>#bridgetjonesbaby is the best thing i've seen ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>Just got back from seeing @GaryDelaney in Burs...</td>\n",
       "      <td>joy</td>\n",
       "      <td>just got back from seeing  in burslem. amazing...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>@IndyMN I thought the holidays could not get a...</td>\n",
       "      <td>joy</td>\n",
       "      <td>i thought the holidays could not get any more...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>I'm just still . So happy .\\nA blast</td>\n",
       "      <td>joy</td>\n",
       "      <td>i'm just still . so happy .\\na blast</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>It's meant to be!! #happy #happy</td>\n",
       "      <td>joy</td>\n",
       "      <td>it's meant to be!! #happy #happy</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>💥⚖️Yeah‼️ PAUL‼️⚖️💥  #glorious #BB18</td>\n",
       "      <td>joy</td>\n",
       "      <td>yeah paul #glorious #bb18</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>My morning started off amazing!! Hopefully the...</td>\n",
       "      <td>joy</td>\n",
       "      <td>my morning started off amazing!! hopefully the...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>😱 @cailamarsai you've had me 😂 😂 the whole tim...</td>\n",
       "      <td>joy</td>\n",
       "      <td>you've had me the whole time watching  after ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>@iamTinaDatta love you so much #smile 😊😊</td>\n",
       "      <td>joy</td>\n",
       "      <td>love you so much #smile</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>@WyoWiseGuy @LivingVertical however, REI did o...</td>\n",
       "      <td>joy</td>\n",
       "      <td>however, rei did offer me the job today as w...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>2 days until #GoPackGo and 23 days until #GoGi...</td>\n",
       "      <td>joy</td>\n",
       "      <td>2 days until #gopackgo and 23 days until #gogi...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>@TheMandyMoore You are beyond wonderful.  Your...</td>\n",
       "      <td>joy</td>\n",
       "      <td>you are beyond wonderful. your singing prowes...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>@luckiiCHARM_ Luckii, I'm changing in so many ...</td>\n",
       "      <td>joy</td>\n",
       "      <td>luckii, i'm changing in so many ways bc of hi...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                 text label   \n",
       "0   Just got back from seeing @GaryDelaney in Burs...   joy  \\\n",
       "1   Oh dear an evening of absolute hilarity I don'...   joy   \n",
       "2   Been waiting all week for this game ❤️❤️❤️ #ch...   joy   \n",
       "3   @gardiner_love : Thank you so much, Gloria! Yo...   joy   \n",
       "4   I feel so blessed to work with the family that...   joy   \n",
       "5   Today I reached 1000 subscribers on YT!! , #go...   joy   \n",
       "6   @Singaholic121 Good morning, love! Happy first...   joy   \n",
       "7   #BridgetJonesBaby is the best thing I've seen ...   joy   \n",
       "8   Just got back from seeing @GaryDelaney in Burs...   joy   \n",
       "9   @IndyMN I thought the holidays could not get a...   joy   \n",
       "10               I'm just still . So happy .\\nA blast   joy   \n",
       "11                   It's meant to be!! #happy #happy   joy   \n",
       "12               💥⚖️Yeah‼️ PAUL‼️⚖️💥  #glorious #BB18   joy   \n",
       "13  My morning started off amazing!! Hopefully the...   joy   \n",
       "14  😱 @cailamarsai you've had me 😂 😂 the whole tim...   joy   \n",
       "15           @iamTinaDatta love you so much #smile 😊😊   joy   \n",
       "16  @WyoWiseGuy @LivingVertical however, REI did o...   joy   \n",
       "17  2 days until #GoPackGo and 23 days until #GoGi...   joy   \n",
       "18  @TheMandyMoore You are beyond wonderful.  Your...   joy   \n",
       "19  @luckiiCHARM_ Luckii, I'm changing in so many ...   joy   \n",
       "\n",
       "                                           text_clean  \n",
       "0   just got back from seeing  in burslem. amazing...  \n",
       "1   oh dear an evening of absolute hilarity i don'...  \n",
       "2   been waiting all week for this game #cheer #fr...  \n",
       "3    : thank you so much, gloria! you're so sweet,...  \n",
       "4   i feel so blessed to work with the family that...  \n",
       "5   today i reached 1000 subscribers on yt!! , #go...  \n",
       "6    good morning, love! happy first day of fall. ...  \n",
       "7   #bridgetjonesbaby is the best thing i've seen ...  \n",
       "8   just got back from seeing  in burslem. amazing...  \n",
       "9    i thought the holidays could not get any more...  \n",
       "10               i'm just still . so happy .\\na blast  \n",
       "11                   it's meant to be!! #happy #happy  \n",
       "12                          yeah paul #glorious #bb18  \n",
       "13  my morning started off amazing!! hopefully the...  \n",
       "14   you've had me the whole time watching  after ...  \n",
       "15                            love you so much #smile  \n",
       "16    however, rei did offer me the job today as w...  \n",
       "17  2 days until #gopackgo and 23 days until #gogi...  \n",
       "18   you are beyond wonderful. your singing prowes...  \n",
       "19   luckii, i'm changing in so many ways bc of hi...  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.head(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "27c8f479",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: xlabel='label'>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "data['label'].value_counts().plot(kind=\"bar\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "e73d41d2",
   "metadata": {},
   "outputs": [],
   "source": [
    "g = data.groupby('label')\n",
    "data = pd.DataFrame(g.apply(lambda x: x.sample(g.size().min()).reset_index(drop=True)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "e67b4aa0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: xlabel='label'>"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "data['label'].value_counts().plot(kind=\"bar\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "7e0d50f2",
   "metadata": {},
   "outputs": [],
   "source": [
    "data['label_int'] = LabelEncoder().fit_transform(data['label'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "46267f59",
   "metadata": {},
   "outputs": [],
   "source": [
    "NUM_LABELS = 4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "7ca815da",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_split, test_split = train_test_split(data, train_size = 0.8)\n",
    "train_split, val_split = train_test_split(train_split, train_size = 0.9)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "d3b8eafc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4414\n",
      "1227\n",
      "491\n"
     ]
    }
   ],
   "source": [
    "print(len(train_split))\n",
    "print(len(test_split))\n",
    "print(len(val_split))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "48d59fda",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_df = pd.DataFrame({\n",
    "    \"label\": train_split.label_int.values,\n",
    "    \"text\": train_split.text_clean.values\n",
    "})\n",
    "\n",
    "test_df = pd.DataFrame({\n",
    "    \"label\": test_split.label_int.values,\n",
    "    \"text\": test_split.text_clean.values\n",
    "})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "e21c9547",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_df = datasets.Dataset.from_dict(train_df)\n",
    "test_df = datasets.Dataset.from_dict(test_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "c183fb53",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset_dict = datasets.DatasetDict({\"train\":train_df, \"test\":test_df})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "ef1f934f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    train: Dataset({\n",
       "        features: ['label', 'text'],\n",
       "        num_rows: 4414\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['label', 'text'],\n",
       "        num_rows: 1227\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset_dict"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e3fb73ff",
   "metadata": {},
   "source": [
    "## Create embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "824ffe45",
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenizer = XLNetTokenizer.from_pretrained(\"xlnet-base-cased\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "f9b1f421",
   "metadata": {},
   "outputs": [],
   "source": [
    "def tokenize_function(examples):\n",
    "    return tokenizer(examples[\"text\"], padding = \"max_length\", max_length = 128, truncation=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "2aae982b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8a8a2311c5f14908b7bfbc7ce38ef340",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Map:   0%|          | 0/4414 [00:00<?, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "032f672036b2423c8e2ef6c8ee9fa95a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Map:   0%|          | 0/1227 [00:00<?, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "tokenized_datasets = dataset_dict.map(tokenize_function, batched=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "866efe5a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    train: Dataset({\n",
       "        features: ['label', 'text', 'input_ids', 'token_type_ids', 'attention_mask'],\n",
       "        num_rows: 4414\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['label', 'text', 'input_ids', 'token_type_ids', 'attention_mask'],\n",
       "        num_rows: 1227\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokenized_datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "12e2c20a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " you've had me the whole time watching  after you've lost your #glasses! it was #hilarious! \n"
     ]
    }
   ],
   "source": [
    "print(tokenized_datasets['train']['text'][0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "6b1e74ac",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 44, 26, 189, 54, 110, 18, 856, 92, 2441, 99, 44, 26, 189, 581, 73, 17, 7967, 15275, 202, 136, 36, 30, 17, 7967, 11729, 617, 4672, 136, 4, 3]\n"
     ]
    }
   ],
   "source": [
    "print(tokenized_datasets['train']['input_ids'][0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "137ca077",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'<pad>'"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokenizer.decode(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "da3ba5d8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 2]\n"
     ]
    }
   ],
   "source": [
    "print(tokenized_datasets['train']['token_type_ids'][0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "9fc5d0c2",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]\n"
     ]
    }
   ],
   "source": [
    "print(tokenized_datasets['train']['attention_mask'][0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "8c57bac1",
   "metadata": {},
   "outputs": [],
   "source": [
    "small_train_dataset = tokenized_datasets[\"train\"].shuffle(seed=42).select(range(100))\n",
    "small_eval_dataset = tokenized_datasets[\"test\"].shuffle(seed=42).select(range(100))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9263042e",
   "metadata": {},
   "source": [
    "## Fine tune our model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "e7281188",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Some weights of XLNetForSequenceClassification were not initialized from the model checkpoint at xlnet-base-cased and are newly initialized: ['logits_proj.bias', 'logits_proj.weight', 'sequence_summary.summary.bias', 'sequence_summary.summary.weight']\n",
      "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
     ]
    }
   ],
   "source": [
    "model = XLNetForSequenceClassification.from_pretrained('xlnet-base-cased', \n",
    "                                                       num_labels=NUM_LABELS, \n",
    "                                                       id2label={0: 'anger', 1: 'fear', 2: 'joy', 3: 'sadness'})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "752e80bb",
   "metadata": {},
   "outputs": [],
   "source": [
    "metric = evaluate.load(\"accuracy\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "59ff2579",
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_metrics(eval_pred):\n",
    "    logits, labels = eval_pred\n",
    "    predictions = np.argmax(logits, axis=-1)\n",
    "    return metric.compute(predictions=predictions, references=labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "305ae6fa",
   "metadata": {},
   "outputs": [],
   "source": [
    "training_args = TrainingArguments(output_dir=\"test_trainer\", eval_strategy=\"epoch\", num_train_epochs=3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "ce63ef3e",
   "metadata": {},
   "outputs": [],
   "source": [
    "trainer = Trainer(\n",
    "    model=model, \n",
    "    args=training_args,\n",
    "    train_dataset=small_train_dataset,\n",
    "    eval_dataset=small_eval_dataset,\n",
    "    compute_metrics=compute_metrics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "e9548d91",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "      \n",
       "      <progress value='39' max='39' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      [39/39 04:05, Epoch 3/3]\n",
       "    </div>\n",
       "    <table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       " <tr style=\"text-align: left;\">\n",
       "      <th>Epoch</th>\n",
       "      <th>Training Loss</th>\n",
       "      <th>Validation Loss</th>\n",
       "      <th>Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>No log</td>\n",
       "      <td>1.476399</td>\n",
       "      <td>0.260000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>No log</td>\n",
       "      <td>1.450392</td>\n",
       "      <td>0.230000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>No log</td>\n",
       "      <td>1.440396</td>\n",
       "      <td>0.230000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table><p>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "TrainOutput(global_step=39, training_loss=1.403076660938752, metrics={'train_runtime': 250.6321, 'train_samples_per_second': 1.197, 'train_steps_per_second': 0.156, 'total_flos': 21366375321600.0, 'train_loss': 1.403076660938752, 'epoch': 3.0})"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainer.train()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d7bca737",
   "metadata": {},
   "source": [
    "## Evaluate model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "89327277",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "      \n",
       "      <progress value='13' max='13' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      [13/13 00:18]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "{'eval_loss': 1.4403964281082153,\n",
       " 'eval_accuracy': 0.23,\n",
       " 'eval_runtime': 20.394,\n",
       " 'eval_samples_per_second': 4.903,\n",
       " 'eval_steps_per_second': 0.637,\n",
       " 'epoch': 3.0}"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainer.evaluate()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "b650e62a",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save_pretrained(\"fine_tuned_model\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "13d9f454",
   "metadata": {},
   "outputs": [],
   "source": [
    "fine_tuned_model = XLNetForSequenceClassification.from_pretrained(\"fine_tuned_model\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "452009f7",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Device set to use cpu\n"
     ]
    }
   ],
   "source": [
    "clf = pipeline(\"text-classification\", fine_tuned_model, tokenizer=tokenizer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "fd4d6981",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "you dont have to feel grateful to be grateful, for it is written ' all things work together for good' for those who love god. #cheerful\n",
      "[{'label': 'sadness', 'score': 0.37114980816841125}, {'label': 'anger', 'score': 0.23535294830799103}, {'label': 'fear', 'score': 0.2268558293581009}, {'label': 'joy', 'score': 0.16664142906665802}]\n"
     ]
    }
   ],
   "source": [
    "rand_int = random.randint(0, len(val_split))\n",
    "print(val_split['text_clean'][rand_int])\n",
    "answer = clf(val_split['text_clean'][rand_int], top_k=None)\n",
    "print(answer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "eb30b275",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "llms_course_env",
   "language": "python",
   "name": "llms_course_env"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}