Treinando Corpus VODAN.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "84bd8aec",
   "metadata": {},
   "source": [
    "# Treinando Corpus VODAN BR"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d6956083",
   "metadata": {},
   "source": [
    "## Importando bibliotecas"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "bb0536d3",
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "import re\n",
    "import unicodedata\n",
    "import multiprocessing\n",
    "import random\n",
    "\n",
    "import gensim\n",
    "from gensim.models import Word2Vec, KeyedVectors\n",
    "from gensim.similarities import WmdSimilarity\n",
    "from pyemd import emd\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "plt.rcParams['figure.figsize'] = [10, 5]\n",
    "\n",
    "import nltk\n",
    "from nltk.corpus import stopwords\n",
    "\n",
    "import numpy as np\n",
    "import scipy as sp\n",
    "import pandas as pd\n",
    "\n",
    "from sklearn.decomposition import TruncatedSVD\n",
    "from sklearn.decomposition import PCA\n",
    "\n",
    "np.random.seed(0)\n",
    "random.seed(0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "73c02e55",
   "metadata": {},
   "source": [
    "## Lendo o arquivo de entrada"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "a28fda79",
   "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>anamnese</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>#Admissão pela clinica médica \\r\\n# Paciente f...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Paciente proveniente da residência, foi atendi...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>EVOLUÇÃO FISIOTERAPIA - PLANTÃO DIURNO SEGUNDA...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Paciente do sexo feminino, 20 anos, acometida ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td># ADMISSÃO COORTE COVID-19 #\\r\\n\\r\\n-       41...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                            anamnese\n",
       "0  #Admissão pela clinica médica \\r\\n# Paciente f...\n",
       "1  Paciente proveniente da residência, foi atendi...\n",
       "2  EVOLUÇÃO FISIOTERAPIA - PLANTÃO DIURNO SEGUNDA...\n",
       "3  Paciente do sexo feminino, 20 anos, acometida ...\n",
       "4  # ADMISSÃO COORTE COVID-19 #\\r\\n\\r\\n-       41..."
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = pd.read_csv('Dados/anamnese.csv')\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "52454ccb",
   "metadata": {},
   "source": [
    "## Baixando stopwords em Português"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "3efbaca6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['de', 'a', 'o', 'que', 'e', 'é', 'do', 'da', 'em', 'um']"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "stopwords_list = stopwords.words('portuguese')\n",
    "stopwords_list[:10]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "28abcec3",
   "metadata": {},
   "source": [
    "## Limpando os dados"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "0a3911ca",
   "metadata": {},
   "outputs": [],
   "source": [
    "def clean_data(word):\n",
    "    word = word.lower()\n",
    "    word = re.sub(r'[^\\w\\s]','', word)\n",
    "    word = re.sub(r\"([0-9])\", r\" \", word)\n",
    "    \n",
    "    words = word.split() \n",
    "    \n",
    "    clean_words = [word for word in words if (word not in stopwords_list) and len(word) > 2]\n",
    "    \n",
    "    return \" \".join(clean_words)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "139e4caa",
   "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>anamnese</th>\n",
       "      <th>clean_text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>#Admissão pela clinica médica \\r\\n# Paciente f...</td>\n",
       "      <td>admissão clinica médica paciente feminina anos...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Paciente proveniente da residência, foi atendi...</td>\n",
       "      <td>paciente proveniente residência atendida spa i...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>EVOLUÇÃO FISIOTERAPIA - PLANTÃO DIURNO SEGUNDA...</td>\n",
       "      <td>evolução fisioterapia plantão diurno segunda f...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Paciente do sexo feminino, 20 anos, acometida ...</td>\n",
       "      <td>paciente sexo feminino anos acometida tipo hep...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td># ADMISSÃO COORTE COVID-19 #\\r\\n\\r\\n-       41...</td>\n",
       "      <td>admissão coorte covid anos srag suspeita covid...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                            anamnese  \\\n",
       "0  #Admissão pela clinica médica \\r\\n# Paciente f...   \n",
       "1  Paciente proveniente da residência, foi atendi...   \n",
       "2  EVOLUÇÃO FISIOTERAPIA - PLANTÃO DIURNO SEGUNDA...   \n",
       "3  Paciente do sexo feminino, 20 anos, acometida ...   \n",
       "4  # ADMISSÃO COORTE COVID-19 #\\r\\n\\r\\n-       41...   \n",
       "\n",
       "                                          clean_text  \n",
       "0  admissão clinica médica paciente feminina anos...  \n",
       "1  paciente proveniente residência atendida spa i...  \n",
       "2  evolução fisioterapia plantão diurno segunda f...  \n",
       "3  paciente sexo feminino anos acometida tipo hep...  \n",
       "4  admissão coorte covid anos srag suspeita covid...  "
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "clean_text = []\n",
    "\n",
    "for text in data['anamnese']:\n",
    "    clean_text.append(clean_data(text))\n",
    "\n",
    "data['clean_text'] = clean_text\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "cedd6ae3",
   "metadata": {},
   "outputs": [],
   "source": [
    "data_clean_text = data['clean_text'].values.tolist()\n",
    "\n",
    "word_split = []\n",
    "for word in data_clean_text:\n",
    "    word_split.append(word.split(' '))\n",
    "\n",
    "data_clean_text = word_split"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b9ee00ab",
   "metadata": {},
   "source": [
    "## Selecionando apenas palavras distintas"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "4344237d",
   "metadata": {},
   "outputs": [],
   "source": [
    "def distinct_words(corpus):\n",
    "    corpus_words = []\n",
    "    num_corpus_words = -1\n",
    "    \n",
    "    corpus_words =  sorted(list(set([word for sentence in corpus for word in sentence])))\n",
    "    num_corpus_words = len(corpus_words)\n",
    "\n",
    "    return corpus_words, num_corpus_words"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ac711b6f",
   "metadata": {},
   "source": [
    "## Calculando matriz de co-ocorrência"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "5e2560a0",
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_co_occurrence_matrix(corpus, window_size = 5):\n",
    "    words, num_words = distinct_words(corpus)\n",
    "    M = np.zeros((num_words, num_words))\n",
    "    word2ind = dict([(word, index) for index, word in enumerate(words)])\n",
    "    \n",
    "    for sentence in corpus:\n",
    "        current_index = 0\n",
    "        sentence_len = len(sentence)\n",
    "        indices = [word2ind[i] for i in sentence]\n",
    "        \n",
    "        while current_index < sentence_len:\n",
    "            left  = max(current_index - window_size, 0)\n",
    "            right = min(current_index + window_size + 1, sentence_len) \n",
    "            \n",
    "            current_word = sentence[current_index]\n",
    "            current_word_index = word2ind[current_word]\n",
    "            words_around = indices[left:current_index] + indices[current_index+1:right]\n",
    "            \n",
    "            for ind in words_around:\n",
    "                M[current_word_index, ind] += 1\n",
    "            \n",
    "            current_index += 1\n",
    "\n",
    "    return M, word2ind"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a2dce144",
   "metadata": {},
   "source": [
    "## Reduzindo a matriz de coorrência para k-dimensão"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "c0fd9dbf",
   "metadata": {},
   "outputs": [],
   "source": [
    "def reduce_to_k_dim(M, k = 2):\n",
    "    n_iters = 10    \n",
    "    M_reduced = None\n",
    "    \n",
    "    print(\"Running Truncated SVD over %i words...\" % (M.shape[0]))\n",
    "    \n",
    "    TSVD = TruncatedSVD(n_components = k, n_iter = n_iters)\n",
    "    M_reduced = TSVD.fit_transform(M)\n",
    "    \n",
    "    print(\"Done.\")\n",
    "    return M_reduced"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fa9c4275",
   "metadata": {},
   "source": [
    "## Plotando embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "290163d1",
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_embeddings(M_reduced, word2ind, words):\n",
    "    for word in words:\n",
    "        index = word2ind[word]\n",
    "        embedding = M_reduced[index]\n",
    "        \n",
    "        x, y  = embedding[0], embedding[1]\n",
    "        \n",
    "        plt.scatter(x, y, marker='x', color='red')\n",
    "        plt.text(x, y, word, fontsize=9)\n",
    "    \n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "f49d80f7",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Running Truncated SVD over 1846 words...\n",
      "Done.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "M_co_occurrence, word2ind_co_occurrence = compute_co_occurrence_matrix(data_clean_text)\n",
    "M_reduced_co_occurrence = reduce_to_k_dim(M_co_occurrence)\n",
    "\n",
    "M_lengths = np.linalg.norm(M_reduced_co_occurrence, axis = 1)\n",
    "M_normalized = M_reduced_co_occurrence / M_lengths[:, np.newaxis]\n",
    "\n",
    "words = [\n",
    "    'paciente', 'covid', 'feminino', 'masculino',\n",
    "    'febre', 'obesa', 'tabagismo', 'diabetes', 'tosse', 'hepatopata', 'hipertensão',\n",
    "    'fisioterapia', 'enfermaria', 'internação', 'cti'\n",
    "]\n",
    "\n",
    "plot_embeddings(M_normalized, word2ind_co_occurrence, words)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ffdc5eb4",
   "metadata": {},
   "source": [
    "## Criando um modelo vazio"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "2039125b",
   "metadata": {},
   "outputs": [],
   "source": [
    "cores = multiprocessing.cpu_count()\n",
    "\n",
    "model = Word2Vec(\n",
    "    min_count = 10,\n",
    "    window = 5,\n",
    "    workers = cores - 1,\n",
    "    max_vocab_size = 100000\n",
    ")\n",
    "\n",
    "model.build_vocab(data_clean_text)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5d6922a0",
   "metadata": {},
   "source": [
    "## Treinando o modelo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "757c013f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(55317, 306750)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.train(\n",
    "    data_clean_text,\n",
    "    total_examples = model.corpus_count,\n",
    "    epochs = 50\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "07b6b3bf",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['paciente',\n",
       " 'anos',\n",
       " 'covid',\n",
       " 'nega',\n",
       " 'dias',\n",
       " 'uso',\n",
       " 'febre',\n",
       " 'conduta',\n",
       " 'mmhg',\n",
       " 'dia']"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "words = list(model.wv.index_to_key)\n",
    "words[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "8734a2b4",
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "model.save('Anamnese')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "415d9226",
   "metadata": {},
   "outputs": [],
   "source": [
    "trained_model = gensim.models.Word2Vec.load('Anamnese')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "02b13871",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('sintomas', 0.9965553879737854),\n",
       " ('covid', 0.9942978620529175),\n",
       " ('febre', 0.993326723575592),\n",
       " ('seca', 0.9928711652755737),\n",
       " ('tosse', 0.9925109148025513),\n",
       " ('início', 0.9917367100715637),\n",
       " ('quadro', 0.991161584854126),\n",
       " ('cti', 0.991116464138031),\n",
       " ('hospital', 0.9907221794128418),\n",
       " ('tratamento', 0.9902929067611694)]"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trained_model.wv.most_similar('paciente')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "f553d2cb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('feminino', 0.9956920742988586),\n",
       " ('sexo', 0.9948996901512146),\n",
       " ('cti', 0.992834210395813),\n",
       " ('suspeita', 0.9893884658813477),\n",
       " ('enfermaria', 0.9888049364089966),\n",
       " ('internação', 0.988092303276062),\n",
       " ('síndrome', 0.9868777990341187),\n",
       " ('admitido', 0.9861096739768982),\n",
       " ('plantão', 0.9860779643058777),\n",
       " ('proveniente', 0.9812664985656738)]"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trained_model.wv.most_similar(positive = ['paciente', 'covid'], negative = ['febre'])"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}