3_CNN_Att_SemEval.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "9OhptODec0Vo"
   },
   "source": [
    "# Attention et CNN pour l'extraction de relation\n",
    "\n",
    "Dans ce notebook nous allons implémenter un modèle CNN auquel s'ajoute le mécanisme d'attention. Ce modèle est inspiré de [cet article](https://www.aclweb.org/anthology/C16-1238.pdf), néanmoins contrairement à ce qui a été fait dans l'article, nous n'utiliserons pas les tags POS.\n",
    "\n",
    "## Préparation des données\n",
    "\n",
    "### Import des données \n",
    "\n",
    "On charge les données comme dans le notebook précédent. On charge les tokens, les relations ainsi que les positions des entités dans la phrases."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "id": "nFao71Pmc-H_"
   },
   "outputs": [],
   "source": [
    "my_path = \"data_RE\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 68
    },
    "id": "Y7R2_iJWc0WD",
    "outputId": "db10c417-3139-4797-db9a-b714164c838a"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 8000/8000 [00:00<00:00, 9916.52it/s]\n",
      "100%|██████████| 2717/2717 [00:00<00:00, 10994.11it/s]\n",
      "100%|██████████| 2717/2717 [00:00<00:00, 11203.50it/s]\n"
     ]
    }
   ],
   "source": [
    "from tqdm import tqdm\n",
    "import os\n",
    "import json\n",
    "import torch\n",
    "import numpy as np\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "import re\n",
    "import argparse\n",
    "import torch\n",
    "import os\n",
    "import random\n",
    "import json\n",
    "import numpy as np\n",
    "def sentence_process(sentence):\n",
    "    e1 = re.findall(r'<e1>(.*)</e1>', sentence)[0]\n",
    "    e2 = re.findall(r'<e2>(.*)</e2>', sentence)[0]\n",
    "    sentence = sentence.replace('<e1>' + e1 + '</e1>', ' <e1> ' + e1 + ' </e1> ', 1)\n",
    "    sentence = sentence.replace('<e2>' + e2 + '</e2>', ' <e2> ' + e2 + ' </e2> ', 1)\n",
    "    sentence = ''.join(sentence)\n",
    "    sentence = sentence.replace('< e1 >', '<e1>')\n",
    "    sentence = sentence.replace('< e2 >', '<e2>')\n",
    "    sentence = sentence.replace('< /e1 >', '</e1>')\n",
    "    sentence = sentence.replace('< /e2 >', '</e2>')\n",
    "    #sentence = sentence\n",
    "    token = sentence.split()\n",
    " \n",
    "\n",
    "    assert '<e1>' in sentence\n",
    "    assert '<e2>' in sentence\n",
    "    assert '</e1>' in sentence\n",
    "    assert '</e2>' in sentence\n",
    "\n",
    "    return  remove_postion_indicators(token)\n",
    "\n",
    "\n",
    "def remove_postion_indicators(token):\n",
    "    subj_start = subj_end = obj_start = obj_end = 0\n",
    "    pure_token = []\n",
    "    pure_pos = []\n",
    "    for i, word in enumerate(token):\n",
    "        if '<e1>' == word:\n",
    "            subj_start = len(pure_token)\n",
    "            continue\n",
    "        if '</e1>' == word:\n",
    "            subj_end = len(pure_token) - 1\n",
    "            continue\n",
    "        if '<e2>' in word:\n",
    "            obj_start = len(pure_token)\n",
    "            continue\n",
    "        if '</e2>' in word:\n",
    "            obj_end = len(pure_token) - 1\n",
    "            continue\n",
    "        pure_token.append(word)\n",
    "       \n",
    "    res = dict(\n",
    "        token=pure_token,\n",
    "        subj_start=subj_start,\n",
    "        subj_end=subj_end,\n",
    "        obj_start=obj_start,\n",
    "        obj_end=obj_end\n",
    "    )\n",
    "\n",
    "    return res\n",
    "\n",
    "def convert(src_file, des_file):\n",
    "    with open(src_file, 'r', encoding='utf-8') as fr:\n",
    "        file_data = fr.readlines()\n",
    "\n",
    "    with open(des_file, 'w', encoding='utf-8') as fw:\n",
    "        for i in tqdm(range(0, len(file_data), 4)):\n",
    "            meta = {}\n",
    "            s = file_data[i].strip().split('\\t')\n",
    "            assert len(s) == 2\n",
    "            meta['id'] = s[0]\n",
    "            \n",
    "            meta['relation'] = file_data[i+1].strip()\n",
    "        \n",
    "            meta['comment'] = file_data[i+2].strip()\n",
    "       \n",
    "            sen_res = sentence_process(s[1])\n",
    "         \n",
    "           \n",
    "            json.dump({**meta, **sen_res}, fw, ensure_ascii=False)\n",
    "            fw.write('\\n')\n",
    "\n",
    "\n",
    "\n",
    "path_train = my_path + '/TRAIN_FILE.TXT'\n",
    "path_test = my_path +'/TEST_FILE_FULL.TXT'\n",
    "path_valid = my_path +'/TEST_FILE_FULL.TXT'\n",
    "\n",
    "convert(path_train, my_path + '/train.json')\n",
    "convert(path_test, my_path + '/test.json')\n",
    "convert(path_test, my_path + '/valid.json')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "vJjq1Qo5c0Xq"
   },
   "source": [
    "### Import du Word Embedding pré-entraîner"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "id": "WoCj_tZrc0Xs"
   },
   "outputs": [],
   "source": [
    "class WordEmbeddingLoader(object):\n",
    "\n",
    "\n",
    "    def __init__(self, embedding_path, word_dim, min_freq):\n",
    "        self.embedding_path = embedding_path  \n",
    "        self.embedding_dim = word_dim \n",
    "        self.data_dir = my_path\n",
    "        self.min_freq = min_freq\n",
    "\n",
    "    def __build_vocab(self):\n",
    "        vocab = {}\n",
    "        #filename = ['train.json', 'test.json']\n",
    "        filename = ['train.json']\n",
    "        for fn in filename:\n",
    "            with open(os.path.join(self.data_dir, fn), 'r', encoding='utf-8') as fr:\n",
    "                for line in fr:\n",
    "                    line = json.loads(line.strip())\n",
    "                    sentence = line['token']\n",
    "                    for token in sentence:\n",
    "                        token = token.lower()\n",
    "                        vocab[token] = vocab.get(token, 0) + 1\n",
    "        vocab = set([token for token in vocab if vocab[token] > self.min_freq])\n",
    "        return vocab\n",
    "\n",
    "    def load_embedding(self):\n",
    "        vocab = self.__build_vocab()\n",
    "        token2id = {}\n",
    "        token2id['PAD'] = len(token2id) #0\n",
    "        token2id['UNK'] = len(token2id) #1\n",
    "        token_emb = []\n",
    "        with open(self.embedding_path, 'r', encoding='utf-8') as fr:\n",
    "            for line in fr:\n",
    "                line = line.strip().split()\n",
    "                if len(line) != self.embedding_dim + 1:\n",
    "                    continue\n",
    "                if line[0] not in vocab:\n",
    "                    continue\n",
    "                token2id[line[0]] = len(token2id)\n",
    "                token_emb.append(np.asarray(line[1:], dtype=np.float32))\n",
    "        token_emb = np.stack(token_emb).reshape(-1, self.embedding_dim)\n",
    "        \n",
    "        # pour les caractères spéciaux, les embeddings sont une distribution normale aléatoire\n",
    "        special_emb = np.random.uniform(-0.1, 0.1, size=(2, self.embedding_dim))\n",
    "        \n",
    "        token_emb = np.concatenate((special_emb, token_emb), axis=0)\n",
    "        token_emb = token_emb.astype(np.float32).reshape(-1, self.embedding_dim)\n",
    "        token_emb = torch.from_numpy(token_emb)\n",
    "        return token2id, token_emb\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "id": "8Q_QJMJ7c0X0"
   },
   "outputs": [
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-5-647fb76e00dd>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0mMIN_FREQ\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 5\u001b[0;31m \u001b[0mword2id\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mword_vec\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mWordEmbeddingLoader\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mEMBEDDING_PATH\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mWORD_DIM\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mMIN_FREQ\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mload_embedding\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m<ipython-input-4-db1b1f6bb651>\u001b[0m in \u001b[0;36mload_embedding\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m     30\u001b[0m         \u001b[0mtoken_emb\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     31\u001b[0m         \u001b[0;32mwith\u001b[0m \u001b[0mopen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0membedding_path\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'r'\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mencoding\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'utf-8'\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0mfr\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 32\u001b[0;31m             \u001b[0;32mfor\u001b[0m \u001b[0mline\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mfr\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     33\u001b[0m                 \u001b[0mline\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mline\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstrip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msplit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     34\u001b[0m                 \u001b[0;32mif\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mline\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m!=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0membedding_dim\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "EMBEDDING_PATH = my_path +'/embeddings/glove.6B.300d.txt'\n",
    "WORD_DIM = 300\n",
    "MIN_FREQ = 1\n",
    "\n",
    "word2id, word_vec = WordEmbeddingLoader(EMBEDDING_PATH, WORD_DIM, MIN_FREQ).load_embedding()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "sca_cvRqc0X6"
   },
   "source": [
    "### Import des relations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "fmIEGBpzc0X7"
   },
   "outputs": [],
   "source": [
    "class RelationLoader(object):\n",
    "    def __init__(self):\n",
    "        self.data_dir = my_path \n",
    "\n",
    "    def __load_relation(self):\n",
    "        relation_file = os.path.join(self.data_dir, 'relation2id.txt')\n",
    "        rel2id = {}\n",
    "        id2rel = {}\n",
    "        with open(relation_file, 'r', encoding='utf-8') as fr:\n",
    "            for line in fr:\n",
    "                relation, id_s = line.strip().split()\n",
    "                id_d = int(id_s)\n",
    "                rel2id[relation] = id_d\n",
    "                id2rel[id_d] = relation\n",
    "        return rel2id, id2rel, len(rel2id)\n",
    "\n",
    "    def get_relation(self):\n",
    "        return self.__load_relation()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "m5PRZCbmc0YB"
   },
   "outputs": [],
   "source": [
    "rel2id, id2rel, class_num = RelationLoader().get_relation()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 119
    },
    "id": "9LrnzGXHc0YI",
    "outputId": "9f77ec6a-b623-4ec3-dd21-94c18cdcd41e"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'<pad>': 0,\n",
       " 'Cause-Effect(e1,e2)': 1,\n",
       " 'Cause-Effect(e2,e1)': 2,\n",
       " 'Component-Whole(e1,e2)': 3,\n",
       " 'Component-Whole(e2,e1)': 4,\n",
       " 'Content-Container(e1,e2)': 5,\n",
       " 'Content-Container(e2,e1)': 6,\n",
       " 'Entity-Destination(e1,e2)': 7,\n",
       " 'Entity-Destination(e2,e1)': 8,\n",
       " 'Entity-Origin(e1,e2)': 9,\n",
       " 'Entity-Origin(e2,e1)': 10,\n",
       " 'Instrument-Agency(e1,e2)': 11,\n",
       " 'Instrument-Agency(e2,e1)': 12,\n",
       " 'Member-Collection(e1,e2)': 13,\n",
       " 'Member-Collection(e2,e1)': 14,\n",
       " 'Message-Topic(e1,e2)': 15,\n",
       " 'Message-Topic(e2,e1)': 16,\n",
       " 'Other': 17,\n",
       " 'Product-Producer(e1,e2)': 18,\n",
       " 'Product-Producer(e2,e1)': 19}"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rel2id"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "r2JNrwe0c0YO"
   },
   "source": [
    "### Traitement des données et itérateurs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "id": "JJAYXDAZc0YR"
   },
   "outputs": [],
   "source": [
    "class BrevetsDateset(Dataset):\n",
    "    def __init__(self, filename, rel2id, word2id, max_len, pos_dis):\n",
    "        self.filename = filename\n",
    "        self.rel2id = rel2id\n",
    "        self.word2id = word2id\n",
    "        self.max_len = max_len\n",
    "        self.pos_dis = pos_dis\n",
    "        self.data_dir = my_path \n",
    "        self.dataset, self.label = self.__load_data()\n",
    "\n",
    "    # Position des mots par rapport aux entités e1 et e2\n",
    "    def __get_pos_index(self, x):\n",
    "        if x < -self.pos_dis:\n",
    "            return 0\n",
    "        if x >= -self.pos_dis and x <= self.pos_dis:\n",
    "            return x + self.pos_dis + 1\n",
    "        if x > self.pos_dis:\n",
    "            return 2 * self.pos_dis + 2\n",
    "\n",
    "    def __get_relative_pos(self, x, entity_pos):\n",
    "        if x < entity_pos[0]:\n",
    "            return self.__get_pos_index(x-entity_pos[0])\n",
    "        elif x > entity_pos[1]:\n",
    "            return self.__get_pos_index(x-entity_pos[1])\n",
    "        else:\n",
    "            return self.__get_pos_index(0)\n",
    "\n",
    "    def __symbolize_sentence(self, e1_pos, e2_pos, sentence):\n",
    "        \"\"\"\n",
    "            Args:\n",
    "                e1_pos (tuple) span of e1\n",
    "                e2_pos (tuple) span of e2\n",
    "                sentence (list)\n",
    "             \n",
    "        \"\"\"\n",
    "        # mask vaut 2 entre les deux entités, 3 après la dernière entité, et 1 avant la première\n",
    "        mask = [1] * len(sentence)\n",
    "        if e1_pos[0] < e2_pos[0]:\n",
    "            for i in range(e1_pos[0], e2_pos[1]+1):\n",
    "                mask[i] = 2\n",
    "            for i in range(e2_pos[1]+1, len(sentence)):\n",
    "                mask[i] = 3\n",
    "        else:\n",
    "            for i in range(e2_pos[0], e1_pos[1]+1):\n",
    "                mask[i] = 2\n",
    "            for i in range(e1_pos[1]+1, len(sentence)):\n",
    "                mask[i] = 3\n",
    "        length = min(self.max_len, len(sentence))\n",
    "        \n",
    "        # e1_mask, e2_mask valent 1 à l'endroit de la phrase où l'entité 1 (resp 2) est présente \n",
    "        e1_mask = [0] * self.max_len\n",
    "        e2_mask = [0] * self.max_len\n",
    "        for i in range(e1_pos[0], e1_pos[1]+1):\n",
    "            if i < length:\n",
    "                e1_mask[i] = 1\n",
    "        for i in range(e2_pos[0], e2_pos[1]+1):\n",
    "            if i < length:\n",
    "                e2_mask[i] = 1\n",
    "\n",
    "        words = []\n",
    "        pos1 = []\n",
    "        pos2 = []\n",
    "       \n",
    "        mask = mask[:length]\n",
    "\n",
    "        for i in range(length):\n",
    "            words.append(self.word2id.get(sentence[i].lower(), self.word2id['UNK']))\n",
    "            pos1.append(self.__get_relative_pos(i, e1_pos))\n",
    "            pos2.append(self.__get_relative_pos(i, e2_pos))\n",
    "\n",
    "        if length < self.max_len:\n",
    "            for i in range(length, self.max_len):\n",
    "                mask.append(0)  # 'PAD' mask vaut 0\n",
    "                words.append(self.word2id['PAD'])\n",
    "\n",
    "                pos1.append(self.__get_relative_pos(i, e1_pos))\n",
    "                pos2.append(self.__get_relative_pos(i, e2_pos))\n",
    "       \n",
    "        unit = np.asarray([words, pos1, pos2, mask, e1_mask, e2_mask], dtype=np.int64)\n",
    "        unit = np.reshape(unit, newshape=(1, 6, self.max_len))\n",
    "        return unit\n",
    "\n",
    "    def __load_data(self):\n",
    "        path_data_file = os.path.join(self.data_dir, self.filename)\n",
    "        data = []\n",
    "        labels = []\n",
    "        with open(path_data_file, 'r', encoding='utf-8') as fr:\n",
    "            for line in fr:\n",
    "                line = json.loads(line.strip())\n",
    "                label = line['relation']\n",
    "                sentence = line['token']\n",
    "                e1_pos = (line['subj_start'], line['subj_end'])\n",
    "                e2_pos = (line['obj_start'], line['obj_end'])\n",
    "                label_idx = self.rel2id[label]\n",
    "\n",
    "                one_sentence = self.__symbolize_sentence(e1_pos, e2_pos, sentence)\n",
    "                data.append(one_sentence)\n",
    "                labels.append(label_idx)\n",
    "        \n",
    "        return data, labels\n",
    "\n",
    "    def __getitem__(self, index):\n",
    "        data = self.dataset[index]\n",
    "        label = self.label[index]\n",
    "        return data, label\n",
    "\n",
    "    def __len__(self):\n",
    "        return len(self.label)\n",
    "    \n",
    "class BrevetsDataLoader(object):\n",
    "    def __init__(self, rel2id, word2id, batch_size, max_len, pos_dis):\n",
    "        self.rel2id = rel2id\n",
    "        self.word2id = word2id\n",
    "        self.batch_size = batch_size\n",
    "        self.max_len = max_len\n",
    "        self.pos_dis = pos_dis\n",
    "    def __collate_fn(self, batch):\n",
    "        data, label = zip(*batch)  # unzip the batch data\n",
    "        data = list(data)\n",
    "        label = list(label)\n",
    "        data = torch.from_numpy(np.concatenate(data, axis=0))\n",
    "        label = torch.from_numpy(np.asarray(label, dtype=np.int64))\n",
    "        return data, label\n",
    "\n",
    "    def __get_data(self, filename, shuffle=False):\n",
    "        dataset = BrevetsDateset(filename, self.rel2id, self.word2id, self.max_len, self.pos_dis)\n",
    "        loader = DataLoader(\n",
    "            dataset=dataset,\n",
    "            batch_size=self.batch_size,\n",
    "            shuffle=shuffle,\n",
    "            num_workers=2,\n",
    "            collate_fn=self.__collate_fn\n",
    "        )\n",
    "        return loader\n",
    "\n",
    "    def get_train(self):\n",
    "        return self.__get_data('train.json', shuffle=True)\n",
    "\n",
    "    def get_dev(self):\n",
    "        return self.__get_data('valid.json', shuffle=False)\n",
    "\n",
    "    def get_test(self):\n",
    "        return self.__get_data('test.json', shuffle=False)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "id": "-wdMngGHc0YW"
   },
   "outputs": [],
   "source": [
    "REL2ID = rel2id\n",
    "WORD2ID = word2id\n",
    "BATCH_SIZE = 32\n",
    "MAX_LEN = 500\n",
    "POS_DIS = 50\n",
    "\n",
    "loader = BrevetsDataLoader(REL2ID, WORD2ID, BATCH_SIZE, MAX_LEN, POS_DIS)\n",
    "\n",
    "test_loader = loader.get_test()\n",
    "train_loader = loader.get_train()\n",
    "valid_loader = loader.get_dev()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "dCmU5BZ3c0Yc"
   },
   "source": [
    "## Modèle\n",
    "\n",
    "Le modèle implémenté utilise la convolution et un modèle d'attention qui sont illustrés ci-dessous.\n",
    "\n",
    "\n",
    " - Architecture générale : \n",
    " \n",
    "![alt text](images/cnnatt2.png)\n",
    "\n",
    " - CNN : \n",
    " \n",
    "![alt text](images/cnnatt1.png)\n",
    "\n",
    " - Attention :\n",
    " \n",
    "![alt text](images/cnnatt3.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "id": "PFFUqTGgc0Yd"
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "from torch.nn import init\n",
    "\n",
    "\n",
    "class Attention_CNN(nn.Module):\n",
    "    def __init__(self, word_vec, class_num, max_len, word_dim, pos_dim, pos_dis, dropout, filter_num, window):\n",
    "        super().__init__()\n",
    "        self.word_vec = word_vec\n",
    "        self.class_num = class_num\n",
    "\n",
    "        self.max_len = max_len\n",
    "        self.word_dim = word_dim\n",
    "        self.pos_dim = pos_dim\n",
    "        self.pos_dis = pos_dis\n",
    "     \n",
    "        self.dropout_value = dropout\n",
    "        self.filter_num = filter_num\n",
    "        self.window = window\n",
    "\n",
    "        self.dim = self.word_dim + 2 * self.pos_dim \n",
    "\n",
    "        self.word_embedding = nn.Embedding.from_pretrained(\n",
    "            embeddings=self.word_vec,\n",
    "            freeze=False,\n",
    "        )\n",
    "        self.pos1_embedding = nn.Embedding(\n",
    "            num_embeddings=2 * self.pos_dis + 3,\n",
    "            embedding_dim=self.pos_dim\n",
    "        )\n",
    "        self.pos2_embedding = nn.Embedding(\n",
    "            num_embeddings=2 * self.pos_dis + 3,\n",
    "            embedding_dim=self.pos_dim\n",
    "        )\n",
    "        self.conv = nn.Conv2d(\n",
    "            in_channels=1,\n",
    "            out_channels=self.filter_num,\n",
    "            kernel_size=(self.window, self.dim),\n",
    "            stride=(1, 1),\n",
    "            bias=True,\n",
    "            padding=(1, 0), \n",
    "            padding_mode='zeros'\n",
    "        )\n",
    "        self.maxpool = nn.MaxPool2d((self.max_len, 1))\n",
    "        self.tanh = nn.Tanh()\n",
    "        self.we = nn.Linear(\n",
    "            in_features=self.dim * 2,\n",
    "            out_features=self.dim * 2,\n",
    "            bias=True\n",
    "        )\n",
    "        self.wa = nn.Linear(\n",
    "            in_features=self.dim*2,\n",
    "            out_features=1,\n",
    "            bias=True\n",
    "        )\n",
    "        \n",
    "        self.dense = nn.Linear(\n",
    "            in_features=self.filter_num + 2 * self.dim,\n",
    "            out_features=self.class_num,\n",
    "            bias=True\n",
    "        )\n",
    "        # initialise les poids\n",
    "        init.uniform_(self.pos1_embedding.weight, a=-0.1, b=0.1)\n",
    "        init.uniform_(self.pos2_embedding.weight, a=-0.1, b=0.1)\n",
    "        init.uniform_(self.conv.weight, a=-0.1, b=0.1)\n",
    "        init.constant_(self.conv.bias, 0.)\n",
    "        init.uniform_(self.we.weight, a=-0.1, b=0.1)\n",
    "        init.constant_(self.we.bias, 0.)\n",
    "        init.uniform_(self.wa.weight, a=-0.1, b=0.1)\n",
    "        init.constant_(self.wa.bias, 0.)\n",
    "        init.uniform_(self.dense.weight, a=-0.1, b=0.1)\n",
    "        init.constant_(self.dense.bias, 0.)\n",
    "\n",
    "    # couche embedding\n",
    "    def encoder_layer(self, token, pos1, pos2):\n",
    "        \n",
    "        word_emb = self.word_embedding(token) \n",
    "        pos1_emb = self.pos1_embedding(pos1)\n",
    "        pos2_emb = self.pos2_embedding(pos2)\n",
    "   \n",
    "        emb = torch.cat(tensors=[word_emb, pos1_emb, pos2_emb], dim=-1)\n",
    "        # emb : [BATCH_SIZE x MAX_LEN x (WORD_dim + 2*POS_DIM)]\n",
    "        return emb \n",
    "\n",
    "    # couche cnn\n",
    "    def conv_layer(self, emb, mask):\n",
    "        emb = emb.unsqueeze(dim=1)  \n",
    "        conv = self.conv(emb)\n",
    "        # mask, supprimer les effets de PAD'\n",
    "        conv = conv.view(-1, self.filter_num, self.max_len)\n",
    "        mask = mask.unsqueeze(dim=1)\n",
    "        \n",
    "        mask = mask.expand(-1, self.filter_num, -1)        \n",
    "        conv = conv.masked_fill_(mask.eq(0), float('-inf')) \n",
    "        conv = conv.unsqueeze(dim=-1)\n",
    "        return conv\n",
    "\n",
    "    def single_maxpool_layer(self, conv):\n",
    "        pool = self.maxpool(conv)\n",
    "        pool = pool.view(-1, self.filter_num)\n",
    "        return pool\n",
    "\n",
    "\n",
    "    def entity_average(self, emb, e_mask):\n",
    "        lengths = torch.sum(e_mask.eq(1), dim=-1).view(-1, 1)  \n",
    "        mask = e_mask.unsqueeze(dim=1).float() \n",
    "        sum_emb = torch.bmm(mask, emb).squeeze(dim=1)  \n",
    "        avg_emb = sum_emb / lengths\n",
    "        # avg_emb : [BATCH_SIZE x (WORD_dim + 2*POS_DIM)]\n",
    "        return avg_emb\n",
    "    \n",
    "\n",
    "\n",
    "    def attention_layer(self, emb, entity, mask):\n",
    "        entity = entity.unsqueeze(dim=1).expand(-1, self.max_len, -1)\n",
    "        h = torch.cat(tensors=[emb, entity], dim=-1)  \n",
    "        h_flat = h.view(-1, 2*self.dim)  \n",
    "        output = self.tanh(self.we(h_flat))  \n",
    "        u_flat = self.wa(output)  \n",
    "        u = u_flat.view(-1, self.max_len)  \n",
    "        # supprimer les effets de PAD'\n",
    "        att_score = u.masked_fill(mask.eq(0), float('-inf'))  \n",
    "        att_weight = F.softmax(att_score, dim=-1).unsqueeze(dim=-1) \n",
    "        reps = torch.bmm(emb.transpose(1, 2), att_weight).squeeze(dim=-1) \n",
    "        # reps : [BATCH_SIZE x (WORD_dim + 2*POS_DIM)]\n",
    "        return reps\n",
    "    \n",
    "    def forward(self, data):\n",
    "\n",
    "        token = data[:, 0, :].view(-1, self.max_len) \n",
    "        pos1 = data[:, 1, :].view(-1, self.max_len)\n",
    "        pos2 = data[:, 2, :].view(-1, self.max_len) \n",
    "        mask = data[:, 3, :].view(-1, self.max_len)  \n",
    "        e1_mask = data[:, 4, :].view(-1, self.max_len)\n",
    "        e2_mask = data[:, 5, :].view(-1, self.max_len)\n",
    "        \n",
    "        emb = self.encoder_layer(token, pos1, pos2)\n",
    "   \n",
    "        conv = self.conv_layer(emb, mask)\n",
    "        conv = self.tanh(conv)\n",
    "        \n",
    "        pool = self.single_maxpool_layer(conv)\n",
    "        \n",
    "        e1_emb = self.entity_average(emb, e1_mask)\n",
    "        e2_emb = self.entity_average(emb, e2_mask)\n",
    "        \n",
    "        e1_context = self.attention_layer(emb, e1_emb, mask)\n",
    "        e2_context = self.attention_layer(emb, e2_emb, mask)\n",
    "        e1_context = self.tanh(e1_context)\n",
    "        e2_context = self.tanh(e2_context)\n",
    "     \n",
    "        feature = torch.cat(tensors=[pool, e1_context, e2_context], dim=-1)\n",
    "       \n",
    "        logits = self.dense(feature)\n",
    "\n",
    "        return logits"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "id": "c0mUK1Bcc0Yi"
   },
   "outputs": [],
   "source": [
    "# on utilise le GPU si possible\n",
    "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
    "#device = 'cpu'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "id": "kqMc3GkZc0Yn"
   },
   "outputs": [],
   "source": [
    "WORD_VEC = word_vec\n",
    "CLASS_NUM = class_num\n",
    "MAX_LEN = 500\n",
    "WORD_DIM = 300\n",
    "POS_DIM = 5 # dimension de l'embedding de position\n",
    "POS_DIS = 50\n",
    "DROPOUT = 0.8\n",
    "FILTER_NUM = 100\n",
    "WINDOW = 3\n",
    "\n",
    "model = Attention_CNN(WORD_VEC , CLASS_NUM, MAX_LEN, WORD_DIM, POS_DIM, POS_DIS, DROPOUT, FILTER_NUM, WINDOW).to(device) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "aNUTU5Cdc0Ys"
   },
   "source": [
    "On affiche le nombre de paramètres de ce modèle."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "id": "ZIBVwUlQc0Yt",
    "outputId": "ccbc47f4-de67-43ba-f317-ade0fa58dc23"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Le modèle a 2,895,091 paramètres à entraîner\n"
     ]
    }
   ],
   "source": [
    "def count_parameters(model): \n",
    "    s = sum(p.numel() for p in model.parameters() if p.requires_grad)\n",
    "    print(f'Le modèle a {s:,} paramètres à entraîner')\n",
    "count_parameters(model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "ihhm_lc3c0Yz"
   },
   "source": [
    "### Optimiseur"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "id": "sNpzKMA_c0Y0"
   },
   "outputs": [],
   "source": [
    "import torch.optim as optim\n",
    "criterion = nn.CrossEntropyLoss().to(device)\n",
    "\n",
    "def optimiseur(model, lr=1e-5, eps=1e-6, weight_decay_rate=0.001, second_weight_decay_rate=0.0):\n",
    "    param_optimizer = list(model.named_parameters())\n",
    "    no_decay = ['bias', 'gamma', 'beta']\n",
    "    optimizer_grouped_parameters = [\n",
    "        {'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)],\n",
    "            'weight_decay_rate': weight_decay_rate},\n",
    "        {'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)],\n",
    "            'weight_decay_rate': second_weight_decay_rate}]\n",
    "    return optim.Adam(\n",
    "        optimizer_grouped_parameters,\n",
    "        lr=lr,\n",
    "        eps=eps\n",
    "    )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "JK3qsu9Wc0Y5"
   },
   "source": [
    "### Métriques"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "id": "u8m-ECS4c0Y6"
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import f1_score\n",
    "import math\n",
    "\n",
    "\n",
    "\n",
    "def f1_loss(predict_label, true_label, class_num = class_num):\n",
    "    res = dict()\n",
    "    assert true_label.shape[0] == predict_label.shape[0]\n",
    "    confusion_matrix = np.zeros(shape=[class_num, class_num], dtype=np.float32)\n",
    "    xDIRx = np.zeros(shape=[class_num], dtype=np.float32)\n",
    "    for i in range(true_label.shape[0]):\n",
    "        true_idx = math.ceil(true_label[i]/2)\n",
    "        predict_idx = math.ceil(predict_label[i]/2)\n",
    "        if true_label[i] == predict_label[i]:\n",
    "            confusion_matrix[predict_idx][true_idx] += 1\n",
    "        else:\n",
    "            if true_idx == predict_idx:\n",
    "                xDIRx[predict_idx] += 1\n",
    "            else:\n",
    "                confusion_matrix[predict_idx][true_idx] += 1\n",
    "\n",
    "    col_sum = np.sum(confusion_matrix, axis=0).reshape(-1)\n",
    "    row_sum = np.sum(confusion_matrix, axis=1).reshape(-1)\n",
    "    f1 = np.zeros(shape=[class_num], dtype=np.float32)\n",
    "    recall = np.zeros(shape=[class_num], dtype=np.float32)\n",
    "    precision = np.zeros(shape=[class_num], dtype=np.float32)\n",
    "    \n",
    "    for i in range(0, class_num):  \n",
    "        try:\n",
    "            p = float(confusion_matrix[i][i]) / float(col_sum[i] + xDIRx[i])\n",
    "            precision[i] = p\n",
    "            r = float(confusion_matrix[i][i]) / float(row_sum[i] + xDIRx[i])\n",
    "            recall[i] = r\n",
    "            f1[i] = (2 * p * r / (p + r))\n",
    "        except:\n",
    "            pass\n",
    "    actual_class = 0\n",
    "    total_f1 = 0.0\n",
    "    total_p = 0.0\n",
    "    total_r = 0.0\n",
    "    for i in range(1, class_num):\n",
    "        if f1[i] > 0.0:  \n",
    "            actual_class += 1\n",
    "            total_f1 += f1[i]\n",
    "            total_p += precision[i]\n",
    "            total_r += recall[i]\n",
    "    try:\n",
    "        macro_f1 = total_f1 / actual_class\n",
    "        macro_p = total_p / actual_class\n",
    "        macro_r = total_r / actual_class\n",
    "    except:\n",
    "        macro_f1 = 0.0\n",
    "        macro_p = 0.0\n",
    "        macro_r = 0.0\n",
    "        \n",
    "    res['f1_avg'] = macro_f1\n",
    "    res['f1'] = f1\n",
    "    res['recall_avg'] = macro_r\n",
    "    res['recall'] = recall\n",
    "    res['precision_avg'] = macro_p\n",
    "    res['precision'] = precision\n",
    "    \n",
    "    return res "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "XM82y7Sdc0ZA"
   },
   "source": [
    "### Entraînement"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "id": "crZXawzyc0ZA"
   },
   "outputs": [],
   "source": [
    "def train_model(model, iterator, optimizer, criterion):\n",
    "    predict_label = []\n",
    "    true_label = []\n",
    "    total_loss = 0.0\n",
    "    \n",
    "    for _, (data, label) in enumerate(iterator):\n",
    "        \n",
    "        model.train()\n",
    "        data = data.to(device)\n",
    "        label = label.to(device)\n",
    "        optimizer.zero_grad()\n",
    "        logits = model(data)\n",
    "        loss = criterion(logits, label)\n",
    "    \n",
    "        \n",
    "        loss.backward()\n",
    "        \n",
    "        nn.utils.clip_grad_value_(model.parameters(), clip_value=5)\n",
    "        optimizer.step()\n",
    "        total_loss += loss.item() * logits.shape[0]\n",
    "\n",
    "        _, pred = torch.max(logits, dim=1) \n",
    "        pred = pred.cpu().detach().numpy().reshape((-1, 1))\n",
    "        label = label.cpu().detach().numpy().reshape((-1, 1))\n",
    "        predict_label.append(pred)\n",
    "        true_label.append(label)\n",
    "    \n",
    "        \n",
    "    predict_label = np.concatenate(predict_label, axis=0).reshape(-1).astype(np.int64)\n",
    "    true_label = np.concatenate(true_label, axis=0).reshape(-1).astype(np.int64)\n",
    "    eval_loss = total_loss / predict_label.shape[0]\n",
    "\n",
    "    results = f1_loss(predict_label, true_label)\n",
    "    \n",
    "    return results, eval_loss\n",
    "\n",
    "\n",
    "def evaluate(model, iterator, criterion):\n",
    "    \n",
    "    \n",
    "    predict_label = []\n",
    "    true_label = []\n",
    "    total_loss = 0.0\n",
    "    acc = []\n",
    "    with torch.no_grad():\n",
    "        model.eval()\n",
    "        for _, (data, label) in enumerate(iterator):\n",
    "\n",
    "            data = data.to(device)\n",
    "            label = label.to(device)\n",
    "            logits = model(data)\n",
    "            loss = criterion(logits, label)\n",
    "            \n",
    "            total_loss += loss.item() * logits.shape[0]\n",
    "\n",
    "            _, pred = torch.max(logits, dim=1)  \n",
    "            pred = pred.cpu().detach().numpy().reshape((-1, 1))\n",
    "            label = label.cpu().detach().numpy().reshape((-1, 1))\n",
    "            predict_label.append(pred)\n",
    "            true_label.append(label)\n",
    "    predict_label = np.concatenate(predict_label, axis=0).reshape(-1).astype(np.int64)\n",
    "    true_label = np.concatenate(true_label, axis=0).reshape(-1).astype(np.int64)\n",
    "    eval_loss = total_loss / predict_label.shape[0]\n",
    "\n",
    "    results = f1_loss(predict_label, true_label)\n",
    "    \n",
    "    return results, eval_loss, predict_label, true_label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "id": "8LkmbRYCc0ZE"
   },
   "outputs": [],
   "source": [
    "import time\n",
    "def epoch_time(start_time, end_time):\n",
    "    elapsed_time = end_time - start_time\n",
    "    elapsed_mins = int(elapsed_time / 60)\n",
    "    elapsed_secs = int(elapsed_time - (elapsed_mins * 60))\n",
    "    return elapsed_mins, elapsed_secs\n",
    "\n",
    "def entrainement(model, optimizer, batch_size = 10, N_EPOCHS = 100):\n",
    "    \n",
    "    res = dict()\n",
    "    t_loss = []\n",
    "    t_f1 = []\n",
    "    t_p = []\n",
    "    t_r = []\n",
    "    \n",
    "    v_loss = []\n",
    "    v_f1 = []\n",
    "    v_p = []\n",
    "    v_r = []\n",
    "    best_valid_loss = float('inf')\n",
    "\n",
    "    for epoch in range(N_EPOCHS):\n",
    "\n",
    "        start_time = time.time()\n",
    "\n",
    "        results_train, train_loss = train_model(model, train_loader, optimizer, criterion)\n",
    "        t_loss.append(train_loss)\n",
    "        t_f1.append(results_train['f1_avg']) \n",
    "        t_p.append(results_train['precision_avg']) \n",
    "        t_r.append(results_train['recall_avg']) \n",
    "        \n",
    "        results_valid, valid_loss,_,_ = evaluate(model, valid_loader, criterion)\n",
    "        v_loss.append(valid_loss)\n",
    "        v_f1.append(results_valid['f1_avg'])\n",
    "        v_p.append(results_valid['precision_avg']) \n",
    "        v_r.append(results_valid['recall_avg'])\n",
    "        \n",
    "        end_time = time.time()\n",
    "\n",
    "        epoch_mins, epoch_secs = epoch_time(start_time, end_time)\n",
    "\n",
    "        if valid_loss < best_valid_loss:\n",
    "            best_valid_loss = valid_loss\n",
    "            torch.save(model.state_dict(), 'tut3-model_re.pt')\n",
    "\n",
    "        f1_t = results_train['f1_avg']\n",
    "        f1_v = results_valid['f1_avg']\n",
    "        if epoch%10 == 0: \n",
    "            print(f'Epoch: {epoch+1:02} | Epoch Time: {epoch_mins}m {epoch_secs}s')\n",
    "            print(f'\\tTrain Loss: {train_loss:.3f} | Train F1 score: {f1_t*100:.2f}%')\n",
    "            print(f'\\t Val. Loss: {valid_loss:.3f} |  Val. F1 score: {f1_v*100:.2f}%')\n",
    "            \n",
    "    res['f1_avg_train'] = t_f1 \n",
    "    res['f1_avg_valid'] = v_f1 \n",
    "    res['p_avg_train'] = t_p\n",
    "    res['p_avg_valid'] = v_p \n",
    "    res['r_avg_train'] = t_r\n",
    "    res['r_avg_valid'] = v_r\n",
    "    \n",
    "    res['loss_train'] = t_loss\n",
    "    res['loss_valid'] = v_loss\n",
    "  \n",
    "    res['f1_train'] = results_train['f1']\n",
    "    res['p_train'] = results_train['precision']\n",
    "    res['r_train'] = results_train['recall']\n",
    "    \n",
    "    res['f1_valid'] = results_valid['f1']\n",
    "    res['p_valid'] = results_valid['precision']\n",
    "    res['r_valid'] = results_valid['recall']\n",
    "    \n",
    "    return res"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 272
    },
    "id": "DiFYwK2ic0ZR",
    "outputId": "4aae2a6f-cee4-47db-ef77-aa096dc77804"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 01 | Epoch Time: 0m 12s\n",
      "\tTrain Loss: 3.115 | Train F1 score: 4.65%\n",
      "\t Val. Loss: 3.009 |  Val. F1 score: 6.69%\n",
      "Epoch: 11 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 2.290 | Train F1 score: 25.51%\n",
      "\t Val. Loss: 2.278 |  Val. F1 score: 26.11%\n",
      "Epoch: 21 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 1.935 | Train F1 score: 33.60%\n",
      "\t Val. Loss: 1.946 |  Val. F1 score: 39.85%\n",
      "Epoch: 31 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 1.685 | Train F1 score: 43.11%\n",
      "\t Val. Loss: 1.726 |  Val. F1 score: 40.69%\n",
      "Epoch: 41 | Epoch Time: 0m 10s\n",
      "\tTrain Loss: 1.485 | Train F1 score: 53.16%\n",
      "\t Val. Loss: 1.555 |  Val. F1 score: 49.65%\n",
      "Epoch: 51 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 1.328 | Train F1 score: 59.29%\n",
      "\t Val. Loss: 1.427 |  Val. F1 score: 54.13%\n",
      "Epoch: 61 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 1.195 | Train F1 score: 64.27%\n",
      "\t Val. Loss: 1.327 |  Val. F1 score: 57.88%\n",
      "Epoch: 71 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 1.082 | Train F1 score: 68.79%\n",
      "\t Val. Loss: 1.248 |  Val. F1 score: 60.50%\n",
      "Epoch: 81 | Epoch Time: 0m 10s\n",
      "\tTrain Loss: 0.983 | Train F1 score: 72.20%\n",
      "\t Val. Loss: 1.189 |  Val. F1 score: 62.68%\n",
      "Epoch: 91 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 0.897 | Train F1 score: 75.19%\n",
      "\t Val. Loss: 1.144 |  Val. F1 score: 64.53%\n",
      "Epoch: 101 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 0.819 | Train F1 score: 78.01%\n",
      "\t Val. Loss: 1.108 |  Val. F1 score: 65.66%\n",
      "Epoch: 111 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 0.747 | Train F1 score: 80.53%\n",
      "\t Val. Loss: 1.080 |  Val. F1 score: 66.67%\n",
      "Epoch: 121 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 0.681 | Train F1 score: 82.85%\n",
      "\t Val. Loss: 1.057 |  Val. F1 score: 67.58%\n",
      "Epoch: 131 | Epoch Time: 0m 10s\n",
      "\tTrain Loss: 0.620 | Train F1 score: 84.69%\n",
      "\t Val. Loss: 1.039 |  Val. F1 score: 68.35%\n",
      "Epoch: 141 | Epoch Time: 0m 10s\n",
      "\tTrain Loss: 0.562 | Train F1 score: 86.59%\n",
      "\t Val. Loss: 1.026 |  Val. F1 score: 68.95%\n",
      "Epoch: 151 | Epoch Time: 0m 10s\n",
      "\tTrain Loss: 0.507 | Train F1 score: 88.34%\n",
      "\t Val. Loss: 1.015 |  Val. F1 score: 69.05%\n",
      "Epoch: 161 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 0.456 | Train F1 score: 89.88%\n",
      "\t Val. Loss: 1.008 |  Val. F1 score: 68.97%\n",
      "Epoch: 171 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 0.408 | Train F1 score: 91.33%\n",
      "\t Val. Loss: 1.002 |  Val. F1 score: 69.00%\n",
      "Epoch: 181 | Epoch Time: 0m 9s\n",
      "\tTrain Loss: 0.364 | Train F1 score: 92.88%\n",
      "\t Val. Loss: 1.000 |  Val. F1 score: 69.23%\n",
      "Epoch: 191 | Epoch Time: 0m 10s\n",
      "\tTrain Loss: 0.322 | Train F1 score: 94.19%\n",
      "\t Val. Loss: 0.999 |  Val. F1 score: 69.28%\n"
     ]
    }
   ],
   "source": [
    "N_EPOCHS = 200\n",
    "\n",
    "res = entrainement(model, optimiseur(model), N_EPOCHS = N_EPOCHS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 725
    },
    "id": "XPKp0yBoc0ZV",
    "outputId": "c28a2d45-ec29-4818-e516-c13397cd96ab"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 4 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt \n",
    "import numpy as np \n",
    "sns.set()\n",
    "x = np.linspace(0, N_EPOCHS,N_EPOCHS)\n",
    "f = plt.figure(figsize=(10, 10))\n",
    "gs = f.add_gridspec(2, 2)\n",
    "\n",
    "t_l = res['loss_train']\n",
    "v_l = res['loss_valid']\n",
    "\n",
    "t_f1 = res['f1_avg_train'] \n",
    "v_f1 = res['f1_avg_valid'] \n",
    "\n",
    "t_p = res['p_avg_train'] \n",
    "v_p = res['p_avg_valid']\n",
    "\n",
    "t_r = res['p_avg_train'] \n",
    "v_r = res['p_avg_valid']\n",
    "\n",
    "with sns.axes_style(\"darkgrid\"):\n",
    "    ax = f.add_subplot(gs[0,0])\n",
    "    plt.plot(x,t_l, label='Train Loss')\n",
    "    plt.plot(x,v_l, label='Valid Loss')\n",
    "    plt.title(\"Loss\")\n",
    "    plt.legend()\n",
    "        \n",
    "    ax = f.add_subplot(gs[0,1])\n",
    "    plt.plot(x, t_f1, label='Train F1 score')\n",
    "    plt.plot(x, v_f1, label='Valid F1 score')\n",
    "    plt.title(\"F1 score\")\n",
    "    plt.legend()\n",
    "    \n",
    "    ax = f.add_subplot(gs[1,0])\n",
    "    plt.plot(x, t_p, label='Train Precision')\n",
    "    plt.plot(x, v_p, label='Valid Precision')\n",
    "    plt.title(\"Precision\")\n",
    "    plt.legend()\n",
    "    \n",
    "    ax = f.add_subplot(gs[1,1])\n",
    "    plt.plot(x, t_r, label='Train Recall')\n",
    "    plt.plot(x, v_r, label='Valid Recall')\n",
    "    plt.title(\"Recall\")\n",
    "    plt.legend()\n",
    "\n",
    "       \n",
    "f.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "N8aDc3Rzc0Zl"
   },
   "source": [
    "## Test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "id": "wVnBu3Pcc0Zl",
    "outputId": "8225f2a0-bd92-4f57-8460-5e1edea2e672"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Loss: 0.998 |  Test F1 score: 69.27%\n"
     ]
    }
   ],
   "source": [
    "model.load_state_dict(torch.load('tut3-model_re.pt'))\n",
    "res_test, eval_loss, preds, labels = evaluate(model, test_loader, criterion)\n",
    "f1 = res_test['f1_avg']\n",
    "print(f'Test Loss: {eval_loss:.3f} |  Test F1 score: {f1*100:.2f}%')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 221
    },
    "id": "lNA3RgJIc0Zq",
    "outputId": "9c08300f-129f-4032-9f86-237c8e061942"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                           precision    recall  f1-score   support\n",
      "\n",
      "      Cause-Effect(e1,e2)       0.85      0.84      0.85       134\n",
      "      Cause-Effect(e2,e1)       0.85      0.86      0.85       194\n",
      "   Component-Whole(e1,e2)       0.72      0.72      0.72       162\n",
      "   Component-Whole(e2,e1)       0.60      0.51      0.55       150\n",
      " Content-Container(e1,e2)       0.76      0.84      0.80       153\n",
      " Content-Container(e2,e1)       0.68      0.67      0.68        39\n",
      "Entity-Destination(e1,e2)       0.81      0.86      0.83       291\n",
      "Entity-Destination(e2,e1)       0.00      0.00      0.00         1\n",
      "     Entity-Origin(e1,e2)       0.77      0.82      0.80       211\n",
      "     Entity-Origin(e2,e1)       0.86      0.77      0.81        47\n",
      " Instrument-Agency(e1,e2)       0.50      0.41      0.45        22\n",
      " Instrument-Agency(e2,e1)       0.66      0.55      0.60       134\n",
      " Member-Collection(e1,e2)       0.44      0.25      0.32        32\n",
      " Member-Collection(e2,e1)       0.77      0.85      0.81       201\n",
      "     Message-Topic(e1,e2)       0.73      0.73      0.73       210\n",
      "     Message-Topic(e2,e1)       0.72      0.57      0.64        51\n",
      "                    Other       0.38      0.43      0.40       454\n",
      "  Product-Producer(e1,e2)       0.68      0.62      0.65       108\n",
      "  Product-Producer(e2,e1)       0.62      0.47      0.54       123\n",
      "\n",
      "                 accuracy                           0.68      2717\n",
      "                macro avg       0.65      0.62      0.63      2717\n",
      "             weighted avg       0.68      0.68      0.68      2717\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.7/site-packages/sklearn/metrics/_classification.py:1221: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
      "  _warn_prf(average, modifier, msg_start, len(result))\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import classification_report, confusion_matrix\n",
    "predict =  preds\n",
    "true =  labels\n",
    "\n",
    "positive_labels = [i for i in range(len(rel2id)) if i != 0]\n",
    "positive_names = [list(rel2id.keys())[list(rel2id.values()).index(i)] for i in range(len(rel2id)) if i != 0]\n",
    "print(classification_report(\n",
    "                y_true=true,\n",
    "                y_pred=predict,\n",
    "                labels=positive_labels,\n",
    "                target_names=positive_names\n",
    "            ))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 204
    },
    "id": "PLs55cCOeJHH",
    "outputId": "64e5127a-4105-4c85-a869-2b1b2d5c36f3"
   },
   "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>Cause-Effect(e1,e2)</th>\n",
       "      <th>Cause-Effect(e2,e1)</th>\n",
       "      <th>Component-Whole(e1,e2)</th>\n",
       "      <th>Component-Whole(e2,e1)</th>\n",
       "      <th>Content-Container(e1,e2)</th>\n",
       "      <th>Content-Container(e2,e1)</th>\n",
       "      <th>Entity-Destination(e1,e2)</th>\n",
       "      <th>Entity-Destination(e2,e1)</th>\n",
       "      <th>Entity-Origin(e1,e2)</th>\n",
       "      <th>Entity-Origin(e2,e1)</th>\n",
       "      <th>Instrument-Agency(e1,e2)</th>\n",
       "      <th>Instrument-Agency(e2,e1)</th>\n",
       "      <th>Member-Collection(e1,e2)</th>\n",
       "      <th>Member-Collection(e2,e1)</th>\n",
       "      <th>Message-Topic(e1,e2)</th>\n",
       "      <th>Message-Topic(e2,e1)</th>\n",
       "      <th>Other</th>\n",
       "      <th>Product-Producer(e1,e2)</th>\n",
       "      <th>Product-Producer(e2,e1)</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Cause-Effect(e1,e2)</th>\n",
       "      <td>112</td>\n",
       "      <td>11</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>8</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Cause-Effect(e2,e1)</th>\n",
       "      <td>3</td>\n",
       "      <td>167</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>10</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Component-Whole(e1,e2)</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>116</td>\n",
       "      <td>9</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>7</td>\n",
       "      <td>3</td>\n",
       "      <td>1</td>\n",
       "      <td>19</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Component-Whole(e2,e1)</th>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>10</td>\n",
       "      <td>76</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>12</td>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>33</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Content-Container(e1,e2)</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>128</td>\n",
       "      <td>2</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>14</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Content-Container(e2,e1)</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "      <td>1</td>\n",
       "      <td>26</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>7</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Entity-Destination(e1,e2)</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>15</td>\n",
       "      <td>1</td>\n",
       "      <td>249</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>22</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Entity-Destination(e2,e1)</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Entity-Origin(e1,e2)</th>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>174</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>19</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Entity-Origin(e2,e1)</th>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>36</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Instrument-Agency(e1,e2)</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>9</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>8</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Instrument-Agency(e2,e1)</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>8</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>74</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>33</td>\n",
       "      <td>5</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Member-Collection(e1,e2)</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>8</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Member-Collection(e2,e1)</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>7</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>171</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Message-Topic(e1,e2)</th>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>154</td>\n",
       "      <td>2</td>\n",
       "      <td>38</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Message-Topic(e2,e1)</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>8</td>\n",
       "      <td>29</td>\n",
       "      <td>12</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Other</th>\n",
       "      <td>6</td>\n",
       "      <td>9</td>\n",
       "      <td>18</td>\n",
       "      <td>18</td>\n",
       "      <td>18</td>\n",
       "      <td>3</td>\n",
       "      <td>34</td>\n",
       "      <td>0</td>\n",
       "      <td>26</td>\n",
       "      <td>4</td>\n",
       "      <td>4</td>\n",
       "      <td>15</td>\n",
       "      <td>7</td>\n",
       "      <td>35</td>\n",
       "      <td>29</td>\n",
       "      <td>7</td>\n",
       "      <td>193</td>\n",
       "      <td>18</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Product-Producer(e1,e2)</th>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>21</td>\n",
       "      <td>67</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Product-Producer(e2,e1)</th>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>5</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>6</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>6</td>\n",
       "      <td>0</td>\n",
       "      <td>34</td>\n",
       "      <td>4</td>\n",
       "      <td>58</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                           Cause-Effect(e1,e2)  Cause-Effect(e2,e1)  \\\n",
       "Cause-Effect(e1,e2)                        112                   11   \n",
       "Cause-Effect(e2,e1)                          3                  167   \n",
       "Component-Whole(e1,e2)                       1                    0   \n",
       "Component-Whole(e2,e1)                       2                    0   \n",
       "Content-Container(e1,e2)                     0                    0   \n",
       "Content-Container(e2,e1)                     0                    0   \n",
       "Entity-Destination(e1,e2)                    0                    0   \n",
       "Entity-Destination(e2,e1)                    0                    0   \n",
       "Entity-Origin(e1,e2)                         0                    5   \n",
       "Entity-Origin(e2,e1)                         3                    0   \n",
       "Instrument-Agency(e1,e2)                     0                    0   \n",
       "Instrument-Agency(e2,e1)                     0                    0   \n",
       "Member-Collection(e1,e2)                     0                    0   \n",
       "Member-Collection(e2,e1)                     0                    0   \n",
       "Message-Topic(e1,e2)                         2                    1   \n",
       "Message-Topic(e2,e1)                         0                    0   \n",
       "Other                                        6                    9   \n",
       "Product-Producer(e1,e2)                      0                    4   \n",
       "Product-Producer(e2,e1)                      2                    0   \n",
       "\n",
       "                           Component-Whole(e1,e2)  Component-Whole(e2,e1)  \\\n",
       "Cause-Effect(e1,e2)                             0                       2   \n",
       "Cause-Effect(e2,e1)                             0                       0   \n",
       "Component-Whole(e1,e2)                        116                       9   \n",
       "Component-Whole(e2,e1)                         10                      76   \n",
       "Content-Container(e1,e2)                        1                       2   \n",
       "Content-Container(e2,e1)                        1                       3   \n",
       "Entity-Destination(e1,e2)                       1                       0   \n",
       "Entity-Destination(e2,e1)                       0                       0   \n",
       "Entity-Origin(e1,e2)                            2                       0   \n",
       "Entity-Origin(e2,e1)                            0                       0   \n",
       "Instrument-Agency(e1,e2)                        0                       1   \n",
       "Instrument-Agency(e2,e1)                        1                       8   \n",
       "Member-Collection(e1,e2)                        0                       1   \n",
       "Member-Collection(e2,e1)                        7                       2   \n",
       "Message-Topic(e1,e2)                            1                       0   \n",
       "Message-Topic(e2,e1)                            0                       0   \n",
       "Other                                          18                      18   \n",
       "Product-Producer(e1,e2)                         0                       0   \n",
       "Product-Producer(e2,e1)                         2                       5   \n",
       "\n",
       "                           Content-Container(e1,e2)  Content-Container(e2,e1)  \\\n",
       "Cause-Effect(e1,e2)                               0                         0   \n",
       "Cause-Effect(e2,e1)                               0                         0   \n",
       "Component-Whole(e1,e2)                            3                         0   \n",
       "Component-Whole(e2,e1)                            1                         2   \n",
       "Content-Container(e1,e2)                        128                         2   \n",
       "Content-Container(e2,e1)                          1                        26   \n",
       "Entity-Destination(e1,e2)                        15                         1   \n",
       "Entity-Destination(e2,e1)                         0                         0   \n",
       "Entity-Origin(e1,e2)                              1                         0   \n",
       "Entity-Origin(e2,e1)                              0                         0   \n",
       "Instrument-Agency(e1,e2)                          0                         0   \n",
       "Instrument-Agency(e2,e1)                          0                         1   \n",
       "Member-Collection(e1,e2)                          0                         0   \n",
       "Member-Collection(e2,e1)                          0                         1   \n",
       "Message-Topic(e1,e2)                              0                         0   \n",
       "Message-Topic(e2,e1)                              1                         0   \n",
       "Other                                            18                         3   \n",
       "Product-Producer(e1,e2)                           0                         0   \n",
       "Product-Producer(e2,e1)                           1                         2   \n",
       "\n",
       "                           Entity-Destination(e1,e2)  \\\n",
       "Cause-Effect(e1,e2)                                0   \n",
       "Cause-Effect(e2,e1)                                0   \n",
       "Component-Whole(e1,e2)                             0   \n",
       "Component-Whole(e2,e1)                             2   \n",
       "Content-Container(e1,e2)                           5   \n",
       "Content-Container(e2,e1)                           0   \n",
       "Entity-Destination(e1,e2)                        249   \n",
       "Entity-Destination(e2,e1)                          1   \n",
       "Entity-Origin(e1,e2)                               5   \n",
       "Entity-Origin(e2,e1)                               1   \n",
       "Instrument-Agency(e1,e2)                           0   \n",
       "Instrument-Agency(e2,e1)                           2   \n",
       "Member-Collection(e1,e2)                           2   \n",
       "Member-Collection(e2,e1)                           0   \n",
       "Message-Topic(e1,e2)                               5   \n",
       "Message-Topic(e2,e1)                               0   \n",
       "Other                                             34   \n",
       "Product-Producer(e1,e2)                            0   \n",
       "Product-Producer(e2,e1)                            2   \n",
       "\n",
       "                           Entity-Destination(e2,e1)  Entity-Origin(e1,e2)  \\\n",
       "Cause-Effect(e1,e2)                                0                     0   \n",
       "Cause-Effect(e2,e1)                                0                    10   \n",
       "Component-Whole(e1,e2)                             0                     0   \n",
       "Component-Whole(e2,e1)                             0                     1   \n",
       "Content-Container(e1,e2)                           0                     1   \n",
       "Content-Container(e2,e1)                           0                     0   \n",
       "Entity-Destination(e1,e2)                          0                     1   \n",
       "Entity-Destination(e2,e1)                          0                     0   \n",
       "Entity-Origin(e1,e2)                               0                   174   \n",
       "Entity-Origin(e2,e1)                               0                     0   \n",
       "Instrument-Agency(e1,e2)                           0                     0   \n",
       "Instrument-Agency(e2,e1)                           0                     2   \n",
       "Member-Collection(e1,e2)                           0                     1   \n",
       "Member-Collection(e2,e1)                           0                     1   \n",
       "Message-Topic(e1,e2)                               0                     2   \n",
       "Message-Topic(e2,e1)                               0                     1   \n",
       "Other                                              0                    26   \n",
       "Product-Producer(e1,e2)                            0                     5   \n",
       "Product-Producer(e2,e1)                            0                     0   \n",
       "\n",
       "                           Entity-Origin(e2,e1)  Instrument-Agency(e1,e2)  \\\n",
       "Cause-Effect(e1,e2)                           1                         0   \n",
       "Cause-Effect(e2,e1)                           0                         0   \n",
       "Component-Whole(e1,e2)                        0                         1   \n",
       "Component-Whole(e2,e1)                        0                         1   \n",
       "Content-Container(e1,e2)                      0                         0   \n",
       "Content-Container(e2,e1)                      0                         0   \n",
       "Entity-Destination(e1,e2)                     1                         0   \n",
       "Entity-Destination(e2,e1)                     0                         0   \n",
       "Entity-Origin(e1,e2)                          0                         0   \n",
       "Entity-Origin(e2,e1)                         36                         0   \n",
       "Instrument-Agency(e1,e2)                      0                         9   \n",
       "Instrument-Agency(e2,e1)                      0                         1   \n",
       "Member-Collection(e1,e2)                      0                         0   \n",
       "Member-Collection(e2,e1)                      0                         0   \n",
       "Message-Topic(e1,e2)                          0                         1   \n",
       "Message-Topic(e2,e1)                          0                         0   \n",
       "Other                                         4                         4   \n",
       "Product-Producer(e1,e2)                       0                         1   \n",
       "Product-Producer(e2,e1)                       0                         0   \n",
       "\n",
       "                           Instrument-Agency(e2,e1)  Member-Collection(e1,e2)  \\\n",
       "Cause-Effect(e1,e2)                               0                         0   \n",
       "Cause-Effect(e2,e1)                               0                         0   \n",
       "Component-Whole(e1,e2)                            1                         0   \n",
       "Component-Whole(e2,e1)                           12                         0   \n",
       "Content-Container(e1,e2)                          0                         0   \n",
       "Content-Container(e2,e1)                          1                         0   \n",
       "Entity-Destination(e1,e2)                         0                         0   \n",
       "Entity-Destination(e2,e1)                         0                         0   \n",
       "Entity-Origin(e1,e2)                              0                         0   \n",
       "Entity-Origin(e2,e1)                              0                         0   \n",
       "Instrument-Agency(e1,e2)                          2                         0   \n",
       "Instrument-Agency(e2,e1)                         74                         0   \n",
       "Member-Collection(e1,e2)                          0                         8   \n",
       "Member-Collection(e2,e1)                          0                         2   \n",
       "Message-Topic(e1,e2)                              1                         0   \n",
       "Message-Topic(e2,e1)                              0                         0   \n",
       "Other                                            15                         7   \n",
       "Product-Producer(e1,e2)                           0                         1   \n",
       "Product-Producer(e2,e1)                           6                         0   \n",
       "\n",
       "                           Member-Collection(e2,e1)  Message-Topic(e1,e2)  \\\n",
       "Cause-Effect(e1,e2)                               0                     0   \n",
       "Cause-Effect(e2,e1)                               0                     0   \n",
       "Component-Whole(e1,e2)                            7                     3   \n",
       "Component-Whole(e2,e1)                            4                     3   \n",
       "Content-Container(e1,e2)                          0                     0   \n",
       "Content-Container(e2,e1)                          0                     0   \n",
       "Entity-Destination(e1,e2)                         0                     0   \n",
       "Entity-Destination(e2,e1)                         0                     0   \n",
       "Entity-Origin(e1,e2)                              0                     2   \n",
       "Entity-Origin(e2,e1)                              0                     0   \n",
       "Instrument-Agency(e1,e2)                          1                     0   \n",
       "Instrument-Agency(e2,e1)                          0                     3   \n",
       "Member-Collection(e1,e2)                          3                     0   \n",
       "Member-Collection(e2,e1)                        171                     3   \n",
       "Message-Topic(e1,e2)                              1                   154   \n",
       "Message-Topic(e2,e1)                              0                     8   \n",
       "Other                                            35                    29   \n",
       "Product-Producer(e1,e2)                           0                     0   \n",
       "Product-Producer(e2,e1)                           1                     6   \n",
       "\n",
       "                           Message-Topic(e2,e1)  Other  \\\n",
       "Cause-Effect(e1,e2)                           0      8   \n",
       "Cause-Effect(e2,e1)                           0     12   \n",
       "Component-Whole(e1,e2)                        1     19   \n",
       "Component-Whole(e2,e1)                        0     33   \n",
       "Content-Container(e1,e2)                      0     14   \n",
       "Content-Container(e2,e1)                      0      7   \n",
       "Entity-Destination(e1,e2)                     0     22   \n",
       "Entity-Destination(e2,e1)                     0      0   \n",
       "Entity-Origin(e1,e2)                          0     19   \n",
       "Entity-Origin(e2,e1)                          0      5   \n",
       "Instrument-Agency(e1,e2)                      0      8   \n",
       "Instrument-Agency(e2,e1)                      0     33   \n",
       "Member-Collection(e1,e2)                      1     16   \n",
       "Member-Collection(e2,e1)                      0     12   \n",
       "Message-Topic(e1,e2)                          2     38   \n",
       "Message-Topic(e2,e1)                         29     12   \n",
       "Other                                         7    193   \n",
       "Product-Producer(e1,e2)                       0     21   \n",
       "Product-Producer(e2,e1)                       0     34   \n",
       "\n",
       "                           Product-Producer(e1,e2)  Product-Producer(e2,e1)  \n",
       "Cause-Effect(e1,e2)                              0                        0  \n",
       "Cause-Effect(e2,e1)                              2                        0  \n",
       "Component-Whole(e1,e2)                           0                        1  \n",
       "Component-Whole(e2,e1)                           0                        3  \n",
       "Content-Container(e1,e2)                         0                        0  \n",
       "Content-Container(e2,e1)                         0                        0  \n",
       "Entity-Destination(e1,e2)                        0                        1  \n",
       "Entity-Destination(e2,e1)                        0                        0  \n",
       "Entity-Origin(e1,e2)                             2                        1  \n",
       "Entity-Origin(e2,e1)                             0                        2  \n",
       "Instrument-Agency(e1,e2)                         1                        0  \n",
       "Instrument-Agency(e2,e1)                         5                        4  \n",
       "Member-Collection(e1,e2)                         0                        0  \n",
       "Member-Collection(e2,e1)                         0                        2  \n",
       "Message-Topic(e1,e2)                             0                        2  \n",
       "Message-Topic(e2,e1)                             0                        0  \n",
       "Other                                           18                       10  \n",
       "Product-Producer(e1,e2)                         67                        9  \n",
       "Product-Producer(e2,e1)                          4                       58  "
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "\n",
    "confusion = confusion_matrix(true, predict)\n",
    "l = positive_names\n",
    "\n",
    "confusion_df =pd.DataFrame(confusion)\n",
    "\n",
    "confusion_df.columns=l\n",
    "s = pd.Series(l)\n",
    "confusion_df = confusion_df.set_index([s])\n",
    "\n",
    "\n",
    "confusion_df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "A9oSCbOBc0Zw"
   },
   "source": [
    "## Références : \n",
    "\n",
    " - https://www.aclweb.org/anthology/C16-1238.pdf\n",
    " - https://github.com/onehaitao/Attention-CNN-relation-extraction"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "name": "3-CNN_Att-RE.ipynb",
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}