Predicting Heart Failure Readmission from Clinical Notes Using Deep Learning 논문 구현.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "from sklearn.model_selection import train_test_split, cross_val_score\n",
    "from sklearn.metrics import precision_score, recall_score, f1_score, accuracy_score\n",
    "import warnings\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "from gensim.models import KeyedVectors\n",
    "from torch.utils.data import DataLoader, Dataset\n",
    "import numpy as np\n",
    "import re\n",
    "import matplotlib.pyplot as plt\n",
    "from tqdm import tqdm\n",
    "import math\n",
    "import nltk\n",
    "from nltk.corpus import stopwords\n",
    "from scipy.stats import chi2_contingency\n",
    "from collections import Counter\n",
    "\n",
    "nltk.download('stopwords')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "note = pd.read_parquet(\"discharge_summaries.parquet\")\n",
    "d_icd_diagnoses = pd.read_parquet(\"d_icd_diagnoses.parquet\")\n",
    "icustays = pd.read_parquet(\"icustays.parquet\")\n",
    "chartevents = pd.read_parquet(\"chartevents.parquet\")\n",
    "d_items = pd.read_parquet(\"d_items.parquet\")\n",
    "diagnoses_icd = pd.read_parquet(\"diagnoses_icd.parquet\")\n",
    "admission = pd.read_parquet(\"admissions.parquet\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "icd_10 = d_icd_diagnoses[d_icd_diagnoses['icd_version']==10]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- I5040\tUnspecified combined systolic (congestive) and diastolic (congestive) heart failure\n",
    "- I5041\tAcute combined systolic (congestive) and diastolic (congestive) heart failure\n",
    "- I5042\tChronic combined systolic (congestive) and diastolic (congestive) heart failure\n",
    "- I5043\tAcute on chronic combined systolic (congestive) and diastolic (congestive) heart failure\n",
    "- I50810\tRight heart failure, unspecified\n",
    "- I50811\tAcute right heart failure\n",
    "- I50812\tChronic right heart failure\n",
    "- I50813\tAcute on chronic right heart failure\n",
    "- I50814\tRight heart failure due to left heart failure\n",
    "- I5082\tBiventricular heart failure\n",
    "- I5083\tHigh output heart failure\n",
    "- I5084\tEnd stage heart failure\n",
    "- I5089\tOther heart failure\n",
    "- I509\tHeart failure, unspecified\n",
    "- R570\tCardiogenic shock\n",
    "- R579\tShock, unspecified"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "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>icd_code</th>\n",
       "      <th>icd_version</th>\n",
       "      <th>long_title</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>24793</th>\n",
       "      <td>I5040</td>\n",
       "      <td>10</td>\n",
       "      <td>Unspecified combined systolic (congestive) and...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      icd_code  icd_version                                         long_title\n",
       "24793    I5040           10  Unspecified combined systolic (congestive) and..."
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "icd_10[icd_10['icd_code'].str.startswith('I5040')]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "heart_failure_code = ['I5040','I5041','I5042','I5043','I50810','I50811','I50812','I50813','I50814','I5082','I5083','I5084','I5089','I509','R570','R579']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "heart_failure_df = icd_10[icd_10['icd_code'].isin(heart_failure_code)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "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>subject_id</th>\n",
       "      <th>hadm_id</th>\n",
       "      <th>seq_num</th>\n",
       "      <th>icd_code</th>\n",
       "      <th>icd_version</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>10000032</td>\n",
       "      <td>22595853</td>\n",
       "      <td>1</td>\n",
       "      <td>5723</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>10000032</td>\n",
       "      <td>22595853</td>\n",
       "      <td>2</td>\n",
       "      <td>78959</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>10000032</td>\n",
       "      <td>22595853</td>\n",
       "      <td>3</td>\n",
       "      <td>5715</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>10000032</td>\n",
       "      <td>22595853</td>\n",
       "      <td>4</td>\n",
       "      <td>07070</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>10000032</td>\n",
       "      <td>22595853</td>\n",
       "      <td>5</td>\n",
       "      <td>496</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4756321</th>\n",
       "      <td>19999987</td>\n",
       "      <td>23865745</td>\n",
       "      <td>7</td>\n",
       "      <td>41401</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4756322</th>\n",
       "      <td>19999987</td>\n",
       "      <td>23865745</td>\n",
       "      <td>8</td>\n",
       "      <td>78039</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4756323</th>\n",
       "      <td>19999987</td>\n",
       "      <td>23865745</td>\n",
       "      <td>9</td>\n",
       "      <td>0413</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4756324</th>\n",
       "      <td>19999987</td>\n",
       "      <td>23865745</td>\n",
       "      <td>10</td>\n",
       "      <td>36846</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4756325</th>\n",
       "      <td>19999987</td>\n",
       "      <td>23865745</td>\n",
       "      <td>11</td>\n",
       "      <td>7810</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>4756326 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "         subject_id   hadm_id  seq_num icd_code  icd_version\n",
       "0          10000032  22595853        1     5723            9\n",
       "1          10000032  22595853        2    78959            9\n",
       "2          10000032  22595853        3     5715            9\n",
       "3          10000032  22595853        4    07070            9\n",
       "4          10000032  22595853        5      496            9\n",
       "...             ...       ...      ...      ...          ...\n",
       "4756321    19999987  23865745        7    41401            9\n",
       "4756322    19999987  23865745        8    78039            9\n",
       "4756323    19999987  23865745        9     0413            9\n",
       "4756324    19999987  23865745       10    36846            9\n",
       "4756325    19999987  23865745       11     7810            9\n",
       "\n",
       "[4756326 rows x 5 columns]"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "diagnoses_icd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "hf_patient = diagnoses_icd[diagnoses_icd['icd_code'].isin(heart_failure_code)][['subject_id', 'hadm_id']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "hf_note = note.merge(hf_patient,how='right', on=['subject_id', 'hadm_id'])\n",
    "hf_note['text_length'] = hf_note['text'].apply(lambda x: len(x) if pd.notna(x) else 0)\n",
    "\n",
    "# 'subject_id', 'hadm_id', 'note_id가 동일한 경우에는 text length가 가장 긴 행만 남김\n",
    "hf_note_max_text = hf_note.loc[hf_note.groupby(['subject_id', 'hadm_id', 'note_id'])['text_length'].idxmax()]\n",
    "hf_note_max_text = hf_note_max_text.drop(columns=['text_length'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "admission_time = admission[['subject_id', 'hadm_id', 'dischtime', 'deathtime']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "hf_note_max_text= hf_note_max_text.merge(admission_time, how='left', on=['subject_id', 'hadm_id'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "hf_note_max_text['dischtime'] = pd.to_datetime(hf_note_max_text['dischtime'])\n",
    "admission['admittime'] = pd.to_datetime(admission['admittime'])\n",
    "admission['dischtime'] = pd.to_datetime(admission['dischtime'])\n",
    "hf_note_max_text['label'] = 0\n",
    "\n",
    "admission_general = hf_note_max_text.copy()\n",
    "admission_30days = hf_note_max_text.copy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 단순히 readmission이 있기만 하면 1로 라벨링함 / 논문에 따르면 scenario A\n",
    "for idx, note_row in admission_general.iterrows():\n",
    "    subject_id = note_row['subject_id']\n",
    "    dischtime = note_row['dischtime']\n",
    "    \n",
    "    general_admissions = admission[admission['subject_id'] == subject_id]\n",
    "    within_admission = general_admissions[\n",
    "        (general_admissions['admittime'] > dischtime)]\n",
    "    \n",
    "    if not within_admission.empty:\n",
    "        admission_general.at[idx, 'label'] = 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 30일 이내에 readmission 기록이 있는 경우에만 라벨링함 / 논문에 따르면 scenario B\n",
    "for idx, note_row in admission_30days.iterrows():\n",
    "    subject_id = note_row['subject_id']\n",
    "    dischtime = note_row['dischtime']\n",
    "    \n",
    "    admissions_30days = admission[admission['subject_id'] == subject_id]\n",
    "    within_30_days = admissions_30days[\n",
    "        (admissions_30days['admittime'] > dischtime) & \n",
    "        (admissions_30days['admittime'] <= dischtime + pd.Timedelta(days=30))]\n",
    "    \n",
    "    if not within_30_days.empty:\n",
    "        admission_30days.at[idx, 'label'] = 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "단순히 readmission한 환자:  2138\n",
      "30일 이내 readmission한 환자:  889\n"
     ]
    }
   ],
   "source": [
    "print(\"단순히 readmission한 환자: \", len(admission_general[admission_general['label']==1]))\n",
    "print(\"30일 이내 readmission한 환자: \", len(admission_30days[admission_30days['label']==1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "label\n",
       "0    2507\n",
       "1    2138\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "admission_general['label'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "label\n",
       "0    3756\n",
       "1     889\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "admission_30days['label'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "pos_general_num = len(admission_general[admission_general['label']==1])\n",
    "pos_30days_num = len(admission_30days[admission_30days['label']==1])\n",
    "pos_general = admission_general[admission_general['label']==1]\n",
    "pos_30days = admission_30days[admission_30days['label']==1]\n",
    "neg_general = admission_general[admission_general['label']==0].sample(n=pos_general_num)\n",
    "neg_30days = admission_30days[admission_30days['label']==0].sample(n=pos_30days_num)\n",
    "\n",
    "general = pd.concat([pos_general, neg_general])\n",
    "days_30 = pd.concat([pos_30days, neg_30days])\n",
    "\n",
    "general = general[['text', 'label']].copy()\n",
    "days_30 = days_30[['text', 'label']].copy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Text Preprocessing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "tqdm.pandas()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4276/4276 [00:21<00:00, 203.31it/s]\n",
      "100%|██████████| 1778/1778 [00:09<00:00, 196.46it/s]\n"
     ]
    }
   ],
   "source": [
    "def preprocess_text(text) : \n",
    "\n",
    "        # text = re.sub(r'[^a-zA-Z0-9\\s]', '', text)\n",
    "        text = text.lower()\n",
    "        text = re.sub(r'[^a-zA-Z0-9\\s\\.\\n]', '', text)\n",
    "        text = re.sub(r'\\s+', ' ', text) # 연속된 공백을 하나의 공백으로 변환\n",
    "        text = text.strip() # 양쪽 끝의 공백 제거\n",
    "\n",
    "        # 익명화되어 알 수 없는 개인 정보 단어 삭제\n",
    "        anony_words = ['name', 'unit', 'no', 'admission date', 'discharge date', 'date of birth', 'date', 'birth', 'attending', 'followup instructions', 'sex', 'service', 'chief complaint', 'major surgical or invasive procedure', 'history of present illness', 'past medical history', 'allergies',\n",
    "                       'medications', 'social history', 'physical examination','laboratory data', 'hospital course', 'discharge instructions','discharge medications', 'discharge condition', 'diagnosis', '.']\n",
    "        for word in anony_words:\n",
    "                text = re.sub(r'\\b' + re.escape(word) + r'\\b', '', text)\n",
    "\n",
    "        # 숫자 제거\n",
    "        text = re.sub(r'\\d+', '', text)\n",
    "\n",
    "        # 불용어 제거\n",
    "        stop_words = list(stopwords.words('english'))\n",
    "        text_words = text.split()\n",
    "        text_words = [word for word in text_words if word not in stop_words]\n",
    "        text = ' '.join(text_words)\n",
    "\n",
    "        return text\n",
    "\n",
    "general['text'] = general['text'].progress_apply(preprocess_text)\n",
    "days_30['text'] = days_30['text'].progress_apply(preprocess_text)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "9417.702993451825"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "general['text_length'] = general['text'].apply(len)\n",
    "average_length = general['text_length'].mean()\n",
    "average_length"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "최대 글자수:  38855\n",
      "최소 글자수:  182\n"
     ]
    }
   ],
   "source": [
    "print('최대 글자수: ', general['text_length'].max())\n",
    "print('최소 글자수: ', general['text_length'].min())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAkcAAAHFCAYAAAD40125AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjguNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8fJSN1AAAACXBIWXMAAA9hAAAPYQGoP6dpAAA76ElEQVR4nO3de1RVdf7/8ddR4HARGYTkSIJR4i3UUsq0i3jP1C42YzfNyvpaliOFq+9Y359SOWI2mo19u455naKZyqZvV7XUdNRJMUvNqFFHUEEGIi4KB4TP74+WZ/YRTDgBBw7Px1pnLc/e77N5f87OfLH3Z+9tM8YYAQAAQJLUxtsNAAAANCeEIwAAAAvCEQAAgAXhCAAAwIJwBAAAYEE4AgAAsCAcAQAAWBCOAAAALAhHAAAAFoQjAGhgSUlJSkpK8nYbADxEOAIAALAgHAHAzzDGqKyszNttAGhChCMATepvf/ub+vTpI7vdrgsvvFDPPfecUlNTZbPZXDXGGL3wwgu65JJLFBQUpPDwcP3617/WwYMH3baVlJSkhIQE7dixQ1dffbWCg4N14YUXav78+aqurnarLS4u1syZMxUXF6eAgACdf/75Sk5O1okTJ9zqbDabHnroIb300kvq2bOn7Ha7VqxYIUl64oknNGDAAHXo0EHt27dXv379tHTpUvH8bsC3+Hm7AQCtx8cff6zx48frmmuu0ZtvvqlTp07pD3/4g44fP+5WN3XqVC1fvly//e1v9fTTT+uHH37Qk08+qUGDBumrr75SVFSUqzY3N1d33HGHUlJSNGfOHK1Zs0azZs1SdHS07rzzTknSyZMnNXjwYB05ckSPPfaY+vTpo3379mn27Nnas2eP1q9f7xbO3n33XW3evFmzZ8+Ww+FQx44dJUn/+te/NHXqVMXGxkqStm/frunTp+vo0aOaPXt2Y399AJqKAYAmctlll5mYmBjjdDpdy0pKSkxERIQ5/b+jbdu2GUlm4cKFbp/Nzs42QUFB5tFHH3UtGzx4sJFk/vGPf7jV9urVy4waNcr1Pi0tzbRp08bs2LHDre6tt94yksyHH37oWibJhIWFmR9++OFnx1JVVWUqKyvNk08+aSIiIkx1dbVbX4MHDz7HtwGgueK0GoAmceLECe3cuVM33nijAgICXMvbtWuncePGud6///77stlsmjhxok6dOuV6ORwO9e3bVxs3bnTbrsPh0OWXX+62rE+fPjp8+LDbNhMSEnTJJZe4bXPUqFGy2Ww1tjl06FCFh4fXGMNnn32m4cOHKywsTG3btpW/v79mz56tgoIC5eXl/YJvB0Bzwmk1AE2isLBQxhi3U2KnWZcdP378rHWSdOGFF7q9j4iIqFFjt9vdJlEfP35c//znP+Xv71/rNvPz893ed+rUqUbNF198oZEjRyopKUmvvvqqOnfurICAAL377rv6/e9/z6RtwIcQjgA0ifDwcNlsthrzi6Sf5g2dFhkZKZvNps2bN8tut9eorW3ZuURGRiooKEivvfbaWddbWecfnZaeni5/f3+9//77CgwMdC1/9913690PgOaNcASgSYSEhCgxMVHvvvuu/vCHP7hOrZWWlur999931Y0dO1bz58/X0aNHNWHChAb52WPHjtW8efMUERGhuLg4j7Zhs9nk5+entm3bupaVlZVp1apVDdIjgOaDcASgyTz55JMaM2aMRo0apRkzZqiqqkrPPPOM2rVrpx9++EGSdOWVV+q//uu/dPfdd2vnzp265pprFBISopycHG3ZskW9e/fWAw88UK+fm5ycrLffflvXXHONHn74YfXp00fV1dXKysrS2rVrlZKSogEDBvzsNsaMGaNFixbp9ttv13/913+poKBAf/jDHzw6kgWgeSMcAWgy1157rd5++23Nnj1bt9xyixwOh6ZNm6Zjx465HYF5+eWXdcUVV+jll1/WCy+8oOrqakVHR+vKK6+sMfm6LkJCQrR582bNnz9fr7zyig4dOqSgoCDFxsZq+PDhuuCCC865jaFDh+q1117T008/rXHjxun888/Xfffdp44dO2rKlCn17glA82UzhruXAfCeyspKXXLJJTr//PO1du1ab7cDABw5AtC0pkyZohEjRqhTp07Kzc3VSy+9pP379+u5557zdmsAIIlwBKCJlZSUaObMmfr3v/8tf39/9evXTx9++KGGDx/u7dYAQBKn1QAAANxwh2wAAAALwhEAAIAF4QgAAMCCCdmSqqurdezYMYWGhtb62AAAAND8GGNUUlKi6OhotWnTcMd7CEeSjh07ppiYGG+3AQAAPJCdna3OnTs32PYIR5JCQ0Ml/fTltm/f3svdAACAuiguLlZMTIzr3/GGQjjSf57A3b59e8IRAAAtTENPiWFCNgAAgAXhCAAAwIJwBAAAYEE4AgAAsCAcAQAAWBCOAAAALAhHAAAAFoQjAAAAC8IRAACABeEIAADAgnAEAABgQTgCAACwIBwBAABYEI4AAAAs/LzdAHAuWVlZys/Pr3N9ZGSkYmNjG7EjAIAvIxyhWcvKylL3Hj1VXnayzp8JDApW5rf7CUgAAI8QjtCs5efnq7zspCLGpsg/Iuac9ZUF2Sp4f6Hy8/MJRwAAjxCO0CL4R8TI7ujq7TYAAK0AE7IBAAAsvBqOUlNTZbPZ3F4Oh8O13hij1NRURUdHKygoSElJSdq3b5/bNpxOp6ZPn67IyEiFhITo+uuv15EjR5p6KAAAwEd4/cjRxRdfrJycHNdrz549rnULFizQokWL9Pzzz2vHjh1yOBwaMWKESkpKXDXJyclas2aN0tPTtWXLFpWWlmrs2LGqqqryxnAAAEAL5/U5R35+fm5Hi04zxmjx4sV6/PHHNX78eEnSihUrFBUVpddff11Tp05VUVGRli5dqlWrVmn48OGSpNWrVysmJkbr16/XqFGjmnQsAACg5fP6kaPvv/9e0dHRiouL06233qqDBw9Kkg4dOqTc3FyNHDnSVWu32zV48GBt3bpVkpSRkaHKykq3mujoaCUkJLhqAAAA6sOrR44GDBiglStXqlu3bjp+/Ljmzp2rQYMGad++fcrNzZUkRUVFuX0mKipKhw8fliTl5uYqICBA4eHhNWpOf742TqdTTqfT9b64uLihhgQAAFo4r4aj0aNHu/7cu3dvDRw4UBdddJFWrFihK664QpJks9ncPmOMqbHsTOeqSUtL0xNPPPELOgcAAL7K66fVrEJCQtS7d299//33rnlIZx4BysvLcx1NcjgcqqioUGFh4VlrajNr1iwVFRW5XtnZ2Q08EgAA0FI1q3DkdDq1f/9+derUSXFxcXI4HFq3bp1rfUVFhTZt2qRBgwZJkvr37y9/f3+3mpycHO3du9dVUxu73a727du7vQAAACQvn1abOXOmxo0bp9jYWOXl5Wnu3LkqLi7W5MmTZbPZlJycrHnz5ik+Pl7x8fGaN2+egoODdfvtt0uSwsLCNGXKFKWkpCgiIkIdOnTQzJkz1bt3b9fVawAAAPXh1XB05MgR3XbbbcrPz9d5552nK664Qtu3b1eXLl0kSY8++qjKyso0bdo0FRYWasCAAVq7dq1CQ0Nd23j22Wfl5+enCRMmqKysTMOGDdPy5cvVtm1bbw0LAAC0YDZjjPF2E95WXFyssLAwFRUVcYqtmdm1a5f69+8vx+TFdXq2mjP3n8pdkayMjAz169evCToEAHhLY/373azmHAEAAHgb4QgAAMCCcAQAAGBBOAIAALAgHAEAAFgQjgAAACwIRwAAABaEIwAAAAvCEQAAgAXhCAAAwIJwBAAAYEE4AgAAsCAcAQAAWBCOAAAALAhHAAAAFn7ebgBoDPv3769zbWRkpGJjYxuxGwBAS0I4gk+pKi2UbDZNnDixzp8JDApW5rf7CUgAAEmEI/iYamepZIwixqbIPyLmnPWVBdkqeH+h8vPzCUcAAEmEI/go/4gY2R1dvd0GAKAFYkI2AACABeEIAADAgnAEAABgQTgCAACwIBwBAABYcLUaoLrfNJIbRgKA7yMcoVWr700juWEkAPg+whFatfrcNJIbRgJA60A4AsRNIwEA/8GEbAAAAAvCEQAAgAXhCAAAwIJwBAAAYEE4AgAAsCAcAQAAWBCOAAAALAhHAAAAFoQjAAAAC+6QDa/IyspSfn7+Oevq+kBYAAAaCuEITS4rK0vde/RUedlJb7cCAEANhCM0ufz8fJWXnazTw17LDu5U0ebVTdQZAACEI3hRXR72WlmQ3UTdAADwEyZkAwAAWBCOAAAALAhHAAAAFoQjAAAAC8IRAACABeEIAADAgnAEAABgQTgCAACwIBwBAABYEI4AAAAsCEcAAAAWhCMAAAALwhEAAIAF4QgAAMCCcAQAAGBBOAIAALAgHAEAAFg0m3CUlpYmm82m5ORk1zJjjFJTUxUdHa2goCAlJSVp3759bp9zOp2aPn26IiMjFRISouuvv15Hjhxp4u4BAICvaBbhaMeOHXrllVfUp08ft+ULFizQokWL9Pzzz2vHjh1yOBwaMWKESkpKXDXJyclas2aN0tPTtWXLFpWWlmrs2LGqqqpq6mEAAAAf4PVwVFpaqjvuuEOvvvqqwsPDXcuNMVq8eLEef/xxjR8/XgkJCVqxYoVOnjyp119/XZJUVFSkpUuXauHChRo+fLguvfRSrV69Wnv27NH69eu9NSQAANCCeT0cPfjggxozZoyGDx/utvzQoUPKzc3VyJEjXcvsdrsGDx6srVu3SpIyMjJUWVnpVhMdHa2EhARXTW2cTqeKi4vdXgAAAJLk580fnp6erl27dmnHjh011uXm5kqSoqKi3JZHRUXp8OHDrpqAgAC3I06na05/vjZpaWl64oknfmn7AADAB3ntyFF2drZmzJih1atXKzAw8Kx1NpvN7b0xpsayM52rZtasWSoqKnK9srOz69c8AADwWV4LRxkZGcrLy1P//v3l5+cnPz8/bdq0SX/84x/l5+fnOmJ05hGgvLw81zqHw6GKigoVFhaetaY2drtd7du3d3sBAABIXgxHw4YN0549e7R7927XKzExUXfccYd2796tCy+8UA6HQ+vWrXN9pqKiQps2bdKgQYMkSf3795e/v79bTU5Ojvbu3euqAQAAqA+vzTkKDQ1VQkKC27KQkBBFRES4licnJ2vevHmKj49XfHy85s2bp+DgYN1+++2SpLCwME2ZMkUpKSmKiIhQhw4dNHPmTPXu3bvGBG8AAIC68OqE7HN59NFHVVZWpmnTpqmwsFADBgzQ2rVrFRoa6qp59tln5efnpwkTJqisrEzDhg3T8uXL1bZtWy923vpkZWUpPz+/TrX79+9v5G4AAPBcswpHGzdudHtvs9mUmpqq1NTUs34mMDBQS5Ys0ZIlSxq3OZxVVlaWuvfoqfKyk95uBQCAX6xZhSO0TPn5+SovO6mIsSnyj4g5Z33ZwZ0q2ry6CToDAKD+CEdoMP4RMbI7up6zrrKAWycAAJovr98hGwAAoDkhHAEAAFgQjgAAACyYcwTUU31uRRAZGanY2NhG7AYA0NAIR0AdVZUWSjabJk6cWOfPBAYFK/Pb/QQkAGhBCEdAHVU7SyVj6nzLgsqCbBW8v1D5+fmEIwBoQQhHQD3V9ZYFAICWiQnZAAAAFoQjAAAAC8IRAACABeEIAADAgnAEAABgQTgCAACwIBwBAABYEI4AAAAsCEcAAAAWhCMAAAALwhEAAIAF4QgAAMCCcAQAAGBBOAIAALAgHAEAAFgQjgAAACwIRwAAABaEIwAAAAvCEQAAgAXhCAAAwIJwBAAAYEE4AgAAsCAcAQAAWBCOAAAALAhHAAAAFoQjAAAAC8IRAACABeEIAADAgnAEAABgQTgCAACwIBwBAABYEI4AAAAsCEcAAAAWhCMAAAALwhEAAIAF4QgAAMCCcAQAAGBBOAIAALAgHAEAAFgQjgAAACwIRwAAABYehaNDhw41dB8AAADNgkfhqGvXrhoyZIhWr16t8vLyhu4JAADAazwKR1999ZUuvfRSpaSkyOFwaOrUqfriiy8aujcAAIAm51E4SkhI0KJFi3T06FEtW7ZMubm5uuqqq3TxxRdr0aJF+ve//93QfQIAADSJXzQh28/PTzfddJP+8pe/6Omnn9aBAwc0c+ZMde7cWXfeeadycnIaqk8AAIAm8YvC0c6dOzVt2jR16tRJixYt0syZM3XgwAF99tlnOnr0qG644YaG6hMAAKBJ+HnyoUWLFmnZsmXKzMzUddddp5UrV+q6665TmzY/Za24uDi9/PLL6tGjR4M2CwAA0Ng8Ckcvvvii7rnnHt19991yOBy11sTGxmrp0qW/qDkAAICm5tFpte+//16zZs06azCSpICAAE2ePPlnt/Piiy+qT58+at++vdq3b6+BAwfqo48+cq03xig1NVXR0dEKCgpSUlKS9u3b57YNp9Op6dOnKzIyUiEhIbr++ut15MgRT4YFAADgWThatmyZ/vrXv9ZY/te//lUrVqyo83Y6d+6s+fPna+fOndq5c6eGDh2qG264wRWAFixYoEWLFun555/Xjh075HA4NGLECJWUlLi2kZycrDVr1ig9PV1btmxRaWmpxo4dq6qqKk+GBgAAWjmPwtH8+fMVGRlZY3nHjh01b968Om9n3Lhxuu6669StWzd169ZNv//979WuXTtt375dxhgtXrxYjz/+uMaPH6+EhAStWLFCJ0+e1Ouvvy5JKioq0tKlS7Vw4UINHz5cl156qVavXq09e/Zo/fr1ngwNAAC0ch6Fo8OHDysuLq7G8i5duigrK8ujRqqqqpSenq4TJ05o4MCBOnTokHJzczVy5EhXjd1u1+DBg7V161ZJUkZGhiorK91qoqOjlZCQ4KqpjdPpVHFxsdsLAABA8jAcdezYUV9//XWN5V999ZUiIiLqta09e/aoXbt2stvtuv/++7VmzRr16tVLubm5kqSoqCi3+qioKNe63NxcBQQEKDw8/Kw1tUlLS1NYWJjrFRMTU6+eAQCA7/IoHN1666367W9/qw0bNqiqqkpVVVX67LPPNGPGDN1666312lb37t21e/dubd++XQ888IAmT56sb775xrXeZrO51Rtjaiw707lqZs2apaKiItcrOzu7Xj0DAADf5dGl/HPnztXhw4c1bNgw+fn9tInq6mrdeeed9ZpzJP10VVvXrl0lSYmJidqxY4eee+45/fd//7ekn44OderUyVWfl5fnOprkcDhUUVGhwsJCt6NHeXl5GjRo0Fl/pt1ul91ur1efAACgdfDoyFFAQIDefPNNffvtt/rzn/+sd955RwcOHNBrr72mgICAX9SQMUZOp1NxcXFyOBxat26da11FRYU2bdrkCj79+/eXv7+/W01OTo727t37s+EIAADgbDw6cnTa6avMPPXYY49p9OjRiomJUUlJidLT07Vx40Z9/PHHstlsSk5O1rx58xQfH6/4+HjNmzdPwcHBuv322yVJYWFhmjJlilJSUhQREaEOHTpo5syZ6t27t4YPH/5LhgYAAFopj8JRVVWVli9frk8//VR5eXmqrq52W//ZZ5/VaTvHjx/XpEmTlJOTo7CwMPXp00cff/yxRowYIUl69NFHVVZWpmnTpqmwsFADBgzQ2rVrFRoa6trGs88+Kz8/P02YMEFlZWUaNmyYli9frrZt23oyNAAA0Mp5FI5mzJih5cuXa8yYMUpISDjnBOmzOdfjRWw2m1JTU5WamnrWmsDAQC1ZskRLlizxqAcAAAArj8JRenq6/vKXv+i6665r6H4AAAC8yuMJ2aevMAMAAPAlHoWjlJQUPffcczLGNHQ/AAAAXuXRabUtW7Zow4YN+uijj3TxxRfL39/fbf0777zTIM0BAAA0NY/C0a9+9SvddNNNDd0LAACA13kUjpYtW9bQfQAAADQLHs05kqRTp05p/fr1evnll1VSUiJJOnbsmEpLSxusOQAAgKbm0ZGjw4cP69prr1VWVpacTqdGjBih0NBQLViwQOXl5XrppZcauk8AAIAm4dGRoxkzZigxMVGFhYUKCgpyLb/pppv06aefNlhzAAAATc3jq9X+/ve/13jIbJcuXXT06NEGaQwAAMAbPDpyVF1draqqqhrLjxw54vbcMwAAgJbGo3A0YsQILV682PXeZrOptLRUc+bM4ZEiAACgRfPotNqzzz6rIUOGqFevXiovL9ftt9+u77//XpGRkXrjjTcaukcAAIAm41E4io6O1u7du/XGG29o165dqq6u1pQpU3THHXe4TdAGIO3fv79OdZGRkYqNjW3kbgAA5+JROJKkoKAg3XPPPbrnnnsash/AZ1SVFko2myZOnFin+sCgYGV+u5+ABABe5lE4Wrly5c+uv/POOz1qBvAl1c5SyRhFjE2Rf0TMz9ZWFmSr4P2Fys/PJxwBgJd5FI5mzJjh9r6yslInT55UQECAgoODCUeAhX9EjOyOrt5uAwBQRx5drVZYWOj2Ki0tVWZmpq666iomZAMAgBbN42ernSk+Pl7z58+vcVQJAACgJfF4QnZt2rZtq2PHjjXkJuFFWVlZys/PP2ddXa/GAgCgJfAoHL333ntu740xysnJ0fPPP68rr7yyQRqDd2VlZal7j54qLzvp7VYAAGhSHoWjG2+80e29zWbTeeedp6FDh2rhwoUN0Re8LD8/X+VlJ+t0pVXZwZ0q2ry6iToDAKBxeRSOqqurG7oPNFN1udKqsiC7iboBAKDxNdiEbAAAAF/g0ZGjRx55pM61ixYt8uRHAAAAeIVH4ejLL7/Url27dOrUKXXv3l2S9N1336lt27bq16+fq85mszVMlwAAAE3Eo3A0btw4hYaGasWKFQoPD5f0040h7777bl199dVKSUlp0CYBAACaikdzjhYuXKi0tDRXMJKk8PBwzZ07l6vVAABAi+ZROCouLtbx48drLM/Ly1NJSckvbgoAAMBbPApHN910k+6++2699dZbOnLkiI4cOaK33npLU6ZM0fjx4xu6RwAAgCbj0Zyjl156STNnztTEiRNVWVn504b8/DRlyhQ988wzDdogAABAU/IoHAUHB+uFF17QM888owMHDsgYo65duyokJKSh+wMAAGhSv+gmkDk5OcrJyVG3bt0UEhIiY0xD9QUAAOAVHoWjgoICDRs2TN26ddN1112nnJwcSdK9997LZfwAAKBF8ygcPfzww/L391dWVpaCg4Ndy2+55RZ9/PHHDdYcAABAU/NoztHatWv1ySefqHPnzm7L4+Pjdfjw4QZpDGiN9u/fX+fayMhIxcbGNmI3ANA6eRSOTpw44XbE6LT8/HzZ7fZf3BTQ2lSVFko2myZOnFjnzwQGBSvz2/0EJABoYB6Fo2uuuUYrV67UU089JemnZ6hVV1frmWee0ZAhQxq0QaA1qHaWSsYoYmyK/CNizllfWZCtgvcXKj8/n3AEAA3Mo3D0zDPPKCkpSTt37lRFRYUeffRR7du3Tz/88IP+/ve/N3SPQKvhHxEju6Ort9sAgFbNownZvXr10tdff63LL79cI0aM0IkTJzR+/Hh9+eWXuuiiixq6RwAAgCZT7yNHlZWVGjlypF5++WU98cQTjdETAACA19T7yJG/v7/27t0rm83WGP0AAAB4lUen1e68804tXbq0oXsBAADwOo8mZFdUVOhPf/qT1q1bp8TExBrPVFu0aFGDNAcAANDU6hWODh48qAsuuEB79+5Vv379JEnfffedWw2n2wAAQEtWr3AUHx+vnJwcbdiwQdJPjwv54x//qKioqEZpDgAAoKnVa86RMcbt/UcffaQTJ040aEMAAADe5NGE7NPODEsAAAAtXb3Ckc1mqzGniDlGAADAl9RrzpExRnfddZfr4bLl5eW6//77a1yt9s477zRchwAAAE2oXuFo8uTJbu/r8wRxAACAlqBe4WjZsmWN1QcAAECz8IsmZAMAAPgawhEAAIAF4QgAAMCCcAQAAGBBOAIAALDwajhKS0vTZZddptDQUHXs2FE33nijMjMz3WqMMUpNTVV0dLSCgoKUlJSkffv2udU4nU5Nnz5dkZGRCgkJ0fXXX68jR4405VAAAICP8Go42rRpkx588EFt375d69at06lTpzRy5Ei357UtWLBAixYt0vPPP68dO3bI4XBoxIgRKikpcdUkJydrzZo1Sk9P15YtW1RaWqqxY8eqqqrKG8MCAAAtWL3uc9TQPv74Y7f3y5YtU8eOHZWRkaFrrrlGxhgtXrxYjz/+uMaPHy9JWrFihaKiovT6669r6tSpKioq0tKlS7Vq1SoNHz5ckrR69WrFxMRo/fr1GjVqVJOPCwAAtFzNas5RUVGRJKlDhw6SpEOHDik3N1cjR4501djtdg0ePFhbt26VJGVkZKiystKtJjo6WgkJCa6aMzmdThUXF7u9AAAApGYUjowxeuSRR3TVVVcpISFBkpSbmytJioqKcquNiopyrcvNzVVAQIDCw8PPWnOmtLQ0hYWFuV4xMTENPRwAANBCNZtw9NBDD+nrr7/WG2+8UWOdzWZze2+MqbHsTD9XM2vWLBUVFble2dnZnjcOAAB8SrMIR9OnT9d7772nDRs2qHPnzq7lDodDkmocAcrLy3MdTXI4HKqoqFBhYeFZa85kt9vVvn17txcAAIDk5XBkjNFDDz2kd955R5999pni4uLc1sfFxcnhcGjdunWuZRUVFdq0aZMGDRokSerfv7/8/f3danJycrR3715XDQAAQF159Wq1Bx98UK+//rr+9re/KTQ01HWEKCwsTEFBQbLZbEpOTta8efMUHx+v+Ph4zZs3T8HBwbr99ttdtVOmTFFKSooiIiLUoUMHzZw5U71793ZdvQb4qv3799epLjIyUrGxsY3cDQD4Bq+GoxdffFGSlJSU5LZ82bJluuuuuyRJjz76qMrKyjRt2jQVFhZqwIABWrt2rUJDQ131zz77rPz8/DRhwgSVlZVp2LBhWr58udq2bdtUQwGaVFVpoWSzaeLEiXWqDwwKVua3+wlIAFAHXg1Hxphz1thsNqWmpio1NfWsNYGBgVqyZImWLFnSgN0BzVe1s1QyRhFjU+Qf8fNXW1YWZKvg/YXKz88nHAFAHXg1HAH4ZfwjYmR3dPV2GwDgU5rF1WoAAADNBeEIAADAgnAEAABgQTgCAACwIBwBAABYEI4AAAAsCEcAAAAWhCMAAAALwhEAAIAF4QgAAMCCcAQAAGBBOAIAALAgHAEAAFgQjgAAACwIRwAAABaEIwAAAAvCEQAAgAXhCAAAwIJwBAAAYEE4AgAAsCAcAQAAWBCOAAAALAhHAAAAFoQjAAAAC8IRAACABeEIAADAgnAEAABgQTgCAACwIBwBAABYEI4AAAAsCEcAAAAWhCMAAAALwhEAAIAF4QgAAMCCcAQAAGBBOAIAALAgHAEAAFgQjgAAACwIRwAAABaEIwAAAAvCEQAAgAXhCAAAwIJwBAAAYEE4AgAAsCAcAQAAWBCOAAAALPy83QCaTlZWlvLz8+tUu3///kbuBgCA5olw1EpkZWWpe4+eKi876e1WAABo1ghHrUR+fr7Ky04qYmyK/CNizllfdnCnijavboLO0FTqczQwMjJSsbGxjdgNADRfhKNWxj8iRnZH13PWVRZkN0E3aApVpYWSzaaJEyfW+TOBQcHK/HY/AQlAq0Q4AnxctbNUMqbORw0rC7JV8P5C5efnE44AtEqEI6CVqOtRQwBo7biUHwAAwIJwBAAAYEE4AgAAsCAcAQAAWBCOAAAALLwajj7//HONGzdO0dHRstlsevfdd93WG2OUmpqq6OhoBQUFKSkpSfv27XOrcTqdmj59uiIjIxUSEqLrr79eR44cacJRAAAAX+LVS/lPnDihvn376u6779bNN99cY/2CBQu0aNEiLV++XN26ddPcuXM1YsQIZWZmKjQ0VJKUnJys//u//1N6eroiIiKUkpKisWPHKiMjQ23btm3qIQE+o6531OZu2gB8jVfD0ejRozV69Oha1xljtHjxYj3++OMaP368JGnFihWKiorS66+/rqlTp6qoqEhLly7VqlWrNHz4cEnS6tWrFRMTo/Xr12vUqFFNNhbAV9T3jtrcTRuAr2m2N4E8dOiQcnNzNXLkSNcyu92uwYMHa+vWrZo6daoyMjJUWVnpVhMdHa2EhARt3br1rOHI6XTK6XS63hcXFzfeQIAWpj531OZu2gB8UbMNR7m5uZKkqKgot+VRUVE6fPiwqyYgIEDh4eE1ak5/vjZpaWl64oknGrhjwLdwR20ArVWzv1rNZrO5vTfG1Fh2pnPVzJo1S0VFRa5XdjYPWQUAAD9ptuHI4XBIUo0jQHl5ea6jSQ6HQxUVFSosLDxrTW3sdrvat2/v9gIAAJCacTiKi4uTw+HQunXrXMsqKiq0adMmDRo0SJLUv39/+fv7u9Xk5ORo7969rhoAAID68Oqco9LSUv3zn/90vT906JB2796tDh06KDY2VsnJyZo3b57i4+MVHx+vefPmKTg4WLfffrskKSwsTFOmTFFKSooiIiLUoUMHzZw5U71793ZdvQYAAFAfXg1HO3fu1JAhQ1zvH3nkEUnS5MmTtXz5cj366KMqKyvTtGnTVFhYqAEDBmjt2rWuexxJ0rPPPis/Pz9NmDBBZWVlGjZsmJYvX849jgAAgEe8Go6SkpJkjDnrepvNptTUVKWmpp61JjAwUEuWLNGSJUsaoUMAANDaNNs5RwAAAN5AOAIAALAgHAEAAFgQjgAAACwIRwAAABaEIwAAAAvCEQAAgAXhCAAAwIJwBAAAYEE4AgAAsCAcAQAAWBCOAAAALAhHAAAAFoQjAAAAC8IRAACABeEIAADAgnAEAABg4eftBgC0fPv3769zbWRkpGJjYxuxGwD4ZQhHADxWVVoo2WyaOHFinT8TGBSszG/3E5AANFuEIwAeq3aWSsYoYmyK/CNizllfWZCtgvcXKj8/n3AEoNkiHAH4xfwjYmR3dPV2GwDQIJiQDQAAYEE4AgAAsCAcAQAAWBCOAAAALAhHAAAAFlytBqDJcdNIAM0Z4QhAk+GmkQBaAsIRgCbDTSMBtASEIwBNjptGAmjOmJANAABgwZGjFi4rK0v5+fnnrKvPBFgAAFozwlELlpWVpe49eqq87KS3WwEAwGcQjlqw/Px8lZedrNPk1rKDO1W0eXUTdQY0rLoe+eSyfwANgXDkA+oyubWyILuJugEaTn0v/eeyfwANgXAEoNmqz6X/XPYPoKEQjgA0e1z6D6ApcSk/AACABeEIAADAgnAEAABgQTgCAACwIBwBAABYEI4AAAAsCEcAAAAW3OcIgE+pz0OWedwIgNoQjgD4hPo+akTicSMAakc4AuAT6vOoEYnHjQA4O8IRAJ9S30eN1PU0HKfggNaDcASgVarvaThOwQGtB+EIQKtUn9Nwp0/Bbd68WT179qzT9p1Op+x2e51qOSoFNC+EIwCtWl1Ow3ky2Vu2NpKprlMpR6WA5oVwBADnUN/J3mUHd6po8+p6HZViYjjQfBCOAKCO6jrZu7Igu1719ZWVlaX8/Pw61XLKDqg/whEAtCBZWVnq3qOnystO1qmeU3ZA/RGOmpn6/EZYnzsBA/AN+fn5Ki87ySk7oBERjpqR+v5GCMB31PWXndN1jXXKDgDhqFmpz2+E0n8mfQJouTy6Eg5Ao/KZcPTCCy/omWeeUU5Oji6++GItXrxYV199tbfb8kh9J30CaLk8vRKuuWByOHyRT4SjN998U8nJyXrhhRd05ZVX6uWXX9bo0aP1zTff8BcRQIvQXH4pqk/YycnJ0c2//o2c5WV1qmdyOFoKnwhHixYt0pQpU3TvvfdKkhYvXqxPPvlEL774otLS0rzcHQB4V13nM9U37JzWWHcZb8wjTRzxws9p8eGooqJCGRkZ+t3vfue2fOTIkdq6dauXuvoPrj4D4C2ezmeq7ym+xrrLuN0eqLfffkudOnU6Z219HtfS3I54NWZQa6nb9rYWH47y8/NVVVWlqKgot+VRUVHKzc2t9TNOp1NOp9P1vqioSJJUXFzcoL1lZ2erf+Jl9f4tzJn7T1VXlJ+z7vTh9brU16e2tWy7OfXCtpt3Ly11285j+yVj1P6y8Wobdt45t11x7Dud+GaDqiudderFnKpotF4q//0vlX71icaOHXvO2p/YJJk61v6kLr1UFf1bxTve0SeffKLu3bvXabtt2rRRdXXdHh1z/PhxTZx0pyqc5/6+JSnAHqjVq1bW+DevuW/bHhikjJ07FBNz7tBdH6f/3Tamfvv+nEwLd/ToUSPJbN261W353LlzTffu3Wv9zJw5c4x++lvEixcvXrx48Wrhr+zs7AbNFi3+yFFkZKTatm1b4yhRXl7eWdPvrFmz9Mgjj7jeV1dX64cfflBERIRsNluD9VZcXKyYmBhlZ2erffv2Dbbd5oZx+hbG6Ttawxglxulr6jNOY4xKSkoUHR3doD20+HAUEBCg/v37a926dbrppptcy9etW6cbbrih1s/Y7fYa56Z/9atfNVqP7du39+n/kE9jnL6FcfqO1jBGiXH6mrqOMywsrMF/dosPR5L0yCOPaNKkSUpMTNTAgQP1yiuvKCsrS/fff7+3WwMAAC2MT4SjW265RQUFBXryySeVk5OjhIQEffjhh+rSpYu3WwMAAC2MT4QjSZo2bZqmTZvm7Tbc2O12zZkzp86Xl7ZUjNO3ME7f0RrGKDFOX9McxmkzpqGvfwMAAGi52ni7AQAAgOaEcAQAAGBBOAIAALAgHAEAAFgQjhrRCy+8oLi4OAUGBqp///7avHmzt1s6q9TUVNlsNreXw+FwrTfGKDU1VdHR0QoKClJSUpL27dvntg2n06np06crMjJSISEhuv7663XkyBG3msLCQk2aNElhYWEKCwvTpEmT9OOPPzbKmD7//HONGzdO0dHRstlsevfdd93WN+WYsrKyNG7cOIWEhCgyMlK//e1vVVFR0STjvOuuu2rs2yuuuKJFjTMtLU2XXXaZQkND1bFjR914443KzMx0q/GF/VmXcfrC/nzxxRfVp08f103+Bg4cqI8++si13hf2ZV3G6Qv78kxpaWmy2WxKTk52LWuR+7NBH0YCl/T0dOPv729effVV880335gZM2aYkJAQc/jwYW+3Vqs5c+aYiy++2OTk5LheeXl5rvXz5883oaGh5u233zZ79uwxt9xyi+nUqZMpLi521dx///3m/PPPN+vWrTO7du0yQ4YMMX379jWnTp1y1Vx77bUmISHBbN261WzdutUkJCSYsWPHNsqYPvzwQ/P444+bt99+20gya9ascVvfVGM6deqUSUhIMEOGDDG7du0y69atM9HR0eahhx5qknFOnjzZXHvttW77tqCgwK2muY9z1KhRZtmyZWbv3r1m9+7dZsyYMSY2NtaUlpa6anxhf9ZlnL6wP9977z3zwQcfmMzMTJOZmWkee+wx4+/vb/bu3WuM8Y19WZdx+sK+tPriiy/MBRdcYPr06WNmzJjhWt4S9yfhqJFcfvnl5v7773db1qNHD/O73/3OSx39vDlz5pi+ffvWuq66uto4HA4zf/5817Ly8nITFhZmXnrpJWOMMT/++KPx9/c36enprpqjR4+aNm3amI8//tgYY8w333xjJJnt27e7arZt22YkmW+//bYRRvUfZ4aGphzThx9+aNq0aWOOHj3qqnnjjTeM3W43RUVFjTpOY376H/ANN9xw1s+0xHHm5eUZSWbTpk3GGN/dn2eO0xjf3J/GGBMeHm7+9Kc/+ey+PHOcxvjWviwpKTHx8fFm3bp1ZvDgwa5w1FL3J6fVGkFFRYUyMjI0cuRIt+UjR47U1q1bvdTVuX3//feKjo5WXFycbr31Vh08eFCSdOjQIeXm5rqNx263a/Dgwa7xZGRkqLKy0q0mOjpaCQkJrppt27YpLCxMAwYMcNVcccUVCgsLa/LvpSnHtG3bNiUkJLg9GHHUqFFyOp3KyMho1HGetnHjRnXs2FHdunXTfffdp7y8PNe6ljjOoqIiSVKHDh0k+e7+PHOcp/nS/qyqqlJ6erpOnDihgQMH+uy+PHOcp/nKvnzwwQc1ZswYDR8+3G15S92fPnOH7OYkPz9fVVVVioqKclseFRWl3NxcL3X18wYMGKCVK1eqW7duOn78uObOnatBgwZp3759rp5rG8/hw4clSbm5uQoICFB4eHiNmtOfz83NVceOHWv87I4dOzb599KUY8rNza3xc8LDwxUQENAk4x49erR+85vfqEuXLjp06JD+3//7fxo6dKgyMjJkt9tb3DiNMXrkkUd01VVXKSEhwfWzT/d85hha6v6sbZyS7+zPPXv2aODAgSovL1e7du20Zs0a9erVy/UPna/sy7ONU/KdfZmenq5du3Zpx44dNda11L+bhKNGZLPZ3N4bY2osay5Gjx7t+nPv3r01cOBAXXTRRVqxYoVrgqAn4zmzprZ6b34vTTUmb477lltucf05ISFBiYmJ6tKliz744AONHz/+rJ9rruN86KGH9PXXX2vLli011vnS/jzbOH1lf3bv3l27d+/Wjz/+qLfffluTJ0/Wpk2bzvqzW+q+PNs4e/Xq5RP7Mjs7WzNmzNDatWsVGBh41rqWtj85rdYIIiMj1bZt2xpJNS8vr0aqba5CQkLUu3dvff/9966r1n5uPA6HQxUVFSosLPzZmuPHj9f4Wf/+97+b/HtpyjE5HI4aP6ewsFCVlZVe+e+hU6dO6tKli77//ntXfy1lnNOnT9d7772nDRs2qHPnzq7lvrY/zzbO2rTU/RkQEKCuXbsqMTFRaWlp6tu3r5577jmf25dnG2dtWuK+zMjIUF5envr37y8/Pz/5+flp06ZN+uMf/yg/Pz/X9lvc/qzXDCXU2eWXX24eeOABt2U9e/ZsthOyz1ReXm7OP/9888QTT7gm1D399NOu9U6ns9YJdW+++aar5tixY7VOqPvHP/7hqtm+fbtXJ2Q3xZhOTxI8duyYqyY9Pb3JJmSfKT8/39jtdrNixQpjTMsYZ3V1tXnwwQdNdHS0+e6772pd7wv781zjrE1L3J+1GTp0qJk8ebLP7MtzjbM2LXFfFhcXmz179ri9EhMTzcSJE82ePXta7P4kHDWS05fyL1261HzzzTcmOTnZhISEmH/961/ebq1WKSkpZuPGjebgwYNm+/btZuzYsSY0NNTV7/z5801YWJh55513zJ49e8xtt91W66WYnTt3NuvXrze7du0yQ4cOrfVSzD59+pht27aZbdu2md69ezfapfwlJSXmyy+/NF9++aWRZBYtWmS+/PJL1+0UmmpMpy8vHTZsmNm1a5dZv3696dy5c4NdRvtz4ywpKTEpKSlm69at5tChQ2bDhg1m4MCB5vzzz29R43zggQdMWFiY2bhxo9tlzydPnnTV+ML+PNc4fWV/zpo1y3z++efm0KFD5uuvvzaPPfaYadOmjVm7dq0xxjf25bnG6Sv7sjbWq9WMaZn7k3DUiP73f//XdOnSxQQEBJh+/fq5XY7b3Jy+74S/v7+Jjo4248ePN/v27XOtr66uNnPmzDEOh8PY7XZzzTXXmD179rhto6yszDz00EOmQ4cOJigoyIwdO9ZkZWW51RQUFJg77rjDhIaGmtDQUHPHHXeYwsLCRhnThg0bjKQar9O/tTXlmA4fPmzGjBljgoKCTIcOHcxDDz1kysvLG32cJ0+eNCNHjjTnnXee8ff3N7GxsWby5Mk1xtDcx1nb+CSZZcuWuWp8YX+ea5y+sj/vuece1/8bzzvvPDNs2DBXMDLGN/blucbpK/uyNmeGo5a4P23GGFO/E3EAAAC+iwnZAAAAFoQjAAAAC8IRAACABeEIAADAgnAEAABgQTgCAACwIBwBAABYEI4AoAX517/+JZvNpt27d3u7FcBnEY6AViovL09Tp05VbGys7Ha7HA6HRo0apW3btnm7Na9rLgHkrrvu0o033ujVHoDWyM/bDQDwjptvvlmVlZVasWKFLrzwQh0/flyffvqpfvjhB2+31mAqKioUEBDg7TYAtDAcOQJaoR9//FFbtmzR008/rSFDhqhLly66/PLLNWvWLI0ZM0ZS7UdPfvzxR9lsNm3cuFGStHHjRtlsNn3yySe69NJLFRQUpKFDhyovL08fffSRevbsqfbt2+u2227TyZMnXdtJSkrS9OnTlZycrPDwcEVFRemVV17RiRMndPfddys0NFQXXXSRPvroI9dnqqqqNGXKFMXFxSkoKEjdu3fXc8895zau00da0tLSFB0drW7duunJJ59U7969a3wH/fv31+zZsz36/owxWrBggS688EIFBQWpb9++euutt1zrT38vn376qRITExUcHKxBgwYpMzPTbTtz585Vx44dFRoaqnvvvVe/+93vdMkll0iSUlNTtWLFCv3tb3+TzWZz+94l6eDBgxoyZIiCg4PVt29fjvgBDaneT2MD0OJVVlaadu3ameTk5LM+lPHQoUNGkvnyyy9dywoLC40ks2HDBmPMfx56e8UVV5gtW7aYXbt2ma5du5rBgwebkSNHml27dpnPP//cREREmPnz57u2M3jwYBMaGmqeeuop891335mnnnrKtGnTxowePdq88sor5rvvvjMPPPCAiYiIMCdOnDDGGFNRUWFmz55tvvjiC3Pw4EGzevVqExwcbN58803XdidPnmzatWtnJk2aZPbu3Wv27NljsrOzTZs2bcwXX3zhqvvqq6+MzWYzBw4cqPPYrR577DHTo0cP8/HHH5sDBw6YZcuWGbvdbjZu3Oj2vQwYMMBs3LjR7Nu3z1x99dVm0KBBrm2sXr3aBAYGmtdee81kZmaaJ554wrRv39707dvXGGNMSUmJmTBhgrn22mtNTk6OycnJMU6n09Vbjx49zPvvv28yMzPNr3/9a9OlSxdTWVlZ+w4HUC+EI6CVeuutt0x4eLgJDAw0gwYNMrNmzTJfffWVa319wtH69etdNWlpaUaSW/CYOnWqGTVqlOv94MGDzVVXXeV6f+rUKRMSEmImTZrkWpaTk2MkmW3btp11DNOmTTM333yz6/3kyZNNVFSUcTqdbnWjR482DzzwgOt9cnKySUpKOut2fy4clZaWmsDAQLN161a35VOmTDG33XabMab27+WDDz4wkkxZWZkxxpgBAwaYBx980G0bV155pSscnR7PDTfcUGtvf/rTn1zL9u3bZySZ/fv3n3VMAOqO02pAK3XzzTfr2LFjeu+99zRq1Cht3LhR/fr10/Lly+u9rT59+rj+HBUVpeDgYF144YVuy/Ly8s76mbZt2yoiIsLt9FdUVJQkuX3upZdeUmJios477zy1a9dOr776qrKysty227t37xrzjO677z698cYbKi8vV2Vlpf785z/rnnvuqfc4Jembb75ReXm5RowYoXbt2rleK1eu1IEDB846xk6dOrmNJzMzU5dffrlb/Znvf87PbRvAL8OEbKAVCwwM1IgRIzRixAjNnj1b9957r+bMmaO77rpLbdr89LuTMcZVX1lZWet2/P39XX+22Wxu708vq66uPutnavuczWaTJNfn/vKXv+jhhx/WwoULNXDgQIWGhuqZZ57RP/7xD7fthISE1Ohv3LhxstvtWrNmjex2u5xOp26++ebav5RzON3PBx98oPPPP99tnd1uP+sYzxyPddlp1u/6XM61bQCeIxwBcOnVq5feffddSdJ5550nScrJydGll14qSV69tH3z5s0aNGiQpk2b5lp25pGas/Hz89PkyZO1bNky2e123XrrrQoODvaoj169eslutysrK0uDBw/2aBuS1L17d33xxReaNGmSa9nOnTvdagICAlRVVeXxzwDgGcIR0AoVFBToN7/5je655x716dNHoaGh2rlzpxYsWKAbbrhBkhQUFKQrrrhC8+fP1wUXXKD8/Hz9z//8j9d67tq1q1auXKlPPvlEcXFxWrVqlXbs2KG4uLg6ff7ee+9Vz549JUl///vf6/SZM68uk34KRzNnztTDDz+s6upqXXXVVSouLtbWrVvVrl07TZ48uU7bnj59uu677z4lJiZq0KBBevPNN/X111+7nY684IIL9MknnygzM1MREREKCwur07YB/DKEI6AVateunQYMGKBnn31WBw4cUGVlpWJiYnTffffpsccec9W99tpruueee5SYmKju3btrwYIFGjlypFd6vv/++7V7927dcsststlsuu222zRt2jS3y/1/Tnx8vAYNGqSCggINGDCgTp+59dZbayw7dOiQnnrqKXXs2FFpaWk6ePCgfvWrX6lfv35u39253HHHHTp48KBmzpyp8vJyTZgwQXfddZe++OILV819992njRs3KjExUaWlpdqwYYMuuOCCOv8MAJ6xmfqc5AaAFsoYox49emjq1Kl65JFHvN1OrUaMGCGHw6FVq1Z5uxWgVePIEQCfl5eXp1WrVuno0aO6++67vd2OJOnkyZN66aWXNGrUKLVt21ZvvPGG1q9fr3Xr1nm7NaDVIxwB8HlRUVGKjIzUK6+8ovDwcG+3I+mnK8w+/PBDzZ07V06nU927d9fbb7+t4cOHe7s1oNXjtBoAAIAFN4EEAACwIBwBAABYEI4AAAAsCEcAAAAWhCMAAAALwhEAAIAF4QgAAMCCcAQAAGBBOAIAALD4/1I8GCel7qS9AAAAAElFTkSuQmCC",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(general['text_length'], bins=range(0,40000,1000), edgecolor='black')\n",
    "plt.xlabel('Summary Length')\n",
    "plt.ylabel('Frequency')\n",
    "plt.title('general')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "lower bound 4223.25\n",
      "upper_bound 16917.75\n"
     ]
    }
   ],
   "source": [
    "lower_bound = general['text_length'].quantile(0.05)\n",
    "upper_bound = general['text_length'].quantile(0.95)\n",
    "print('lower bound', lower_bound)\n",
    "print('upper_bound', upper_bound)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "filtered_general = general[(general['text_length'] >= lower_bound) & (general['text_length'] <= upper_bound)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "상, 하위 5%씩 제거한 후 글자수 평균:  9129.252079002079\n"
     ]
    }
   ],
   "source": [
    "print(\"상, 하위 5%씩 제거한 후 글자수 평균: \", filtered_general['text_length'].mean())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(filtered_general['text_length'], bins=range(0,20000,1000), edgecolor='black')\n",
    "plt.xlabel('Summary Length')\n",
    "plt.ylabel('Frequency')\n",
    "plt.title('general')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "# admission_general\n",
    "X = filtered_general['text']\n",
    "y = filtered_general['label']\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.1, random_state=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "# admission_30days\n",
    "A = days_30['text']\n",
    "b = days_30['label']\n",
    "\n",
    "A_train, A_test, b_train, b_test = train_test_split(A, b, test_size = 0.1, random_state=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## word2vec Embedding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calculate_length(texts):\n",
    "    lengths = []\n",
    "    for text in texts : \n",
    "        parts = text.split(',')\n",
    "        for part in parts : \n",
    "            word_count = len(part.split())\n",
    "            lengths.append(word_count)\n",
    "\n",
    "    avg_length = np.mean(lengths)\n",
    "    max_length = np.max(lengths)\n",
    "    min_length = np.min(lengths)\n",
    "\n",
    "    return avg_length, max_length, min_length\n",
    "\n",
    "# 각 행의 평균 길이 계산\n",
    "general['length'] = general['text'].apply(calculate_length)\n",
    "days_30['length'] = days_30['text'].apply(calculate_length)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "전체 평균: 0.8608037640377574\n",
      "최대 길이: 1.0\n",
      "최소 길이: 0.0\n"
     ]
    }
   ],
   "source": [
    "avg_length = general['length'].apply(lambda x: x[0])\n",
    "max_length = general['length'].apply(lambda x: x[1])\n",
    "min_length = general['length'].apply(lambda x: x[2])\n",
    "\n",
    "average = avg_length.mean()\n",
    "max_average = max_length.mean()\n",
    "min_average = min_length.mean()\n",
    "\n",
    "print(f\"전체 평균: {average}\")\n",
    "print(f\"최대 길이: {max_average}\")\n",
    "print(f\"최소 길이: {min_average}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "전체 평균: 0.8606459642325579\n",
      "최대 길이: 1.0\n",
      "최소 길이: 0.0\n"
     ]
    }
   ],
   "source": [
    "avg_length = days_30['length'].apply(lambda x: x[0])\n",
    "max_length = days_30['length'].apply(lambda x: x[1])\n",
    "min_length = days_30['length'].apply(lambda x: x[2])\n",
    "\n",
    "# 각 리스트의 전체 평균 계산\n",
    "average = avg_length.mean()\n",
    "max_average = max_length.mean()\n",
    "min_average = min_length.mean()\n",
    "\n",
    "print(f\"전체 평균: {average}\")\n",
    "print(f\"최대 길이: {max_average}\")\n",
    "print(f\"최소 길이: {min_average}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "word2vec_path = \"C:/Users/gangmin/dahs/PubMed-and-PMC-w2v.bin\"\n",
    "word2vec = KeyedVectors.load_word2vec_format(word2vec_path, binary=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- max length == 10000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "embedding_dim = 200\n",
    "\n",
    "def get_word_embedding(word) : \n",
    "    if word in word2vec : \n",
    "        return word2vec[word]\n",
    "    else : \n",
    "        return np.random.uniform(-0.25, 0.25, embedding_dim) # Words not present in the set of pre-trained words are initialized randomly\n",
    "\n",
    "def pad_sequences(sequences, maxlen=10000):\n",
    "    if maxlen is None:\n",
    "        maxlen = max(len(seq) for seq in sequences)\n",
    "    padded_sequences = np.zeros((len(sequences), maxlen, embedding_dim))\n",
    "    for i, seq in enumerate(sequences):\n",
    "        length = min(len(seq), maxlen)\n",
    "        padded_sequences[i, :length] = seq[:length]\n",
    "    return padded_sequences\n",
    "\n",
    "def create_padded_embeddings(texts, batch_size=32):\n",
    "    batches = [texts[i:i + batch_size] for i in range(0, len(texts), batch_size)]\n",
    "    all_padded_embeddings = []\n",
    "    for batch in tqdm(batches):\n",
    "        embeddings = []\n",
    "        for text in batch:\n",
    "            words = text.split()\n",
    "            embedding = [get_word_embedding(word) for word in words]\n",
    "            embeddings.append(embedding)\n",
    "        padded_embeddings = pad_sequences(embeddings)\n",
    "        all_padded_embeddings.extend(padded_embeddings)\n",
    "    return np.array(all_padded_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 433/433 [00:05<00:00, 75.38it/s]\n",
      "100%|██████████| 25/25 [00:00<00:00, 35.33it/s]\n"
     ]
    }
   ],
   "source": [
    "padded_embeddings_train = create_padded_embeddings(X_train, batch_size=8)\n",
    "padded_embeddings_test = create_padded_embeddings(X_test, batch_size=16)\n",
    "labels_train = y_train.tolist()\n",
    "labels_test = y_test.tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 200/200 [00:03<00:00, 60.38it/s]\n",
      "100%|██████████| 12/12 [00:00<00:00, 32.28it/s]\n"
     ]
    }
   ],
   "source": [
    "padded_embeddings_train_30days = create_padded_embeddings(A_train, batch_size=8)\n",
    "padded_embeddings_test_30days = create_padded_embeddings(A_test, batch_size=16)\n",
    "labels_train_30days = b_train.tolist()\n",
    "labels_test_30days = b_test.tolist()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- max length == 20000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "embedding_dim = 200\n",
    "\n",
    "def get_word_embedding(word) : \n",
    "    if word in word2vec : \n",
    "        return word2vec[word]\n",
    "    else : \n",
    "        return np.random.uniform(-0.25, 0.25, embedding_dim) # Words not present in the set of pre-trained words are initialized randomly\n",
    "\n",
    "def pad_sequences(sequences, maxlen=20000):\n",
    "    if maxlen is None:\n",
    "        maxlen = max(len(seq) for seq in sequences)\n",
    "    padded_sequences = np.zeros((len(sequences), maxlen, embedding_dim))\n",
    "    for i, seq in enumerate(sequences):\n",
    "        length = min(len(seq), maxlen)\n",
    "        padded_sequences[i, :length] = seq[:length]\n",
    "    return padded_sequences\n",
    "\n",
    "def create_padded_embeddings(texts, batch_size=32):\n",
    "    batches = [texts[i:i + batch_size] for i in range(0, len(texts), batch_size)]\n",
    "    all_padded_embeddings = []\n",
    "    for batch in tqdm(batches):\n",
    "        embeddings = []\n",
    "        for text in batch:\n",
    "            words = text.split()\n",
    "            embedding = [get_word_embedding(word) for word in words]\n",
    "            embeddings.append(embedding)\n",
    "        padded_embeddings = pad_sequences(embeddings)\n",
    "        all_padded_embeddings.extend(padded_embeddings)\n",
    "    return np.array(all_padded_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 433/433 [00:05<00:00, 76.12it/s]\n",
      "100%|██████████| 25/25 [00:00<00:00, 26.99it/s]\n"
     ]
    }
   ],
   "source": [
    "padded_embeddings_train = create_padded_embeddings(X_train, batch_size=8)\n",
    "padded_embeddings_test = create_padded_embeddings(X_test, batch_size=16)\n",
    "labels_train = y_train.tolist()\n",
    "labels_test = y_test.tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 200/200 [00:03<00:00, 59.56it/s]\n",
      "100%|██████████| 12/12 [00:00<00:00, 23.81it/s]\n"
     ]
    }
   ],
   "source": [
    "padded_embeddings_train_30days = create_padded_embeddings(A_train, batch_size=8)\n",
    "padded_embeddings_test_30days = create_padded_embeddings(A_test, batch_size=16)\n",
    "labels_train_30days = b_train.tolist()\n",
    "labels_test_30days = b_test.tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(3463, 20000, 200)"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "padded_embeddings_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(385, 20000, 200)"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "padded_embeddings_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1600, 20000, 200)"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "padded_embeddings_train_30days.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(178, 20000, 200)"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "padded_embeddings_test_30days.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Modeling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ['CUDA_LAUNCH_BLOCKING'] = '1'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ClinicalnotesDataset(Dataset) : \n",
    "    def __init__(self, data, labels) : \n",
    "        self.data = data\n",
    "        self.labels = labels\n",
    "\n",
    "    def __len__(self) : \n",
    "        return len(self.data)\n",
    "    \n",
    "    def __getitem__(self, idx) : \n",
    "        data_tensor = torch.tensor(self.data[idx], dtype=torch.float32)\n",
    "        label_tensor = torch.tensor(self.labels[idx], dtype=torch.long)\n",
    "        return data_tensor, label_tensor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 단순히 readmission 이력만 있는 경우\n",
    "train_dataset = ClinicalnotesDataset(padded_embeddings_train, labels_train)\n",
    "test_dataset = ClinicalnotesDataset(padded_embeddings_test, labels_test)\n",
    "\n",
    "train_dataloader = DataLoader(train_dataset, batch_size=8, shuffle=True)\n",
    "test_dataloader = DataLoader(test_dataset, batch_size=16, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 30일 이내에 readmission 이력이 있는 경우\n",
    "train_dataset_30days = ClinicalnotesDataset(padded_embeddings_train_30days, labels_train_30days)\n",
    "test_dataset_30days = ClinicalnotesDataset(padded_embeddings_test_30days, labels_test_30days)\n",
    "\n",
    "train_dataloader_30days = DataLoader(train_dataset_30days, batch_size=8, shuffle=True)\n",
    "test_dataloader_30days = DataLoader(test_dataset_30days, batch_size=16, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training data and labels length: 3463 3463\n",
      "Test data and labels length: 385 385\n"
     ]
    }
   ],
   "source": [
    "print(\"Training data and labels length:\", len(padded_embeddings_train), len(labels_train))\n",
    "print(\"Test data and labels length:\", len(padded_embeddings_test), len(labels_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training data and labels length: 1600 1600\n",
      "Test data and labels length: 178 178\n"
     ]
    }
   ],
   "source": [
    "print(\"Training data and labels length:\", len(padded_embeddings_train_30days), len(labels_train_30days))\n",
    "print(\"Test data and labels length:\", len(padded_embeddings_test_30days), len(labels_test_30days))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "class CNN(nn.Module):\n",
    "  def __init__(self, embedding_dim, n_filters, filter_sizes, dropout, output_dim):\n",
    "    \n",
    "    super().__init__()\n",
    "\n",
    "    self.embedding_dim = embedding_dim\n",
    "    self.n_filters = n_filters\n",
    "    self.filter_sizes = filter_sizes\n",
    "    self.output_dim = output_dim\n",
    "    self.dropout = nn.Dropout(dropout)\n",
    "\n",
    "    self.conv_0 = nn.Conv1d(in_channels = embedding_dim, \n",
    "                            out_channels = n_filters, \n",
    "                            kernel_size = filter_sizes[0]) # 1D CNN 연산에서는 커널의 높이만으로 해당 커널의 크기를 간주함(커널의 너비는 문장 행렬에서의 임베딩 벡터의 차원과 동일함)\n",
    "    \n",
    "    self.conv_1 = nn.Conv1d(in_channels = embedding_dim,\n",
    "                            out_channels = n_filters,\n",
    "                            kernel_size = filter_sizes[1]) \n",
    "    \n",
    "    self.conv_2 = nn.Conv1d(in_channels = embedding_dim,\n",
    "                            out_channels = n_filters,\n",
    "                            kernel_size = filter_sizes[2])\n",
    "    \n",
    "    self.fc = nn.Linear(len(filter_sizes) * n_filters, output_dim)\n",
    "\n",
    "  def forward(self, embedded):\n",
    "\n",
    "    embedded = embedded.permute(0, 2, 1)\n",
    "\n",
    "    conved_0 = F.relu(self.conv_0(embedded))\n",
    "    conved_1 = F.relu(self.conv_1(embedded))\n",
    "    conved_2 = F.relu(self.conv_2(embedded))\n",
    "\n",
    "    pooled_0 = F.max_pool1d(conved_0, conved_0.shape[2]).squeeze(2)\n",
    "    pooled_1 = F.max_pool1d(conved_1, conved_1.shape[2]).squeeze(2)\n",
    "    pooled_2 = F.max_pool1d(conved_2, conved_2.shape[2]).squeeze(2)\n",
    "\n",
    "    cat = self.dropout(torch.cat((pooled_0, pooled_1, pooled_2), dim = 1))\n",
    "\n",
    "    return F.softmax(self.fc(cat), dim=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "embedding_dim = 200\n",
    "n_filters = 90\n",
    "filter_sizes = [1, 2, 3]\n",
    "dropout = 0.2\n",
    "output_dim = 2\n",
    "learning_rate = 0.0002\n",
    "num_epochs = 3000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "device = 'cuda' if torch.cuda.is_available() else 'cpu'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 단순히 readmission만 한 경우"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "Tracking run with wandb version 0.17.4"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Run data is saved locally in <code>c:\\Users\\gangmin\\dahs\\project\\wandb\\run-20240709_123153-wclf18qg</code>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Syncing run <strong><a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp/runs/wclf18qg' target=\"_blank\">only_readmission_1st_20000</a></strong> to <a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp' target=\"_blank\">Weights & Biases</a> (<a href='https://wandb.me/run' target=\"_blank\">docs</a>)<br/>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       " View project at <a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp' target=\"_blank\">https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp</a>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       " View run at <a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp/runs/wclf18qg' target=\"_blank\">https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp/runs/wclf18qg</a>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 0/3000 [00:00<?, ?it/s]c:\\Users\\gangmin\\AppData\\Local\\miniconda3\\envs\\medical\\lib\\site-packages\\torch\\nn\\modules\\conv.py:306: UserWarning: Plan failed with a cudnnException: CUDNN_BACKEND_EXECUTION_PLAN_DESCRIPTOR: cudnnFinalize Descriptor Failed cudnn_status: CUDNN_STATUS_NOT_SUPPORTED (Triggered internally at C:\\cb\\pytorch_1000000000000\\work\\aten\\src\\ATen\\native\\cudnn\\Conv_v8.cpp:919.)\n",
      "  return F.conv1d(input, weight, bias, self.stride,\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1, Train Loss: 0.6718515655322659, Test Loss : 0.642245020866394,  Accuracy: 0.6649350649350649, Precision: 0.6312849162011173, Recall: 0.6420454545454546, F1 Score: 0.6366197183098592\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 1/3000 [01:13<61:09:45, 73.42s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 2, Train Loss: 0.6355468136609014, Test Loss : 0.6257494139671326,  Accuracy: 0.638961038961039, Precision: 0.5687732342007435, Recall: 0.8693181818181818, F1 Score: 0.6876404494382022\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 2/3000 [01:56<46:05:03, 55.34s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 3, Train Loss: 0.6166470799655341, Test Loss : 0.6179393374919891,  Accuracy: 0.6545454545454545, Precision: 0.5811320754716981, Recall: 0.875, F1 Score: 0.6984126984126984\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 3/3000 [02:37<40:37:31, 48.80s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 4, Train Loss: 0.6048313739255724, Test Loss : 0.6077264308929443,  Accuracy: 0.6753246753246753, Precision: 0.6133333333333333, Recall: 0.7840909090909091, F1 Score: 0.6882793017456358\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 4/3000 [03:04<33:40:06, 40.46s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 5, Train Loss: 0.5938698389926629, Test Loss : 0.6040266418457031,  Accuracy: 0.6831168831168831, Precision: 0.6285714285714286, Recall: 0.75, F1 Score: 0.6839378238341969\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 5/3000 [03:32<29:45:59, 35.78s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 6, Train Loss: 0.5791271790758842, Test Loss : 0.6031780850887298,  Accuracy: 0.6753246753246753, Precision: 0.6268656716417911, Recall: 0.7159090909090909, F1 Score: 0.6684350132625996\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 6/3000 [03:59<27:24:27, 32.96s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 7, Train Loss: 0.5724016841647256, Test Loss : 0.5990850234031677,  Accuracy: 0.6909090909090909, Precision: 0.6266666666666667, Recall: 0.8011363636363636, F1 Score: 0.7032418952618454\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 7/3000 [04:27<25:52:41, 31.13s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 8, Train Loss: 0.5586760875534516, Test Loss : 0.6009025585651397,  Accuracy: 0.6935064935064935, Precision: 0.6106870229007634, Recall: 0.9090909090909091, F1 Score: 0.730593607305936\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 8/3000 [04:54<24:54:07, 29.96s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 9, Train Loss: 0.5565315459534422, Test Loss : 0.5988914632797241,  Accuracy: 0.6987012987012987, Precision: 0.6578947368421053, Recall: 0.7102272727272727, F1 Score: 0.6830601092896175\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 9/3000 [05:22<24:16:13, 29.21s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 10, Train Loss: 0.5470808290168815, Test Loss : 0.5970643901824951,  Accuracy: 0.6883116883116883, Precision: 0.6060606060606061, Recall: 0.9090909090909091, F1 Score: 0.7272727272727273\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 10/3000 [05:49<23:54:02, 28.78s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 11, Train Loss: 0.533733350024763, Test Loss : 0.591816155910492,  Accuracy: 0.7116883116883117, Precision: 0.6431718061674009, Recall: 0.8295454545454546, F1 Score: 0.7245657568238213\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 11/3000 [06:17<23:36:19, 28.43s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 12, Train Loss: 0.5262932125608034, Test Loss : 0.595165057182312,  Accuracy: 0.6987012987012987, Precision: 0.6136363636363636, Recall: 0.9204545454545454, F1 Score: 0.7363636363636363\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 12/3000 [06:45<23:29:55, 28.31s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 13, Train Loss: 0.5199856242599443, Test Loss : 0.5910062634944916,  Accuracy: 0.7116883116883117, Precision: 0.6444444444444445, Recall: 0.8238636363636364, F1 Score: 0.7231920199501246\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 13/3000 [07:14<23:33:08, 28.39s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 14, Train Loss: 0.510879769749234, Test Loss : 0.591003201007843,  Accuracy: 0.7116883116883117, Precision: 0.6701570680628273, Recall: 0.7272727272727273, F1 Score: 0.6975476839237057\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 14/3000 [07:42<23:32:16, 28.38s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 15, Train Loss: 0.5003221826520308, Test Loss : 0.5888497877120972,  Accuracy: 0.7064935064935065, Precision: 0.6245059288537549, Recall: 0.8977272727272727, F1 Score: 0.7365967365967365\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 15/3000 [08:10<23:18:48, 28.12s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 16, Train Loss: 0.4942150389900384, Test Loss : 0.5853653311729431,  Accuracy: 0.7272727272727273, Precision: 0.6577777777777778, Recall: 0.8409090909090909, F1 Score: 0.7381546134663342\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 16/3000 [08:37<23:11:34, 27.98s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 17, Train Loss: 0.49318079075692156, Test Loss : 0.5864282071590423,  Accuracy: 0.7168831168831169, Precision: 0.6618357487922706, Recall: 0.7784090909090909, F1 Score: 0.7154046997389034\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 17/3000 [09:05<23:05:23, 27.87s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 18, Train Loss: 0.4838149707945059, Test Loss : 0.5869072091579437,  Accuracy: 0.7116883116883117, Precision: 0.6326530612244898, Recall: 0.8806818181818182, F1 Score: 0.7363420427553443\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 18/3000 [09:33<23:03:08, 27.83s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 19, Train Loss: 0.47727398302483504, Test Loss : 0.5910935747623444,  Accuracy: 0.6987012987012987, Precision: 0.6578947368421053, Recall: 0.7102272727272727, F1 Score: 0.6830601092896175\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 19/3000 [10:00<23:01:19, 27.80s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 20, Train Loss: 0.47407187504250786, Test Loss : 0.5866075491905213,  Accuracy: 0.7116883116883117, Precision: 0.6666666666666666, Recall: 0.7386363636363636, F1 Score: 0.7008086253369272\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 20/3000 [10:28<22:59:39, 27.78s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Early stopping 기능이 21에서 실행되었습니다.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 20/3000 [10:56<27:09:59, 32.82s/it]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "모델 학습 종료\n"
     ]
    }
   ],
   "source": [
    "model = CNN(embedding_dim, n_filters, filter_sizes, dropout, output_dim).to(device)\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = optim.Adam(model.parameters(), lr=learning_rate)\n",
    "\n",
    "best_loss = np.inf # 초기 설정\n",
    "patience = 5\n",
    "patience_counter = 0\n",
    "\n",
    "for epoch in tqdm(range(num_epochs)):\n",
    "    model.train()\n",
    "    \n",
    "    train_loss = 0.0\n",
    "\n",
    "    for inputs, labels in train_dataloader :\n",
    "        inputs, labels = inputs.to(device), labels.to(device)\n",
    "        optimizer.zero_grad()\n",
    "\n",
    "        train_predictions = model(inputs)\n",
    "\n",
    "        loss = criterion(train_predictions, labels.long())\n",
    "        \n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        train_loss += loss.item()\n",
    "    \n",
    "    model.eval()\n",
    "    with torch.no_grad() : \n",
    "        all_test_labels = []\n",
    "        all_test_predictions = []\n",
    "        test_loss = 0.0\n",
    "\n",
    "        for inputs, labels in test_dataloader : \n",
    "            inputs, labels = inputs.to(device), labels.to(device)\n",
    "            test_predictions = model(inputs)\n",
    "\n",
    "            loss = criterion(test_predictions, labels.long())\n",
    "            test_loss += loss.item()\n",
    "\n",
    "            all_test_labels.extend(labels.cpu().numpy())\n",
    "            all_test_predictions.extend(torch.argmax(test_predictions, dim=1).cpu().numpy())\n",
    "    \n",
    "    train_loss /= len(train_dataloader)\n",
    "    test_loss /= len(test_dataloader)\n",
    "\n",
    "    if test_loss < best_loss:\n",
    "        best_loss = test_loss\n",
    "        patience_counter = 0\n",
    "\n",
    "    else :\n",
    "        patience_counter += 1\n",
    "        if patience_counter >= patience:\n",
    "            print(f'Early stopping 기능이 {epoch+1}에서 실행되었습니다.')\n",
    "            break\n",
    "\n",
    "    all_labels = np.array(all_test_labels)\n",
    "    all_predictions = np.array(all_test_predictions)\n",
    "    \n",
    "    accuracy = accuracy_score(all_test_labels, all_test_predictions)\n",
    "    precision = precision_score(all_test_labels, all_test_predictions, average = 'binary')\n",
    "    recall = recall_score(all_test_labels, all_test_predictions, average = 'binary')\n",
    "    f1 = f1_score(all_test_labels, all_test_predictions, average = 'binary')\n",
    "    \n",
    "    print(f'Epoch {epoch+1}, Train Loss: {train_loss}, Test Loss : {test_loss},  Accuracy: {accuracy}, Precision: {precision}, Recall: {recall}, F1 Score: {f1}')\n",
    "\n",
    "print(\"모델 학습 종료\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 276/276 [00:42<00:00,  6.55it/s]\n",
      "100%|██████████| 109/109 [00:16<00:00,  6.66it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: impella, Chi2: 51.52622510126866, p-value: 7.064706041556759e-13\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: expired, Chi2: 47.995152578728806, p-value: 4.272742024520224e-12\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: copd, Chi2: 45.38882046167588, p-value: 1.615539011728561e-11\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: malaise, Chi2: 42.5766931608721, p-value: 6.796440902868595e-11\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: pain, Chi2: 39.93337049689079, p-value: 2.627752400369123e-10\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: ms, Chi2: 34.82467804714551, p-value: 3.607713850643471e-09\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: wheezing, Chi2: 34.66814617712602, p-value: 3.909770101914784e-09\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: take, Chi2: 34.3832685331659, p-value: 4.525973812269287e-09\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: puncture, Chi2: 30.861886667319027, p-value: 2.7705989558490992e-08\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: carotid, Chi2: 30.71108129332426, p-value: 2.9945002842713326e-08\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: metoprolol, Chi2: 29.968955628997595, p-value: 4.3901903375349056e-08\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: neb, Chi2: 28.418263009335135, p-value: 9.773892621580926e-08\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: follow, Chi2: 28.296704338955582, p-value: 1.0407302205866559e-07\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: transplant, Chi2: 27.672229933825616, p-value: 1.4371102365912053e-07\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: citalopram, Chi2: 27.24242275976928, p-value: 1.7947711617297674e-07\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: hd, Chi2: 26.996495842387716, p-value: 2.038246352957472e-07\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: daily, Chi2: 25.68349005595362, p-value: 4.0225369687136106e-07\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: gain, Chi2: 24.833483293560278, p-value: 6.250229221882971e-07\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: discoloration, Chi2: 24.346591773069136, p-value: 8.046899002425923e-07\n",
      "단순 재입원 환자 Chi-square Feature Anaylsis - Word: lactate, Chi2: 24.19250190219783, p-value: 8.71709988479633e-07\n"
     ]
    }
   ],
   "source": [
    "vector_to_word = {tuple(word2vec[word]): word for word in word2vec.index_to_key}\n",
    "\n",
    "def get_word_from_vector(vector, vector_to_word):\n",
    "    vector_tuple = tuple(vector)\n",
    "    return vector_to_word.get(vector_tuple, \"없음\")\n",
    "\n",
    "correct_indices = np.where(np.array(all_test_labels) == np.array(all_test_predictions))[0]\n",
    "incorrect_indices = np.where(np.array(all_test_labels) != np.array(all_test_predictions))[0]\n",
    "\n",
    "positive_words = []\n",
    "negative_words = []\n",
    "exclude_words = {\".\", \"없음\"}\n",
    "\n",
    "for idx in tqdm(correct_indices):\n",
    "    embedding_vectors = test_dataloader.dataset[idx][0].cpu().numpy()\n",
    "    for vector in embedding_vectors : \n",
    "        words = get_word_from_vector(vector, vector_to_word)\n",
    "        positive_words.append(words)\n",
    "\n",
    "for idx in tqdm(incorrect_indices):\n",
    "    embedding_vectors = test_dataloader.dataset[idx][0].cpu().numpy()\n",
    "    for vector in embedding_vectors : \n",
    "        words = get_word_from_vector(vector, vector_to_word)\n",
    "        negative_words.append(words)\n",
    "\n",
    "filtered_positive_words = [word for word in positive_words if word not in exclude_words]\n",
    "filtered_negative_words = [word for word in negative_words if word not in exclude_words]\n",
    "\n",
    "positive_word_counts = Counter(filtered_positive_words)\n",
    "negative_word_counts = Counter(filtered_negative_words)\n",
    "\n",
    "chi2_results = {}\n",
    "all_words = set(filtered_positive_words).union(set(filtered_negative_words))\n",
    "\n",
    "for word in all_words:\n",
    "    pos_count = positive_word_counts[word]\n",
    "    neg_count = negative_word_counts[word]\n",
    "    pos_total = len(filtered_positive_words)\n",
    "    neg_total = len(filtered_negative_words)\n",
    "    \n",
    "    observed = [[pos_count, pos_total - pos_count],\n",
    "                [neg_count, neg_total - neg_count]]\n",
    "    \n",
    "    chi2, p, _, _ = chi2_contingency(observed)\n",
    "    chi2_results[word] = (chi2, p)\n",
    "\n",
    "sorted_chi2_results = sorted(chi2_results.items(), key=lambda item: item[1][0], reverse=True)[:20]\n",
    "\n",
    "for word, (chi2, p) in sorted_chi2_results :\n",
    "    print(f\"단순 재입원 환자 Chi-square Feature Anaylsis - Word: {word}, Chi2: {chi2}, p-value: {p}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 30일 이내 Readmission"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "Finishing last run (ID:wclf18qg) before initializing another..."
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e24fc76794ed4137961de21624bc4d27",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(Label(value='0.001 MB of 0.001 MB uploaded\\r'), FloatProgress(value=1.0, max=1.0)))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<style>\n",
       "    table.wandb td:nth-child(1) { padding: 0 10px; text-align: left ; width: auto;} td:nth-child(2) {text-align: left ; width: 100%}\n",
       "    .wandb-row { display: flex; flex-direction: row; flex-wrap: wrap; justify-content: flex-start; width: 100% }\n",
       "    .wandb-col { display: flex; flex-direction: column; flex-basis: 100%; flex: 1; padding: 10px; }\n",
       "    </style>\n",
       "<div class=\"wandb-row\"><div class=\"wandb-col\"><h3>Run history:</h3><br/><table class=\"wandb\"><tr><td>Accuracy</td><td>▃▁▂▄▅▄▅▅▆▅▇▆▇▇▆█▇▇▆▇</td></tr><tr><td>F1 Score</td><td>▁▅▅▅▄▃▆▇▄▇▇█▇▅██▆█▄▅</td></tr><tr><td>Precision</td><td>▅▁▂▄▅▅▅▄▇▄▆▄▆█▅▇▇▅▇█</td></tr><tr><td>Recall</td><td>▁▇▇▅▄▃▅█▃█▆█▆▃▇▆▄▇▃▃</td></tr><tr><td>Test loss</td><td>█▆▅▄▃▃▃▃▃▂▂▂▂▂▁▁▁▁▂▁</td></tr><tr><td>Train loss</td><td>█▇▆▆▅▅▄▄▄▄▃▃▃▂▂▂▂▁▁▁</td></tr></table><br/></div><div class=\"wandb-col\"><h3>Run summary:</h3><br/><table class=\"wandb\"><tr><td>Accuracy</td><td>0.71169</td></tr><tr><td>F1 Score</td><td>0.70081</td></tr><tr><td>Precision</td><td>0.66667</td></tr><tr><td>Recall</td><td>0.73864</td></tr><tr><td>Test loss</td><td>0.58661</td></tr><tr><td>Train loss</td><td>0.47407</td></tr></table><br/></div></div>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       " View run <strong style=\"color:#cdcd00\">only_readmission_1st_20000</strong> at: <a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp/runs/wclf18qg' target=\"_blank\">https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp/runs/wclf18qg</a><br/> View project at: <a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp' target=\"_blank\">https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp</a><br/>Synced 5 W&B file(s), 0 media file(s), 0 artifact file(s) and 0 other file(s)"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Find logs at: <code>.\\wandb\\run-20240709_123153-wclf18qg\\logs</code>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "The new W&B backend becomes opt-out in version 0.18.0; try it out with `wandb.require(\"core\")`! See https://wandb.me/wandb-core for more information."
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Successfully finished last run (ID:wclf18qg). Initializing new run:<br/>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5124e99c68fc42df884b38bc5e3bd8c6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(Label(value='Waiting for wandb.init()...\\r'), FloatProgress(value=0.011111111111111112, max=1.0…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Tracking run with wandb version 0.17.4"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Run data is saved locally in <code>c:\\Users\\gangmin\\dahs\\project\\wandb\\run-20240709_124508-37x6m32c</code>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "Syncing run <strong><a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp/runs/37x6m32c' target=\"_blank\">30days_readmission_1st_20000</a></strong> to <a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp' target=\"_blank\">Weights & Biases</a> (<a href='https://wandb.me/run' target=\"_blank\">docs</a>)<br/>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       " View project at <a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp' target=\"_blank\">https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp</a>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       " View run at <a href='https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp/runs/37x6m32c' target=\"_blank\">https://wandb.ai/kanovatio-Korea%20University7022/cnn_nlp/runs/37x6m32c</a>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 0/3000 [00:00<?, ?it/s]c:\\Users\\gangmin\\AppData\\Local\\miniconda3\\envs\\medical\\lib\\site-packages\\torch\\nn\\modules\\conv.py:306: UserWarning: Plan failed with a cudnnException: CUDNN_BACKEND_EXECUTION_PLAN_DESCRIPTOR: cudnnFinalize Descriptor Failed cudnn_status: CUDNN_STATUS_NOT_SUPPORTED (Triggered internally at C:\\cb\\pytorch_1000000000000\\work\\aten\\src\\ATen\\native\\cudnn\\Conv_v8.cpp:919.)\n",
      "  return F.conv1d(input, weight, bias, self.stride,\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1, Train Loss: 0.6931528145074844, Test Loss : 0.6890684068202972,  Accuracy: 0.5, Precision: 0.0, Recall: 0.0, F1 Score: 0.0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 1/3000 [00:18<15:07:51, 18.16s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 2, Train Loss: 0.6811825555562973, Test Loss : 0.6812518934408823,  Accuracy: 0.6235955056179775, Precision: 0.5945945945945946, Recall: 0.75, F1 Score: 0.6633165829145727\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 2/3000 [00:31<12:34:29, 15.10s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 3, Train Loss: 0.6674493536353111, Test Loss : 0.6794674644867579,  Accuracy: 0.5561797752808989, Precision: 0.5283018867924528, Recall: 0.9545454545454546, F1 Score: 0.6801619433198381\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 3/3000 [00:43<11:40:30, 14.02s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 4, Train Loss: 0.6599271216988564, Test Loss : 0.6692431767781576,  Accuracy: 0.6123595505617978, Precision: 0.5904761904761905, Recall: 0.7045454545454546, F1 Score: 0.6424870466321243\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 4/3000 [00:56<11:15:26, 13.53s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 5, Train Loss: 0.6436101284623146, Test Loss : 0.6694747706254324,  Accuracy: 0.5617977528089888, Precision: 0.6388888888888888, Recall: 0.26136363636363635, F1 Score: 0.3709677419354838\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 5/3000 [01:09<11:01:29, 13.25s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 6, Train Loss: 0.6340027981996537, Test Loss : 0.6675437887509664,  Accuracy: 0.5898876404494382, Precision: 0.6829268292682927, Recall: 0.3181818181818182, F1 Score: 0.43410852713178294\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 6/3000 [01:22<10:51:48, 13.06s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 7, Train Loss: 0.6216226014494896, Test Loss : 0.6582017888625463,  Accuracy: 0.6404494382022472, Precision: 0.6304347826086957, Recall: 0.6590909090909091, F1 Score: 0.6444444444444444\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 7/3000 [01:34<10:46:33, 12.96s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 8, Train Loss: 0.6155771872401238, Test Loss : 0.6550833135843277,  Accuracy: 0.6348314606741573, Precision: 0.6575342465753424, Recall: 0.5454545454545454, F1 Score: 0.5962732919254657\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 8/3000 [01:47<10:43:41, 12.91s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 9, Train Loss: 0.6015823899209499, Test Loss : 0.6591647416353226,  Accuracy: 0.5842696629213483, Precision: 0.547945205479452, Recall: 0.9090909090909091, F1 Score: 0.6837606837606837\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 9/3000 [02:00<10:39:58, 12.84s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 10, Train Loss: 0.5867339931428432, Test Loss : 0.6539326558510462,  Accuracy: 0.6123595505617978, Precision: 0.6557377049180327, Recall: 0.45454545454545453, F1 Score: 0.5369127516778524\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 10/3000 [02:13<10:38:13, 12.81s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 11, Train Loss: 0.578752840012312, Test Loss : 0.6495296458403269,  Accuracy: 0.6067415730337079, Precision: 0.575, Recall: 0.7840909090909091, F1 Score: 0.6634615384615384\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 11/3000 [02:25<10:36:33, 12.78s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 12, Train Loss: 0.5645960871875286, Test Loss : 0.6456436912218729,  Accuracy: 0.6292134831460674, Precision: 0.6018518518518519, Recall: 0.7386363636363636, F1 Score: 0.663265306122449\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 12/3000 [02:38<10:36:40, 12.78s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 13, Train Loss: 0.5545532323420048, Test Loss : 0.6462490459283193,  Accuracy: 0.6292134831460674, Precision: 0.6486486486486487, Recall: 0.5454545454545454, F1 Score: 0.5925925925925926\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 13/3000 [02:51<10:37:02, 12.80s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 14, Train Loss: 0.545673212558031, Test Loss : 0.6437655737002691,  Accuracy: 0.6404494382022472, Precision: 0.6132075471698113, Recall: 0.7386363636363636, F1 Score: 0.6701030927835051\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 14/3000 [03:04<10:35:43, 12.77s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 15, Train Loss: 0.532528159469366, Test Loss : 0.6444281389315923,  Accuracy: 0.6179775280898876, Precision: 0.6351351351351351, Recall: 0.5340909090909091, F1 Score: 0.5802469135802468\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 15/3000 [03:16<10:33:23, 12.73s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 16, Train Loss: 0.5233469368517398, Test Loss : 0.6403903712828954,  Accuracy: 0.6460674157303371, Precision: 0.6506024096385542, Recall: 0.6136363636363636, F1 Score: 0.631578947368421\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 16/3000 [03:29<10:32:44, 12.72s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 17, Train Loss: 0.5122062231600285, Test Loss : 0.6467373271783193,  Accuracy: 0.6179775280898876, Precision: 0.65625, Recall: 0.4772727272727273, F1 Score: 0.5526315789473685\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 17/3000 [03:42<10:32:12, 12.72s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 18, Train Loss: 0.49951554119586944, Test Loss : 0.6411334524552027,  Accuracy: 0.6460674157303371, Precision: 0.6190476190476191, Recall: 0.7386363636363636, F1 Score: 0.6735751295336787\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 18/3000 [03:54<10:31:44, 12.71s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 19, Train Loss: 0.4988689522445202, Test Loss : 0.6392611712217331,  Accuracy: 0.6573033707865169, Precision: 0.6483516483516484, Recall: 0.6704545454545454, F1 Score: 0.6592178770949721\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 19/3000 [04:07<10:32:36, 12.73s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 20, Train Loss: 0.4850210615992546, Test Loss : 0.6441791007916132,  Accuracy: 0.601123595505618, Precision: 0.5714285714285714, Recall: 0.7727272727272727, F1 Score: 0.6570048309178744\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 20/3000 [04:20<10:33:51, 12.76s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 21, Train Loss: 0.4783703704178333, Test Loss : 0.6390364716450373,  Accuracy: 0.6404494382022472, Precision: 0.6538461538461539, Recall: 0.5795454545454546, F1 Score: 0.6144578313253013\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 21/3000 [04:33<10:32:59, 12.75s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 22, Train Loss: 0.468686513453722, Test Loss : 0.6381220916906992,  Accuracy: 0.651685393258427, Precision: 0.625, Recall: 0.7386363636363636, F1 Score: 0.6770833333333333\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 22/3000 [04:45<10:31:31, 12.72s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 23, Train Loss: 0.45906116351485254, Test Loss : 0.6392822364966074,  Accuracy: 0.6629213483146067, Precision: 0.64, Recall: 0.7272727272727273, F1 Score: 0.6808510638297872\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 23/3000 [04:58<10:31:37, 12.73s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 24, Train Loss: 0.45448625266551973, Test Loss : 0.6385059108336767,  Accuracy: 0.6573033707865169, Precision: 0.6626506024096386, Recall: 0.625, F1 Score: 0.6432748538011697\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 24/3000 [05:11<10:31:12, 12.73s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 25, Train Loss: 0.45092905893921853, Test Loss : 0.6374810487031937,  Accuracy: 0.6348314606741573, Precision: 0.6263736263736264, Recall: 0.6477272727272727, F1 Score: 0.6368715083798883\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 25/3000 [05:24<10:32:31, 12.76s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 26, Train Loss: 0.4444343896210194, Test Loss : 0.6370650430520376,  Accuracy: 0.6573033707865169, Precision: 0.6421052631578947, Recall: 0.6931818181818182, F1 Score: 0.6666666666666667\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 26/3000 [05:36<10:31:35, 12.74s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 27, Train Loss: 0.4374312192201614, Test Loss : 0.6461252917846044,  Accuracy: 0.601123595505618, Precision: 0.6491228070175439, Recall: 0.42045454545454547, F1 Score: 0.510344827586207\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 27/3000 [05:49<10:31:02, 12.74s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 28, Train Loss: 0.43344648540019987, Test Loss : 0.6413228213787079,  Accuracy: 0.6179775280898876, Precision: 0.631578947368421, Recall: 0.5454545454545454, F1 Score: 0.5853658536585366\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 28/3000 [06:02<10:30:00, 12.72s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 29, Train Loss: 0.4237756736576557, Test Loss : 0.6382179806629816,  Accuracy: 0.6573033707865169, Precision: 0.6363636363636364, Recall: 0.7159090909090909, F1 Score: 0.6737967914438503\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 29/3000 [06:14<10:29:43, 12.72s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 30, Train Loss: 0.4213789837062359, Test Loss : 0.6409169832865397,  Accuracy: 0.5955056179775281, Precision: 0.6052631578947368, Recall: 0.5227272727272727, F1 Score: 0.5609756097560975\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 30/3000 [06:27<10:29:26, 12.72s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Early stopping 기능이 31에서 실행되었습니다.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 30/3000 [06:40<11:00:41, 13.35s/it]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "모델 학습 종료\n"
     ]
    }
   ],
   "source": [
    "model = CNN(embedding_dim, n_filters, filter_sizes, dropout, output_dim).to(device)\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = optim.Adam(model.parameters(), lr=learning_rate)\n",
    "\n",
    "best_loss = np.inf # 초기 설정\n",
    "patience = 5\n",
    "patience_counter = 0\n",
    "\n",
    "for epoch in tqdm(range(num_epochs)):\n",
    "    model.train()\n",
    "    \n",
    "    train_loss = 0.0\n",
    "\n",
    "    for inputs, labels in train_dataloader_30days :\n",
    "        inputs, labels = inputs.to(device), labels.to(device)\n",
    "        optimizer.zero_grad()\n",
    "\n",
    "        train_predictions = model(inputs)\n",
    "\n",
    "        loss = criterion(train_predictions, labels.long())\n",
    "        \n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        train_loss += loss.item()\n",
    "    \n",
    "    model.eval()\n",
    "    with torch.no_grad() : \n",
    "        all_test_labels = []\n",
    "        all_test_predictions = []\n",
    "        test_loss = 0.0\n",
    "\n",
    "        for inputs, labels in test_dataloader_30days : \n",
    "            inputs, labels = inputs.to(device), labels.to(device)\n",
    "            test_predictions = model(inputs)\n",
    "\n",
    "            loss = criterion(test_predictions, labels.long())\n",
    "            test_loss += loss.item()\n",
    "\n",
    "            all_test_labels.extend(labels.cpu().numpy())\n",
    "            all_test_predictions.extend(torch.argmax(test_predictions, dim=1).cpu().numpy())\n",
    "    \n",
    "    train_loss /= len(train_dataloader_30days)\n",
    "    test_loss /= len(test_dataloader_30days)\n",
    "\n",
    "    if test_loss < best_loss:\n",
    "        best_loss = test_loss\n",
    "        patience_counter = 0\n",
    "\n",
    "    else:\n",
    "        patience_counter += 1\n",
    "        if patience_counter >= patience:\n",
    "            print(f'Early stopping 기능이 {epoch+1}에서 실행되었습니다.')\n",
    "            break\n",
    "\n",
    "    all_labels = np.array(all_test_labels)\n",
    "    all_predictions = np.array(all_test_predictions)\n",
    "    \n",
    "    accuracy = accuracy_score(all_test_labels, all_test_predictions)\n",
    "    precision = precision_score(all_test_labels, all_test_predictions, average = 'binary')\n",
    "    recall = recall_score(all_test_labels, all_test_predictions, average = 'binary')\n",
    "    f1 = f1_score(all_test_labels, all_test_predictions, average = 'binary')\n",
    "    \n",
    "    print(f'Epoch {epoch+1}, Train Loss: {train_loss}, Test Loss : {test_loss},  Accuracy: {accuracy}, Precision: {precision}, Recall: {recall}, F1 Score: {f1}')\n",
    "\n",
    "print(\"모델 학습 종료\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 118/118 [00:17<00:00,  6.68it/s]\n",
      "100%|██████████| 60/60 [00:09<00:00,  6.59it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: trach, Chi2: 46.62999205861445, p-value: 8.573792163551856e-12\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: po, Chi2: 41.30128440189896, p-value: 1.3048312701248091e-10\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: mg, Chi2: 38.95454122019472, p-value: 4.3379001687301745e-10\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: peak, Chi2: 33.78739922891153, p-value: 6.147576391616123e-09\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: ventricular, Chi2: 29.78009106912895, p-value: 4.839362075932311e-08\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: pericardial, Chi2: 27.988698435437637, p-value: 1.2202603966380087e-07\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: tracheostomy, Chi2: 23.404957646635662, p-value: 1.3123987175502007e-06\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: collection, Chi2: 23.404957646635662, p-value: 1.3123987175502007e-06\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: right, Chi2: 23.147428286708926, p-value: 1.5004373923207648e-06\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: home, Chi2: 22.694390732508793, p-value: 1.8992147055339385e-06\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: valve, Chi2: 22.65546446489384, p-value: 1.938082456189709e-06\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: av, Chi2: 22.415055370894233, p-value: 2.1964549845960705e-06\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: linezolid, Chi2: 21.46428905394384, p-value: 3.6047971799533053e-06\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: type, Chi2: 21.24684216542601, p-value: 4.03775031866445e-06\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: tunneled, Chi2: 19.32480859413251, p-value: 1.102643610959448e-05\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: losartan, Chi2: 19.219361796542817, p-value: 1.1652552259514217e-05\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: tricuspid, Chi2: 18.615766025736203, p-value: 1.5989274377776327e-05\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: bid, Chi2: 18.16473886130103, p-value: 2.0259535145839572e-05\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: subdural, Chi2: 17.86236175011373, p-value: 2.3747160730023595e-05\n",
      "30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: virus, Chi2: 17.602425865489696, p-value: 2.7224085023785927e-05\n"
     ]
    }
   ],
   "source": [
    "vector_to_word = {tuple(word2vec[word]): word for word in word2vec.index_to_key}\n",
    "\n",
    "def get_word_from_vector(vector, vector_to_word):\n",
    "    vector_tuple = tuple(vector)\n",
    "    return vector_to_word.get(vector_tuple, \"없음\")\n",
    "\n",
    "correct_indices = np.where(np.array(all_test_labels) == np.array(all_test_predictions))[0]\n",
    "incorrect_indices = np.where(np.array(all_test_labels) != np.array(all_test_predictions))[0]\n",
    "\n",
    "positive_words = []\n",
    "negative_words = []\n",
    "exclude_words = {\".\", \"없음\"}\n",
    "\n",
    "for idx in tqdm(correct_indices):\n",
    "    embedding_vectors = test_dataloader.dataset[idx][0].cpu().numpy()\n",
    "    for vector in embedding_vectors : \n",
    "        words = get_word_from_vector(vector, vector_to_word)\n",
    "        positive_words.append(words)\n",
    "\n",
    "for idx in tqdm(incorrect_indices):\n",
    "    embedding_vectors = test_dataloader.dataset[idx][0].cpu().numpy()\n",
    "    for vector in embedding_vectors : \n",
    "        words = get_word_from_vector(vector, vector_to_word)\n",
    "        negative_words.append(words)\n",
    "\n",
    "filtered_positive_words = [word for word in positive_words if word not in exclude_words]\n",
    "filtered_negative_words = [word for word in negative_words if word not in exclude_words]\n",
    "\n",
    "positive_word_counts = Counter(filtered_positive_words)\n",
    "negative_word_counts = Counter(filtered_negative_words)\n",
    "\n",
    "chi2_results = {}\n",
    "all_words = set(filtered_positive_words).union(set(filtered_negative_words))\n",
    "\n",
    "for word in all_words:\n",
    "    pos_count = positive_word_counts[word]\n",
    "    neg_count = negative_word_counts[word]\n",
    "    pos_total = len(filtered_positive_words)\n",
    "    neg_total = len(filtered_negative_words)\n",
    "    \n",
    "    observed = [[pos_count, pos_total - pos_count],\n",
    "                [neg_count, neg_total - neg_count]]\n",
    "    \n",
    "    chi2, p, _, _ = chi2_contingency(observed)\n",
    "    chi2_results[word] = (chi2, p)\n",
    "\n",
    "sorted_chi2_results = sorted(chi2_results.items(), key=lambda item: item[1][0], reverse=True)[:20]\n",
    "\n",
    "for word, (chi2, p) in sorted_chi2_results :\n",
    "    print(f\"30일 이내 재입원 환자 Chi-square Feature Anaylsis - Word: {word}, Chi2: {chi2}, p-value: {p}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "medical",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.19"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}