(Experiment 7) FR_Test_Character_Embedding .ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import csv\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "import pickle"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "DS = pd.read_csv(\"FR_Object_tagged_Subset_150SentenceLength.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "DS['tag'].fillna('text', inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text_ID</th>\n",
       "      <th>word</th>\n",
       "      <th>tag</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>51616</td>\n",
       "      <td>toonid</td>\n",
       "      <td>text</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>51616</td>\n",
       "      <td>rütmilised,</td>\n",
       "      <td>text</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>51616</td>\n",
       "      <td>fr</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>51616</td>\n",
       "      <td>72,</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>101498</td>\n",
       "      <td>rütm</td>\n",
       "      <td>text</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>101498</td>\n",
       "      <td>fr</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>101498</td>\n",
       "      <td>83</td>\n",
       "      <td>object</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>101498</td>\n",
       "      <td>ii,</td>\n",
       "      <td>text</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    text_ID         word     tag\n",
       "7     51616       toonid    text\n",
       "8     51616  rütmilised,    text\n",
       "9     51616           fr  object\n",
       "10    51616          72,  object\n",
       "15   101498         rütm    text\n",
       "16   101498           fr  object\n",
       "17   101498           83  object\n",
       "20   101498          ii,    text"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "DS.loc[np.r_[7:11,15:18,20:21],:] "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "words = list(set(DS[\"word\"].values))\n",
    "n_words = len(words)\n",
    "tags = list(set(DS[\"tag\"].values))\n",
    "n_tags = len(tags)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "class SentenceGetter(object):\n",
    "    def __init__(self, data):\n",
    "        self.n_sent = 1\n",
    "        self.data = data\n",
    "        self.empty = False\n",
    "        agg_func = lambda s: [(w, t) for w, t in zip(s[\"word\"].values.tolist(),\n",
    "                                                           s[\"tag\"].values.tolist())]\n",
    "        self.grouped = self.data.groupby(\"text_ID\").apply(agg_func)\n",
    "        self.sentences = [s for s in self.grouped]\n",
    "getter = SentenceGetter(DS)\n",
    "sentences = getter.sentences\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_len = 150\n",
    "max_len_char = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "tag2idx = {t: i + 1 for i, t in enumerate(tags)}\n",
    "tag2idx[\"PAD\"] = 0\n",
    "idx2tag = {i: w for w, i in tag2idx.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'object': 1, 'text': 2, 'PAD': 0}"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tag2idx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100\n"
     ]
    }
   ],
   "source": [
    "chars = set([w_i for w in words for w_i in w])\n",
    "n_chars = len(chars)\n",
    "print(n_chars)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "char2idx = {c: i + 2 for i, c in enumerate(chars)}\n",
    "char2idx[\"UNK\"] = 1 #For predictions\n",
    "char2idx[\"PAD\"] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "idx2char = {i: w for w, i in char2idx.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_char = []\n",
    "for sentence in sentences:\n",
    "    sent_seq = []\n",
    "    for i in range(max_len):\n",
    "        word_seq = []\n",
    "        for j in range(max_len_char):\n",
    "            try:\n",
    "                word_seq.append(char2idx.get(sentence[i][0][j]))\n",
    "            except:\n",
    "                word_seq.append(char2idx.get(\"PAD\"))\n",
    "        sent_seq.append(word_seq)\n",
    "    X_char.append(np.array(sent_seq))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "y = [[tag2idx[w[1]] for w  in s] for s in sentences]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "4072"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(sentences)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "y = pad_sequences(maxlen=max_len, sequences=y, value=tag2idx[\"PAD\"], padding='post', truncating='post')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0])"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'object': 1, 'text': 2, 'PAD': 0}"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tag2idx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "X_char_tr, X_char_te, y_tr,  y_te= train_test_split(X_char, y, test_size=0.8, random_state=2018)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "max_len_char"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Kasutaja\\Anaconda3\\envs\\myEnv\\lib\\site-packages\\tensorflow_core\\python\\framework\\indexed_slices.py:433: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
      "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         (None, 150, 10)           0         \n",
      "_________________________________________________________________\n",
      "time_distributed_1 (TimeDist (None, 150, 10, 20)       2120      \n",
      "_________________________________________________________________\n",
      "time_distributed_2 (TimeDist (None, 150, 20)           3280      \n",
      "_________________________________________________________________\n",
      "spatial_dropout1d_1 (Spatial (None, 150, 20)           0         \n",
      "_________________________________________________________________\n",
      "bidirectional_1 (Bidirection (None, 150, 40)           6560      \n",
      "_________________________________________________________________\n",
      "time_distributed_3 (TimeDist (None, 150, 3)            123       \n",
      "=================================================================\n",
      "Total params: 12,083\n",
      "Trainable params: 12,083\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras.models import load_model\n",
    " \n",
    "# load model\n",
    "model = load_model('CharacterEmbeddingObject-1.00.hdf5')\n",
    "# summarize model.\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model.layers[4:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 732 samples, validate on 82 samples\n",
      "Epoch 1/10\n",
      "732/732 [==============================] - 6s 8ms/step - loss: 0.0313 - acc: 0.9909 - val_loss: 0.0249 - val_acc: 0.9922\n",
      "Epoch 2/10\n",
      "732/732 [==============================] - 5s 7ms/step - loss: 0.0257 - acc: 0.9920 - val_loss: 0.0240 - val_acc: 0.9924\n",
      "Epoch 3/10\n",
      "732/732 [==============================] - 5s 7ms/step - loss: 0.0228 - acc: 0.9922 - val_loss: 0.0198 - val_acc: 0.9926\n",
      "Epoch 4/10\n",
      "732/732 [==============================] - 5s 7ms/step - loss: 0.0190 - acc: 0.9933 - val_loss: 0.0155 - val_acc: 0.9952\n",
      "Epoch 5/10\n",
      "732/732 [==============================] - 5s 7ms/step - loss: 0.0165 - acc: 0.9944 - val_loss: 0.0130 - val_acc: 0.9956\n",
      "Epoch 6/10\n",
      "732/732 [==============================] - 5s 7ms/step - loss: 0.0142 - acc: 0.9952 - val_loss: 0.0110 - val_acc: 0.9968\n",
      "Epoch 7/10\n",
      "732/732 [==============================] - 5s 7ms/step - loss: 0.0119 - acc: 0.9960 - val_loss: 0.0085 - val_acc: 0.9977\n",
      "Epoch 8/10\n",
      "732/732 [==============================] - 5s 7ms/step - loss: 0.0077 - acc: 0.9977 - val_loss: 0.0058 - val_acc: 0.9985\n",
      "Epoch 9/10\n",
      "732/732 [==============================] - 5s 7ms/step - loss: 0.0063 - acc: 0.9982 - val_loss: 0.0060 - val_acc: 0.9986\n",
      "Epoch 10/10\n",
      "732/732 [==============================] - 5s 7ms/step - loss: 0.0053 - acc: 0.9987 - val_loss: 0.0055 - val_acc: 0.9987\n"
     ]
    }
   ],
   "source": [
    "from keras.callbacks import ModelCheckpoint\n",
    "from keras.optimizers import Adam, SGD, RMSprop\n",
    "\n",
    "# # Saving the best model only\n",
    "# filepath=\"CharacterEmbeddingObject-{val_acc:.2f}.hdf5\"\n",
    "# checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='max')\n",
    "# callbacks_list = [checkpoint]\n",
    "\n",
    "# for layer in model.layers[:4]:\n",
    "#     layer.trainable = False\n",
    "# for layer in model.layers[4:]:\n",
    "#     layer.trainable = True\n",
    "\n",
    "# early_stopping = EarlyStopping(monitor='val_acc', patience=5)\n",
    "#https://www.depends-on-the-definition.com/transfer-learning-for-dog-breed-identification/\n",
    "\n",
    "# model.compile(optimizer=Adam(lr=0.0001, momentum=0.9),\n",
    "#               loss='sparse_categorical_crossentropy',\n",
    "#               metrics=[\"acc\"])\n",
    "\n",
    "# Fit the best model\n",
    "history = model.fit(np.array(X_char_tr).reshape((len(X_char_tr), max_len, max_len_char)),\n",
    "                    np.array(y_tr).reshape(len(y_tr), max_len, 1),\n",
    "                    batch_size=32, epochs=10, validation_split=0.1, verbose=1)\n",
    "#,callbacks=callbacks_list"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<keras.engine.input_layer.InputLayer at 0x20b8afca848>,\n",
       " <keras.layers.wrappers.TimeDistributed at 0x20b8afc6588>,\n",
       " <keras.layers.wrappers.TimeDistributed at 0x20b8afc6b88>,\n",
       " <keras.layers.core.SpatialDropout1D at 0x20b8afc6688>,\n",
       " <keras.layers.wrappers.Bidirectional at 0x20b8afca0c8>,\n",
       " <keras.layers.wrappers.TimeDistributed at 0x20b8afca408>]"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "hist = pd.DataFrame(history.history)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x864 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "plt.style.use(\"ggplot\")\n",
    "plt.figure(figsize=(12,12))\n",
    "plt.plot(hist[\"acc\"])\n",
    "plt.plot(hist[\"val_acc\"])\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "def charindx2word(Tword):\n",
    "    word = []\n",
    "    for i in range(len(Tword)):\n",
    "        if(Tword[i] !=0):\n",
    "            word.append(idx2char[Tword[i]])\n",
    "    return \"\".join(word)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = model.predict(\n",
    "                        np.array(X_char_te).reshape((len(X_char_te),\n",
    "                                                     max_len, max_len_char)))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3258"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(np.array(X_char_te).reshape((len(X_char_te),\n",
    "                                                     max_len, max_len_char)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(814, 150, 10)"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.array(X_char_tr).reshape((len(X_char_tr),\n",
    "                                                     max_len, max_len_char)).shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Word           ||True ||Pred\n",
      "==============================\n",
      "EKG            : text  text\n",
      "Holter:        : text  text\n",
      "siinusrütm     : text  text\n",
      "fr             : object object\n",
      "63-127         : object object\n",
      "xmin,          : text  text\n",
      "keskmine       : text  text\n",
      "sagedus        : text  text\n",
      "83             : text  text\n",
      "xmin.          : text  text\n",
      "Rütmihäire     : text  text\n",
      "Kokkuvõtva     : text  text\n",
      "ravi           : text  text\n",
      "ja             : text  text\n",
      "plaanid        : text  text\n",
      "jäävad         : text  text\n",
      "samaks!        : text  text\n"
     ]
    }
   ],
   "source": [
    "i = 1\n",
    "p = np.argmax(y_pred[i], axis=-1)\n",
    "print(\"{:15}||{:5}||{}\".format(\"Word\", \"True\", \"Pred\"))\n",
    "print(30 * \"=\")\n",
    "for w, t, pred in zip(X_char_te[i], y_te[i], p):\n",
    "    if t != 0: #PAD\n",
    "        print(\"{:15}: {:5} {}\".format(charindx2word(w), idx2tag[t], idx2tag[pred]))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "3258"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Wrong Tags\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "49             : text  object\n",
      "mmHg,fr79x     : object text\n",
      "fr78.Kl.ja     : object text\n",
      "90',PQ         : object text\n",
      "mmHg,fr        : object text\n",
      "72             : object text\n",
      "mmHg,fr69x     : object text\n",
      "r              : text  object\n",
      "133            : text  object\n",
      "=              : text  object\n",
      "63             : object text\n",
      "Normofrekv     : object text\n",
      "=              : text  object\n",
      "76             : object text\n",
      "66             : object text\n",
      "120/80;fr      : object text\n",
      "74;            : object text\n",
      "64',St         : object text\n",
      "=              : text  object\n",
      "77             : object text\n",
      "FA.fr          : object text\n",
      "70/min.        : object text\n",
      "EKG:fr67x'     : object text\n",
      "=              : text  object\n",
      "88             : object text\n",
      "x/min,         : text  object\n",
      "nefroloogi     : object text\n",
      "Siinusrütm     : object text\n",
      "76             : object text\n",
      "Siinusrütm     : object text\n",
      "76             : object text\n",
      "Siinusrütm     : object text\n",
      "76             : object text\n",
      "200/100mmH     : object text\n",
      "200/100mmH     : object text\n",
      "Kaal60         : object text\n",
      "Siinusrütm     : object text\n",
      "72.<ANONYM     : object text\n",
      "siinusrütm     : object text\n",
      "71,horizon     : object text\n",
      "100/70;        : object text\n",
      "mmHg,fr        : object text\n",
      "66             : object text\n",
      "fr.92x'.       : object text\n",
      "regulaarne     : object text\n",
      "mmHg,fr        : object text\n",
      "82             : object text\n",
      "70             : object text\n",
      "112            : object text\n",
      "fr111          : object text\n",
      "mmHgfr         : object text\n",
      "x/min,         : text  object\n",
      "Cor-fr         : object text\n",
      "96-100x/mi     : object text\n",
      "fr68           : object PAD\n",
      "regulaarne     : object text\n",
      "SR,fr          : object text\n",
      "51             : object text\n",
      "fr.81          : object text\n",
      "normofrekv     : object text\n",
      "fr70x*         : object text\n",
      "mmHg,fr        : object text\n",
      "65             : object text\n",
      "siinusrütm     : object text\n",
      "77x            : object text\n",
      "mmHg,fr        : object text\n",
      "63             : object text\n",
      "=              : text  object\n",
      "61             : object text\n",
      "130/80         : object text\n",
      "145/82,fr      : object text\n",
      "77             : object text\n",
      "x/min,         : text  object\n",
      "71             : object text\n",
      "mmHg,fr        : object text\n",
      "81             : object text\n",
      "fr.78          : object text\n",
      "=              : text  object\n",
      "79             : object text\n",
      "mmHg,fr        : object text\n",
      "75             : object text\n",
      "fr85x´         : object text\n",
      "mmHg,fr        : object text\n",
      "64x            : object text\n",
      "90',PQ         : object text\n",
      "SR,fr51        : object text\n",
      "=              : text  object\n",
      "57             : object text\n",
      "Siinusrütm     : object text\n",
      "fr65           : object PAD\n",
      "64             : object text\n",
      "frrquent       : object text\n",
      "frrquent       : object text\n",
      "(              : text  object\n",
      "täna           : text  object\n",
      "115/75         : object text\n",
      "<ANONYM        : text  object\n",
      "110/70         : object text\n",
      "Normofrekv     : object text\n",
      "fr102x1,       : object text\n",
      "140780         : object text\n",
      "=              : text  object\n",
      "96             : object text\n",
      "arütmia,fr     : object text\n",
      "55,horizon     : object text\n",
      "arütmia,fr     : object text\n",
      "55,horizon     : object text\n",
      "190.           : object text\n",
      "fr=185         : object text\n",
      "118/80;        : object text\n",
      "üv             : text  object\n",
      "EKG:Siinus     : object text\n",
      "68x1           : object text\n",
      "ebareg.fr.     : object text\n",
      "fr.-102x´      : object text\n",
      "110/70,fr      : object text\n",
      "81.            : object text\n",
      "Siinusrütm     : object text\n",
      "x              : text  object\n",
      "=              : text  object\n",
      "68             : object text\n",
      "fr87,vestl     : object text\n",
      "mmHg,fr        : object text\n",
      "67             : object text\n",
      "N,fr78,kl.     : object text\n",
      "SR,.fr97x/     : object text\n",
      "135/80mmHg     : object text\n",
      "r              : text  object\n",
      "epol           : text  object\n",
      "160/90mmHg     : object text\n",
      "86x            : object text\n",
      "mmHg.fr        : object text\n",
      "62             : object text\n",
      "=              : text  object\n",
      "76             : object text\n",
      "=              : text  object\n",
      "79             : object text\n",
      "Siinusrütm     : object text\n",
      "76x            : object text\n",
      "reg.fr         : object text\n",
      "100            : object text\n",
      "Siinusrütm     : object text\n",
      "65/min.Süg     : object text\n",
      "fr63xmin.,     : object text\n",
      "(N             : text  object\n",
      ">6)            : text  object\n",
      "=              : text  object\n",
      "67             : object text\n",
      "96             : text  object\n",
      "63             : text  object\n",
      "63             : text  object\n",
      "63             : text  object\n",
      "Normofrekv     : object text\n",
      "Siinusrütm     : object text\n",
      "65/min.        : object text\n",
      "fr111          : object text\n",
      "Hg,fr          : object text\n",
      "67             : object text\n",
      "88             : text  object\n",
      "86             : object text\n",
      "rr             : text  object\n",
      "110-72         : text  object\n",
      "normofr.,      : object text\n",
      "bradüarütm     : object text\n",
      "66             : object text\n",
      "96             : text  object\n",
      "siinusrütm     : object text\n",
      "82xmin,        : object text\n",
      "76,            : text  object\n",
      "siinusrütm     : object text\n",
      "68             : object text\n",
      "157/86         : object text\n",
      "fr78,          : object text\n",
      "x/min,         : text  object\n",
      "120/80         : object text\n",
      "(              : text  object\n",
      "täna           : text  object\n",
      "fr-87,         : object text\n",
      "146/80         : object text\n",
      "130/80         : object text\n",
      "EKG.Siinus     : object text\n",
      "87x,           : object text\n",
      "155/80         : object text\n",
      "65             : text  object\n",
      "65             : text  object\n",
      "65             : text  object\n",
      "65             : text  object\n",
      "(N             : text  object\n",
      "<6)_           : text  object\n",
      "=              : text  object\n",
      "65             : object text\n",
      "treemor-,f     : object text\n",
      "siinusrütm     : object text\n",
      "75             : object text\n",
      "mmHg,fr        : object text\n",
      "72             : object text\n",
      "diafragma      : object text\n",
      "diafragma      : object text\n",
      "48*min.        : text  object\n",
      "Siinusrütm     : object text\n",
      "74x            : object text\n",
      "vats.fr.12     : object text\n",
      "EKGSiinusr     : object text\n",
      "80x1           : object text\n",
      "fr.82x'.       : object text\n",
      "(N             : text  object\n",
      ">11)           : text  object\n",
      "fr75xmin       : object text\n",
      "=              : text  object\n",
      "81             : object text\n",
      "regulaarne     : object text\n",
      "70             : object text\n",
      "72             : object text\n",
      "fr.74x'.El     : object text\n",
      "fr.74x'.El     : object text\n",
      "fr.74x'.El     : object text\n",
      "fr.74x'.El     : object text\n",
      "140/90mmHg     : object text\n",
      "80´            : object text\n",
      "(N             : text  object\n",
      "<11)_          : text  object\n",
      "108/73         : object text\n",
      "108/73         : object text\n",
      "126/74,        : object text\n",
      "55             : text  object\n",
      "84,            : text  object\n",
      "168/70,        : object text\n",
      "x/min,         : text  object\n",
      "EKG-fr         : object text\n",
      "71,            : object text\n",
      "mmHg,fr        : object text\n",
      "58             : object text\n",
      "rütm,fr        : object text\n",
      "64             : object text\n",
      "fr78.          : object text\n",
      "frekv.-ga      : object text\n",
      "2000           : text  object\n",
      "=              : text  object\n",
      "65             : object text\n",
      "135/85.fr      : object text\n",
      "66             : object text\n",
      "79             : object text\n",
      "=              : text  object\n",
      "79             : object text\n",
      "mmHg(närve     : object text\n",
      "89x            : object text\n",
      "140/90,fr      : object text\n",
      "160-170x/m     : object text\n",
      "fr73xmin       : object text\n",
      "SR.fr.66x'     : object text\n",
      "EKG:Siinus     : object text\n",
      "61X1min,no     : object text\n",
      "fr79*---RR     : object text\n",
      "fr79*---RR     : object text\n",
      "mmHg,fr        : object text\n",
      "62             : object text\n",
      "=              : text  object\n",
      "64             : object text\n",
      "164            : object text\n",
      "93,0.Kaeba     : object text\n",
      ",kahinata,     : object text\n",
      "70             : object text\n",
      "115/70         : object text\n",
      "f              : text  object\n",
      "reg            : text  object\n",
      "x/min,         : text  object\n",
      "124/62         : object text\n",
      "ole,fr         : object text\n",
      "59x            : object text\n",
      "fr.86,R-R      : object text\n",
      "fr75,tuli      : object text\n",
      "fr72x*         : object text\n",
      "x/min,         : text  object\n",
      "120/80         : object text\n",
      "71             : object text\n",
      "fr62*min       : object text\n",
      "ü              : text  object\n",
      "Elektrokar     : text  object\n",
      "ü              : text  PAD\n",
      "fr76x          : object PAD\n",
      "-              : text  object\n",
      "82             : object text\n",
      "(N             : text  object\n",
      ">11)           : text  object\n",
      "150/90,fr      : object text\n",
      "68;            : object text\n",
      "siinusrütm     : object text\n",
      "88,normaal     : object text\n",
      "Cor-fr         : object text\n",
      "64             : object text\n",
      "=              : text  object\n",
      "66             : object text\n",
      "fr150-170x     : object text\n",
      "f              : text  object\n",
      "reg            : text  object\n",
      "f              : text  object\n",
      "reg            : text  object\n",
      "x.             : text  object\n",
      "70*min.        : object text\n",
      "(N             : text  object\n",
      "<6)_           : text  object\n",
      "fr74,vestl     : object text\n",
      "119            : text  object\n",
      "119            : text  object\n",
      "119            : text  object\n",
      "Siinusrütm     : object text\n",
      "165/83         : object text\n",
      "165/83         : object text\n",
      "55             : text  object\n",
      "55             : text  object\n",
      "55             : text  object\n",
      "55             : text  object\n",
      "x              : text  object\n",
      "=              : text  object\n",
      "110            : object text\n",
      "MEF50          : object text\n",
      "70             : text  object\n",
      "x/min,         : text  object\n",
      "=              : text  object\n",
      "61             : object text\n",
      "mmHg,fr        : object text\n",
      "61x            : object text\n",
      "54             : text  object\n",
      "mmHg__fr63     : object text\n",
      "-              : text  object\n",
      "69             : object text\n",
      "(N             : text  object\n",
      "<11)_          : text  object\n",
      "=              : text  object\n",
      "50             : object text\n",
      "normofr.       : object text\n",
      "115/70,fr      : object text\n",
      "71             : object text\n",
      "fr75           : object PAD\n",
      "n\"/>:fr        : object text\n",
      "70,            : object text\n",
      "arütmia,fr     : object text\n",
      "101,horizo     : object text\n",
      "arütmia,fr     : object text\n",
      "101,horizo     : object text\n",
      "koldeleiut     : object text\n",
      "siinusrütm     : object text\n",
      "81             : object text\n",
      "173/90         : object text\n",
      "173/90         : object text\n",
      "(              : text  object\n",
      "13.10).        : text  object\n",
      "67             : text  object\n",
      "67             : text  object\n",
      "67             : text  object\n",
      "67             : text  object\n",
      "70x',          : object text\n",
      "ref            : text  object\n",
      "Süda,Fa.fr     : object text\n",
      "136/min.       : object text\n",
      "=              : text  object\n",
      "83             : object text\n",
      "=              : text  object\n",
      "61             : object text\n",
      "=              : text  object\n",
      "66             : object text\n",
      "diafragmak     : object text\n",
      "diafragmak     : object text\n",
      "diafragmak     : object text\n",
      "diafragmak     : object text\n",
      "diafragmak     : object text\n",
      "diafragmak     : object text\n",
      "diafragmak     : object text\n",
      "diafragmak     : object text\n",
      "siinusrütm     : object text\n",
      "73             : object text\n",
      "149/78         : object text\n",
      "149/78         : object text\n",
      "66',Pq         : object text\n",
      "160/80         : object text\n",
      "siinusrütm     : object text\n",
      "76             : object text\n",
      "(N             : text  object\n",
      ">3)            : text  object\n",
      "=              : text  object\n",
      "80             : object text\n",
      "mmHg,fr        : object text\n",
      "81             : object text\n",
      "71             : text  object\n",
      "71             : text  object\n",
      "71             : text  object\n",
      "-              : text  object\n",
      "87             : object text\n",
      "x              : text  object\n",
      "õendustoim     : object text\n",
      "60             : text  object\n",
      ",              : text  object\n",
      "siinusbrad     : object text\n",
      "59,normaal     : object text\n",
      "133/75         : object text\n",
      "(              : text  object\n",
      "ravi           : text  object\n",
      "(              : text  object\n",
      "ravi           : text  object\n",
      "mmHg,fr        : object text\n",
      "76             : object text\n",
      "EKG-fr         : object text\n",
      "53,            : object text\n",
      "40.72,         : object text\n",
      "=              : text  object\n",
      "67             : object text\n",
      "siinusbrad     : object text\n",
      "42,normaal     : object text\n",
      "fr.80x`.       : object text\n",
      "127/77,fr      : object text\n",
      "63X            : object text\n",
      "127/77,fr      : object text\n",
      "63X            : object text\n",
      "127/77,fr      : object text\n",
      "63X            : object text\n",
      "140/90         : object text\n",
      "FA,fr          : object text\n",
      "85             : object text\n",
      "FA,fr          : object text\n",
      "85             : object text\n",
      "FA,fr          : object text\n",
      "85             : object text\n",
      "FA,fr          : object text\n",
      "85             : object text\n",
      ".              : text  object\n",
      "72             : object text\n",
      "regul,fr       : object text\n",
      "76             : object text\n",
      "ole,fr         : object text\n",
      "70x            : object text\n",
      "tahhyfrekv     : object text\n",
      "105/65mmHg     : object text\n",
      "mmHg,fr        : object text\n",
      "95x            : object text\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "bradüarütm     : object text\n",
      "58             : object text\n",
      "bradüarütm     : object text\n",
      "58             : object text\n",
      "fr74x`,        : object text\n",
      "mmHg,fr        : object text\n",
      "56             : object text\n",
      "EKG-normof     : object text\n",
      "arütmia,fr     : object text\n",
      "72,horizon     : object text\n",
      "Siinusrütm     : object text\n",
      "80/min.        : object text\n",
      "fr73x1         : object text\n",
      "=              : text  object\n",
      "75             : object text\n",
      "frekvents      : object text\n",
      "frekvents      : object text\n",
      "frekvents      : object text\n",
      "siinusrütm     : object text\n",
      "80             : object text\n",
      "Siinusbrad     : object text\n",
      "55/min.        : object text\n",
      "x/min,         : text  object\n",
      "Siinusrütm     : object text\n",
      "60x'.          : object text\n",
      "ekg-siinus     : object text\n",
      "68x            : object text\n",
      "fr61           : object PAD\n",
      "nefrekoomi     : object text\n",
      "130/85         : object text\n",
      "64,            : text  object\n",
      "416            : object text\n",
      "416            : object text\n",
      "416            : object text\n",
      "416            : object text\n",
      "416            : object text\n",
      "416            : object text\n",
      "Tahhüfrekv     : object text\n",
      "isheemia.,     : object text\n",
      "siinusrütm     : object text\n",
      "81,normaal     : object text\n",
      "fr=76/min      : object text\n",
      "arütmia,fr     : object text\n",
      "96             : object text\n",
      "=              : text  object\n",
      "81             : object text\n",
      "fr.81          : object text\n",
      "fr.71x'.<A     : object text\n",
      "fr.71x'.<A     : object text\n",
      "rr             : text  object\n",
      "150-80         : text  object\n",
      "mmHg,fr        : object text\n",
      "66             : object text\n",
      "(N             : text  object\n",
      ">6)            : text  object\n",
      "bradüarütm     : object text\n",
      "66             : object text\n",
      "P              : text  object\n",
      "70             : object text\n",
      "diafragma      : object text\n",
      "~121           : text  object\n",
      "~121           : text  object\n",
      "~121           : text  object\n",
      "162x`kopsu     : object text\n",
      "140/84,fr      : object text\n",
      "68.            : object text\n",
      "150/80,fr      : object text\n",
      "65,reg.,       : object text\n",
      "=              : text  object\n",
      "58             : object text\n",
      "siinusrütm     : object text\n",
      "66             : object text\n",
      "64,3           : object text\n",
      "64,3           : object text\n",
      "146/70,        : object text\n",
      "190/110        : object text\n",
      "(N             : text  object\n",
      "<3)_           : text  object\n",
      "9061-RR134     : object text\n",
      "fr58           : object PAD\n",
      "siinusrütm     : object text\n",
      "70             : object text\n",
      "mmHg.fr        : object text\n",
      "66             : object text\n",
      "siinusrütm     : object text\n",
      "130/70         : object text\n",
      "EKG:fr         : object text\n",
      "69x',norm.     : object text\n",
      "ole,fr         : object text\n",
      "70x            : object text\n",
      "135/80mmHg     : object text\n",
      "cor-fr75x`     : object text\n",
      "siinustahü     : object text\n",
      "93,normaln     : object text\n",
      "fr100,vest     : object text\n",
      "180/100,fr     : object text\n",
      "99;            : object text\n",
      "55             : object text\n",
      "127782,fr      : object text\n",
      "79.Suhkur      : object text\n"
     ]
    }
   ],
   "source": [
    "j = 0\n",
    "for i in range(len(y_pred)):\n",
    "    p = np.argmax(y_pred[i], axis=-1)\n",
    "#     print(\"{:15}||{:5}||{}\".format(\"Word\", \"True\", \"Pred\"))\n",
    "#     print(30 * \"=\")\n",
    "    for w, t, pred in zip(X_char_te[i], y_te[i], p):\n",
    "        if charindx2word(w) != \"\": #PAD\n",
    "            if idx2tag[t] != idx2tag[pred]:\n",
    "                print(\"{:15}: {:5} {}\".format(charindx2word(w), idx2tag[t], idx2tag[pred]))\n",
    "                j = j+1\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "534"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "j"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Word           ||True ||Pred\n",
      "==============================\n",
      "05.09.2012     : text  text\n",
      "-              : text  text\n",
      "Parem.Süda     : text  text\n",
      "puuduvad.      : text  text\n",
      "Hommikuks      : text  text\n",
      "RR             : text  text\n",
      "kõrgem.        : text  text\n",
      "Obj:           : text  text\n",
      "RR             : text  text\n",
      "165/60         : text  text\n",
      "mmHg,          : text  text\n",
      "fr             : object object\n",
      "64x            : object object\n",
      "regul,turs     : text  text\n",
      "ei             : text  text\n",
      "ole.           : text  text\n"
     ]
    }
   ],
   "source": [
    "i = 1001\n",
    "p = np.argmax(y_pred[i], axis=-1)\n",
    "print(\"{:15}||{:5}||{}\".format(\"Word\", \"True\", \"Pred\"))\n",
    "print(30 * \"=\")\n",
    "for w, t, pred in zip(X_char_te[i], y_te[i], p):\n",
    "    if t != 0: #PAD\n",
    "        print(\"{:15}: {:5} {}\".format(charindx2word(w), idx2tag[t], idx2tag[pred]))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1-score: 89.2%\n",
      "           precision    recall  f1-score   support\n",
      "\n",
      "     text       0.93      0.88      0.91      5198\n",
      "   object       0.92      0.80      0.86      1992\n",
      "\n",
      "micro avg       0.92      0.86      0.89      7190\n",
      "macro avg       0.92      0.86      0.89      7190\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from seqeval.metrics import accuracy_score, precision_score, recall_score, f1_score, classification_report\n",
    "\n",
    "def pred2label(pred):\n",
    "    out = []\n",
    "    for pred_i in pred:\n",
    "        out_i = []\n",
    "        for p in pred_i:\n",
    "            p_i = np.argmax(p)\n",
    "            out_i.append(idx2tag[p_i].replace(\"PAD\", \"text\"))\n",
    "        out.append(out_i)\n",
    "    return out\n",
    "\n",
    "def test2label(pred):\n",
    "    out = []\n",
    "    for pred_i in pred:\n",
    "        out_i = []\n",
    "        for p in pred_i:\n",
    "            out_i.append(idx2tag[p].replace(\"PAD\", \"text\"))\n",
    "        out.append(out_i)\n",
    "    return out\n",
    "\n",
    "pred_labels = pred2label(y_pred)\n",
    "test_labels = test2label(y_te)\n",
    "\n",
    "print(\"F1-score: {:.1%}\".format(f1_score(test_labels, pred_labels)))\n",
    "print(classification_report(test_labels, pred_labels))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9989093513402906"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(test_labels, pred_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "      object       0.96      0.89      0.93      3707\n",
      "        text       1.00      1.00      1.00    484993\n",
      "\n",
      "    accuracy                           1.00    488700\n",
      "   macro avg       0.98      0.95      0.96    488700\n",
      "weighted avg       1.00      1.00      1.00    488700\n",
      "\n",
      "Accuracy 0.9989093513402906\n"
     ]
    }
   ],
   "source": [
    "from sklearn_crfsuite.metrics import flat_classification_report,flat_accuracy_score\n",
    "report = flat_classification_report(y_pred=pred_labels, y_true=test_labels)\n",
    "print(report)\n",
    "print(\"Accuracy\",flat_accuracy_score(y_pred=pred_labels, y_true=test_labels))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn_crfsuite.metrics import flat_f1_score, flat_precision_score, flat_recall_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9989073050951504\n",
      "0.9989073050951504\n",
      "0.9989073050951504\n"
     ]
    }
   ],
   "source": [
    "def pred2label2(pred):\n",
    "    out = []\n",
    "    for pred_i in pred:\n",
    "        out_i = []\n",
    "        for p in pred_i:\n",
    "            p_i = np.argmax(p)\n",
    "            out_i.append(p_i)\n",
    "        out.append(out_i)\n",
    "    return out\n",
    "y_predNew = pred2label2(y_pred)\n",
    "print(flat_recall_score(y_pred=np.array(y_predNew) , y_true= y_te ,average = \"micro\"))\n",
    "print(flat_f1_score(y_pred=np.array(y_predNew),  y_true= y_te ,average = \"micro\"))\n",
    "print(flat_precision_score(y_pred=np.array(y_predNew), y_true= y_te ,average = \"micro\"))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# from sklearn.metrics import precision_recall_fscore_support\n",
    "# precision_recall_fscore_support(y_pred=pred_labels, y_true=test_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# from sklearn.metrics import multilabel_confusion_matrix\n",
    "# multilabel_confusion_matrix(y_pred=pred_labels, y_true=test_labels)\n"
   ]
  }
 ],
 "metadata": {
  "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
}