notebooks/rnn_ner/RNN_NER.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Дисклеймер\n",
    "Эту тетрадку нужно запускать в колабе. Не мучатесь с установкой библиотек и с обучением на cpu."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# %pip install --upgrade keras\n",
    "# %pip install torch torchvision torchaudio"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Рекуррентные нейронные сети"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Рекуррентные нейронные сети (RNN) предназначены для работы с последовательностями (в том числе текстом). По устройству они похожи на цикл, где каждый шаг связан с предудущими через внешние переменные, в которые записывается нужная информация по каким-то условиям. Так как в рекурентных сетях вычисления происходят последовательно, обучение плохо поддается параллелизации. Простыми словами, рекурректные сети медленные. Но их неэффективность часто компенсируется точностью, поэтому рекурректные сети - это стандартный инструмент при работе с текстами.  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Классификация текста с помощью RNN"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Давайте посмотрим, как в тензорфлоу использовать рекуррентные слои и как их соединять с Embedding и Dense слоями."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Код с прошлых занятий"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3.0.5\n"
     ]
    }
   ],
   "source": [
    "# в новой версии кераса можно использовать разные бекэнды, можно попробовать торч\n",
    "# если заменить на tensorflow или jax то код также будет работать\n",
    "# но нужно заранее установить нужный фреймворк\n",
    "import os\n",
    "os.environ[\"KERAS_BACKEND\"] = \"torch\"\n",
    "\n",
    "import keras\n",
    "print(keras.__version__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from string import punctuation\n",
    "from sklearn.model_selection import train_test_split\n",
    "from collections import Counter\n",
    "# from IPython.display import Image\n",
    "# from IPython.core.display import HTML \n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Возьмем небольшой кусок датасета [lenta.ru](https://github.com/yutkin/Lenta.Ru-News-Dataset). Попробуем предсказывать тематику текста"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = pd.read_csv('../../data/lenta_40k.csv.zip')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>topic</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Россия должна сотрудничать с Всемирным антидоп...</td>\n",
       "      <td>Спорт</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Уголовный суд Кувейта 28 июня освободил под за...</td>\n",
       "      <td>Мир</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Французский журнал Charlie Hebdo опубликовал н...</td>\n",
       "      <td>Интернет и СМИ</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>В Петербурге в доме № 53 по улице Лени Голиков...</td>\n",
       "      <td>Россия</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>В московском аэропорту \"Домодедово\" задержан г...</td>\n",
       "      <td>Россия</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                text           topic\n",
       "0  Россия должна сотрудничать с Всемирным антидоп...           Спорт\n",
       "1  Уголовный суд Кувейта 28 июня освободил под за...             Мир\n",
       "2  Французский журнал Charlie Hebdo опубликовал н...  Интернет и СМИ\n",
       "3  В Петербурге в доме № 53 по улице Лени Голиков...          Россия\n",
       "4  В московском аэропорту \"Домодедово\" задержан г...          Россия"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['Спорт', 'Мир', 'Интернет и СМИ', 'Россия', 'Наука и техника',\n",
       "       'Силовые структуры', 'Бывший СССР', 'Культура', 'Экономика',\n",
       "       'Ценности', 'Из жизни', 'Дом', 'Бизнес', '69-я параллель', 'Крым',\n",
       "       'Путешествия', 'Культпросвет ', 'Легпром', 'Библиотека'],\n",
       "      dtype=object)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.topic.unique()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Предобработка нам тут особо не важна."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def preprocess(text):\n",
    "    tokens = text.lower().split()\n",
    "    tokens = [token.strip(punctuation) for token in tokens]\n",
    "    return tokens"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Как и раньше, создаем словарь, фильтруем, чтобы он был адекватного размера и переводим токены в индексы"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# словарь\n",
    "vocab = Counter()\n",
    "\n",
    "for text in data.text:\n",
    "    vocab.update(preprocess(text))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "354611"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# отфильтрованный словарь\n",
    "filtered_vocab = set()\n",
    "\n",
    "for word in vocab:\n",
    "    if vocab[word] > 30:\n",
    "        filtered_vocab.add(word)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "24091"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(filtered_vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# индексируем слова\n",
    "word2id = {'PAD':0, 'UNK':1}\n",
    "\n",
    "for word in filtered_vocab:\n",
    "    word2id[word] = len(word2id)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "id2word = {i:word for word, i in word2id.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# переводим тексты в последовательности индексов\n",
    "X = []\n",
    "\n",
    "for text in data.text:\n",
    "    tokens = preprocess(text)\n",
    "    ids = [word2id.get(token, 1) for token in tokens]\n",
    "    X.append(ids)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "MAX_LEN = max(len(x) for x in X)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Можно посчитать среднюю или медианную длину, если максимальная кажется слишком большой."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "MEAN_LEN = np.median([len(x) for x in X])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1748, 170.0)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "MAX_LEN, MEAN_LEN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "MAX_LEN = int(MEAN_LEN + 30)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# паддинг\n",
    "X = keras.preprocessing.sequence.pad_sequences(X, maxlen=MAX_LEN)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(44356, 200)"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "id2label = {i:label for i, label in enumerate(set(data.topic.values))}\n",
    "label2id = {l:i for i, l in id2label.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "y = keras.utils.to_categorical([label2id[label] for label in data.topic.values])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "19"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(label2id)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "В этот раз добавим стратификацию, т.к. в данных у нас дисбаланс классов."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_valid, y_train, y_valid = train_test_split(X, y, test_size=0.05, stratify=y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Есть несколько разновидностей рекуррентных слоев - RNN (vanila rnn), GRU, LSTM. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Первый - самый простой рекуррентный слой, где на каждом шаге используется предыдущее состояние и текущий вектор, они складываются и умножаются на матрицу весов с функцией активации, полученный вектор передается на следующий шаг. Проблема с обычными RNN в том, что информация о предыдущих состояних очень быстро затирается и учитывать нелокальные зависимости не получается. \n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Можно **схематично** представить Vanilla RNN вот таким циклом в питоне:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "# это вектор внутреннего состояния, можно его рассматривать как память \n",
    "hidden_state = [0, 0, ...] \n",
    "\n",
    "# ванила рнн это цикл по векторам слов в тексте\n",
    "for token_vector in text_vectors:\n",
    "    # на каждом шаге вектор текущего слова складывается с вектором памяти и к ним применяется\n",
    "    # полносвязный слой и функция активации tanh (гиперболический тангенс)\n",
    "    # tanh переводит числа в интервал от -1 до 1\n",
    "    hidden_state = tanh(W(hidden_state + token_vector))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "LSTM и GRU частично решают эту проблему добавлением дополнительных матриц весов и активаций, которые будут отвечать за то, чтобы нужная информация сохранялась на протяжении всей последовательности, а ненужная удалялась."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Схематично LSTM можно представить вот так:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# в LSTM две переменные для памяти (отсюда и название Long-short memory network, \n",
    "#                                одна переменная для короткой памяти другая для долгой)\n",
    "# cell_state используется только внутри ячейки (это long memory)\n",
    "cell_state = [0, 0, ...]\n",
    "# а hidden_state также передается наружу (его можно использовать для классификации)\n",
    "hidden_state = [0, 0, ...]\n",
    "\n",
    "# цикл точно также идет по векторам слов в тексте\n",
    "for token_vector in text_vectors:\n",
    "    # запоминание/забывание в LSTM реализовано с помощью механизма, который называется gate\n",
    "    # есть три гейта - гейт забывания (forget_gate), входной гейт (input gate), и выходной гейт (output_gate)\n",
    "    # гейт - это полносвязный слой с функцией активации сигмоида (она выдает числа от 0 до 1)\n",
    "    # по сути это веса которые пременяются к hidden_state, cell_state\n",
    "    # например, результат forget_gate может выглядеть как [0.02, 0.01, 0.99]\n",
    "    # он будет означать что нужно почти полностью забыть то что хранится в 1 и 2 размерности, но оставить то что в 3\n",
    "    forget_gate = sigmoid(W_forget(hidden_state + token_vector))\n",
    "    input_gate = sigmoid(W_input(hidden_state + token_vector))\n",
    "    output_gate = sigmoid(W_output(hidden_state + token_vector))\n",
    "    \n",
    "    # гейты это веса, а сами значения для векторов cell_state и hidden_state получаются через tanh\n",
    "    # обновление cell_state, hidden_state происходит следующим образом:\n",
    "    # 1. forget_gate применяется к cell_state. Это поточечное умножение двух векторов \n",
    "    # (в питоне * означает обычное умножение, правильно было бы написать pointwise_multiply\n",
    "    #  но т.к. это просто схема я написал просто *) \n",
    "    # Этот шаг отвечает за удаление из долгой памяти\n",
    "    # Т.е гейт смотрит на текущую кратковременную память и текущее слова и решает, что можно удалить из долгой памяти\n",
    "    cell_state_forgotten = forget_gate * cell_state\n",
    "    \n",
    "    # 2. в cell_state сохраняется новая информация\n",
    "    # Чтобы что-то сохранить в долгую память нужно скомбинировать текущее слово и короткую память\n",
    "    # и привести это к формату в котором хранится долговременная память \n",
    "    # Комбинирование - это просто сложение hidden_state и token_vector\n",
    "    # А приведение к нужному формату это полносвязный слой и все тот же tanh\n",
    "    candidate = tanh(W_candidate(hidden_state + token_vector))\n",
    "    # важность новой информации определяет input_gate\n",
    "    input = candidate * input_gate\n",
    "    # итоговый cell_state получается удалением ненужного и добавлением актуального\n",
    "    cell_state = cell_state_forgotten + input\n",
    "    # таким образом на каждом шаге гейты забывания и запоминания решают что и в какой пропорции помнить, а что забыть\n",
    "    # обратите внимание, что это не статичные веса (10 % забываем,  20% запоминаем), \n",
    "    # каждый раз веса получаются разные в зависимости от текущего слова и короткой памяти\n",
    "    \n",
    "    # 3. обновляется короткая память (hidden_state)\n",
    "    # hidden_state вычисляется из обновленного cell_state с помощью output_gate\n",
    "    # Можно интепретировать это как-то так:\n",
    "    # модель выбирает из памяти только наиболее релевантную текущему слову информацию\n",
    "    # Если текущее слово глагол, то в hidden_state может записаться информация о самом глаголе (значение, время, лицо)\n",
    "    # а также информация о связанных с ним словах (местоимении, именной группе, отрицательной частите)\n",
    "    hidden_state = tanh(cell_state) * output_gate\n",
    "    \n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "А GRU вот так:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# в GRU только одна переменная для памяти\n",
    "hidden_state = [0, 0, ...]\n",
    "\n",
    "for token_vector in text_vectors:\n",
    "    # механизм запоминания/забывания тоже реализован через гейты\n",
    "    # В GRU 2 гейта - reset (забывание) и update (запоминание)\n",
    "    # расчитываются они применением полносвязного слоя к текущему слову и памяти с сигмоидой\n",
    "    reset = sigmoid(W_reset(token_vector + hidden_state))\n",
    "    update = sigmoid(W_update(token_vector + hidden_state))\n",
    "    \n",
    "    # Обновление hidden_state устроено следующим образом:\n",
    "    # 1. Из памяти выбирается только нужная информация с помощью reset gate\n",
    "    # На основе актуальной памяти и текущего вектора считается новый потенциальный вектор hidden_state\n",
    "    # Это делается полносвязным слоем и tanh\n",
    "    candidate_hidden_state = tanh(W_candidate((hidden_state * reset) + token_vector)) \n",
    "    \n",
    "    # 2. update gate решает в какой пропорции совместить память и потенциальный hidden_state\n",
    "    # тут два раза используется update но во втором случае он вычитается из единицы\n",
    "    # вычитание из единицы отзеркаливает update (если update [0.2, 0.45], то 1-update = [0.8, 0.55])\n",
    "    # Если значения в update маленькие, то в 1-update будут большие, т.е. \n",
    "    # модель возьмет почти все из памяти и почти ничего из потенциального вектора (и наоборот)\n",
    "    hidden_state = (update * candidate_hidden_state) + ( (1-update) * hidden_state )\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "На картинке все три типа можно изобразить вот так:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![lstm cell](https://habrastorage.org/webt/xt/_q/nj/xt_qnjgfjengqoqd4gizkq4j_wk.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "На практике выбор стоит между LSTM и GRU. По описанию и графику можно заметить, что в LSTM 4 матрицы с весами, а в GRU 3. Таким образом, GRU - немного проще (а значит быстрее) LSTM. Но это не значит, что LSTM лучше работает (сложное не значит хорошее). Как и с другими алгоритмами машинного обучения все зависит от задачи. Лучше пробовать и то и другое (или даже совмещать)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Ссылки, чтобы разобраться подробнее:  \n",
    "    http://karpathy.github.io/2015/05/21/rnn-effectiveness/  \n",
    "    https://towardsdatascience.com/illustrated-guide-to-lstms-and-gru-s-a-step-by-step-explanation-44e9eb85bf21  \n",
    "    https://colah.github.io/posts/2015-08-Understanding-LSTMs/  \n",
    "    \n",
    "    \n",
    "    \n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Использовать RNN слой проще, чем понять, как он устроен. Нужно просто поставить его после другого слоя, который возвращает последовательность (например, после Embedding слоя).\n",
    "\n",
    "При return_sequences=False в следующий слой будет передаваться только последнее состояние - 1 вектор, т.е. не нужно конкатенировать"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = keras.layers.Input(shape=(MAX_LEN,))\n",
    "embeddings = keras.layers.Embedding(input_dim=len(word2id), output_dim=30)(inputs, )\n",
    "\n",
    "\n",
    "# rnn = keras.layers.SimpleRNN(128, return_sequences=False)(embeddings)\n",
    "rnn = keras.layers.LSTM(128, return_sequences=False)(embeddings)\n",
    "# rnn = keras.layers.GRU(128, return_sequences=False)(embeddings)\n",
    "\n",
    "outputs = keras.layers.Dense(len(label2id), activation='softmax')(rnn)\n",
    "\n",
    "model = keras.Model(inputs=inputs, outputs=outputs)\n",
    "optimizer = keras.optimizers.Adam(learning_rate=0.001)\n",
    "model.compile(optimizer=optimizer,\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=[keras.metrics.RecallAtPrecision(0.8, name='rec@prec')])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Схема задачи, чтобы было понятнее:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![image](./image_lstm_clf.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"functional_1\"</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1mModel: \"functional_1\"\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\"> Layer (type)                    </span>┃<span style=\"font-weight: bold\"> Output Shape           </span>┃<span style=\"font-weight: bold\">       Param # </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ input_layer (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)        │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">200</span>)            │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ embedding (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Embedding</span>)           │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">200</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">30</span>)        │       <span style=\"color: #00af00; text-decoration-color: #00af00\">722,790</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ lstm (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">LSTM</span>)                     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)            │        <span style=\"color: #00af00; text-decoration-color: #00af00\">81,408</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                   │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">19</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">2,451</span> │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape          \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ input_layer (\u001b[38;5;33mInputLayer\u001b[0m)        │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m200\u001b[0m)            │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ embedding (\u001b[38;5;33mEmbedding\u001b[0m)           │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m200\u001b[0m, \u001b[38;5;34m30\u001b[0m)        │       \u001b[38;5;34m722,790\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ lstm (\u001b[38;5;33mLSTM\u001b[0m)                     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m128\u001b[0m)            │        \u001b[38;5;34m81,408\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense (\u001b[38;5;33mDense\u001b[0m)                   │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m19\u001b[0m)             │         \u001b[38;5;34m2,451\u001b[0m │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">806,649</span> (3.08 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m806,649\u001b[0m (3.08 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">806,649</span> (3.08 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m806,649\u001b[0m (3.08 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m32s\u001b[0m 749ms/step - loss: 2.6669 - rec@prec: 0.0000e+00 - val_loss: 2.3388 - val_rec@prec: 0.0000e+00\n",
      "Epoch 2/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m32s\u001b[0m 733ms/step - loss: 2.3144 - rec@prec: 9.7443e-05 - val_loss: 2.3152 - val_rec@prec: 0.0000e+00\n",
      "Epoch 3/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m32s\u001b[0m 744ms/step - loss: 2.2636 - rec@prec: 0.0013 - val_loss: 2.1355 - val_rec@prec: 0.0356\n",
      "Epoch 4/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m32s\u001b[0m 756ms/step - loss: 1.9367 - rec@prec: 0.0904 - val_loss: 1.6986 - val_rec@prec: 0.1691\n",
      "Epoch 5/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m32s\u001b[0m 749ms/step - loss: 1.5295 - rec@prec: 0.2631 - val_loss: 1.4169 - val_rec@prec: 0.2831\n",
      "Epoch 6/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m33s\u001b[0m 762ms/step - loss: 1.2786 - rec@prec: 0.4337 - val_loss: 1.4722 - val_rec@prec: 0.3228\n",
      "Epoch 7/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m33s\u001b[0m 764ms/step - loss: 1.1959 - rec@prec: 0.4747 - val_loss: 1.2603 - val_rec@prec: 0.4234\n",
      "Epoch 8/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m33s\u001b[0m 770ms/step - loss: 1.0104 - rec@prec: 0.6088 - val_loss: 1.1568 - val_rec@prec: 0.5167\n",
      "Epoch 9/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m35s\u001b[0m 817ms/step - loss: 0.8841 - rec@prec: 0.6993 - val_loss: 1.1179 - val_rec@prec: 0.5374\n",
      "Epoch 10/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m36s\u001b[0m 842ms/step - loss: 0.7756 - rec@prec: 0.7614 - val_loss: 1.1085 - val_rec@prec: 0.5618\n",
      "Epoch 11/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m38s\u001b[0m 868ms/step - loss: 0.7077 - rec@prec: 0.8020 - val_loss: 1.1220 - val_rec@prec: 0.5812\n",
      "Epoch 12/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m37s\u001b[0m 859ms/step - loss: 0.6225 - rec@prec: 0.8429 - val_loss: 1.0915 - val_rec@prec: 0.6091\n",
      "Epoch 13/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m37s\u001b[0m 868ms/step - loss: 0.5617 - rec@prec: 0.8694 - val_loss: 1.1209 - val_rec@prec: 0.6168\n",
      "Epoch 14/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m35s\u001b[0m 823ms/step - loss: 0.5178 - rec@prec: 0.8881 - val_loss: 1.1195 - val_rec@prec: 0.6186\n",
      "Epoch 15/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m35s\u001b[0m 818ms/step - loss: 0.4437 - rec@prec: 0.9160 - val_loss: 1.1588 - val_rec@prec: 0.6105\n",
      "Epoch 16/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m36s\u001b[0m 827ms/step - loss: 0.4141 - rec@prec: 0.9239 - val_loss: 1.1731 - val_rec@prec: 0.5857\n",
      "Epoch 17/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m36s\u001b[0m 832ms/step - loss: 0.3930 - rec@prec: 0.9336 - val_loss: 1.2109 - val_rec@prec: 0.5812\n",
      "Epoch 18/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m36s\u001b[0m 837ms/step - loss: 0.3378 - rec@prec: 0.9488 - val_loss: 1.2871 - val_rec@prec: 0.5825\n",
      "Epoch 19/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m36s\u001b[0m 838ms/step - loss: 0.3284 - rec@prec: 0.9505 - val_loss: 1.2751 - val_rec@prec: 0.5798\n",
      "Epoch 20/20\n",
      "\u001b[1m43/43\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m36s\u001b[0m 847ms/step - loss: 0.3025 - rec@prec: 0.9594 - val_loss: 1.2920 - val_rec@prec: 0.6028\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.src.callbacks.history.History at 0x2abddfbe0>"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train, y_train, \n",
    "          validation_data=(X_valid, y_valid),\n",
    "          batch_size=1000,\n",
    "         epochs=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(model.history.history['rec@prec'])\n",
    "plt.plot(model.history.history['val_rec@prec'])\n",
    "plt.title('model metrics')\n",
    "plt.ylabel('recall at precision')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Рекуррентные слои можно накладывать друг на друга. Нужно только указать параметр return_sequences=True, т.к. слой ожидает на вход последовательность. Длина последовательности не изменяется, поэтому можно совмещать сколько угодно слоев. Это называется stacked LSTM/GRU."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Основной параметр в LSTM/GRU слое - units. Чем он больше, тем больше весов будет в модели. Он будет соответствовать размеру вектора состояния (hidden_state). Он может быть любой даже при накладывании слоев друг на друга (они могут быть одинаковые как тут, первый может быть больше, второй может быть больше)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = keras.layers.Input(shape=(MAX_LEN,))\n",
    "embeddings = keras.layers.Embedding(input_dim=len(word2id), output_dim=30)(inputs, )\n",
    "\n",
    "lstm_1 = keras.layers.LSTM(128, return_sequences=True)(embeddings)\n",
    "lstm_2 = keras.layers.GRU(128, return_sequences=False)(lstm_1)\n",
    "\n",
    "dense = keras.layers.Dense(64, activation='relu')(lstm_2)\n",
    "outputs = keras.layers.Dense(len(label2id), activation='softmax')(dense)\n",
    "\n",
    "model = keras.Model(inputs=inputs, outputs=outputs)\n",
    "optimizer = keras.optimizers.Adam(learning_rate=0.001)\n",
    "model.compile(optimizer=optimizer,\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=[keras.metrics.RecallAtPrecision(0.8, name='rec@prec')])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"functional_3\"</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1mModel: \"functional_3\"\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\"> Layer (type)                    </span>┃<span style=\"font-weight: bold\"> Output Shape           </span>┃<span style=\"font-weight: bold\">       Param # </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ input_layer_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)      │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">200</span>)            │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ embedding_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Embedding</span>)         │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">200</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">30</span>)        │       <span style=\"color: #00af00; text-decoration-color: #00af00\">722,790</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ lstm_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">LSTM</span>)                   │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">200</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)       │        <span style=\"color: #00af00; text-decoration-color: #00af00\">81,408</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ gru (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">GRU</span>)                       │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)            │        <span style=\"color: #00af00; text-decoration-color: #00af00\">99,072</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                 │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">8,256</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_2 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                 │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">19</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">1,235</span> │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape          \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ input_layer_1 (\u001b[38;5;33mInputLayer\u001b[0m)      │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m200\u001b[0m)            │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ embedding_1 (\u001b[38;5;33mEmbedding\u001b[0m)         │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m200\u001b[0m, \u001b[38;5;34m30\u001b[0m)        │       \u001b[38;5;34m722,790\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ lstm_1 (\u001b[38;5;33mLSTM\u001b[0m)                   │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m200\u001b[0m, \u001b[38;5;34m128\u001b[0m)       │        \u001b[38;5;34m81,408\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ gru (\u001b[38;5;33mGRU\u001b[0m)                       │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m128\u001b[0m)            │        \u001b[38;5;34m99,072\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_1 (\u001b[38;5;33mDense\u001b[0m)                 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m64\u001b[0m)             │         \u001b[38;5;34m8,256\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_2 (\u001b[38;5;33mDense\u001b[0m)                 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m19\u001b[0m)             │         \u001b[38;5;34m1,235\u001b[0m │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">912,761</span> (3.48 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m912,761\u001b[0m (3.48 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">912,761</span> (3.48 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m912,761\u001b[0m (3.48 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m148s\u001b[0m 699ms/step - loss: 2.4180 - rec@prec: 8.1810e-05 - val_loss: 1.7612 - val_rec@prec: 0.0397\n",
      "Epoch 2/15\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m154s\u001b[0m 733ms/step - loss: 1.5459 - rec@prec: 0.1098 - val_loss: 1.1927 - val_rec@prec: 0.4337\n",
      "Epoch 3/15\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m160s\u001b[0m 758ms/step - loss: 1.0095 - rec@prec: 0.5730 - val_loss: 1.1259 - val_rec@prec: 0.5460\n",
      "Epoch 4/15\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m160s\u001b[0m 758ms/step - loss: 0.7695 - rec@prec: 0.7455 - val_loss: 1.0720 - val_rec@prec: 0.6073\n",
      "Epoch 5/15\n",
      "\u001b[1m 10/211\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2:20\u001b[0m 699ms/step - loss: 0.6399 - rec@prec: 0.8350"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "Cell \u001b[0;32mIn [35], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m \u001b[43mmodel\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mfit\u001b[49m\u001b[43m(\u001b[49m\u001b[43mX_train\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43my_train\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\n\u001b[1;32m      2\u001b[0m \u001b[43m          \u001b[49m\u001b[43mvalidation_data\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mX_valid\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43my_valid\u001b[49m\u001b[43m)\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m      3\u001b[0m \u001b[43m          \u001b[49m\u001b[43mbatch_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m200\u001b[39;49m\u001b[43m,\u001b[49m\n\u001b[1;32m      4\u001b[0m \u001b[43m          \u001b[49m\u001b[43mepochs\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;241;43m15\u001b[39;49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:118\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    116\u001b[0m filtered_tb \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    117\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 118\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    119\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    120\u001b[0m     filtered_tb \u001b[38;5;241m=\u001b[39m _process_traceback_frames(e\u001b[38;5;241m.\u001b[39m__traceback__)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/backend/torch/trainer.py:249\u001b[0m, in \u001b[0;36mTorchTrainer.fit\u001b[0;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq)\u001b[0m\n\u001b[1;32m    245\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m step, data \u001b[38;5;129;01min\u001b[39;00m epoch_iterator\u001b[38;5;241m.\u001b[39menumerate_epoch():\n\u001b[1;32m    246\u001b[0m     \u001b[38;5;66;03m# Callbacks\u001b[39;00m\n\u001b[1;32m    247\u001b[0m     callbacks\u001b[38;5;241m.\u001b[39mon_train_batch_begin(step)\n\u001b[0;32m--> 249\u001b[0m     logs \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtrain_function\u001b[49m\u001b[43m(\u001b[49m\u001b[43mdata\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    251\u001b[0m     \u001b[38;5;66;03m# Callbacks\u001b[39;00m\n\u001b[1;32m    252\u001b[0m     callbacks\u001b[38;5;241m.\u001b[39mon_train_batch_end(step, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_pythonify_logs(logs))\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/backend/torch/trainer.py:112\u001b[0m, in \u001b[0;36mTorchTrainer.make_train_function.<locals>.one_step_on_data\u001b[0;34m(data)\u001b[0m\n\u001b[1;32m    110\u001b[0m \u001b[38;5;124;03m\"\"\"Runs a single training step on a batch of data.\"\"\"\u001b[39;00m\n\u001b[1;32m    111\u001b[0m data \u001b[38;5;241m=\u001b[39m data[\u001b[38;5;241m0\u001b[39m]\n\u001b[0;32m--> 112\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtrain_step\u001b[49m\u001b[43m(\u001b[49m\u001b[43mdata\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/backend/torch/trainer.py:43\u001b[0m, in \u001b[0;36mTorchTrainer.train_step\u001b[0;34m(self, data)\u001b[0m\n\u001b[1;32m     41\u001b[0m \u001b[38;5;66;03m# Compute predictions\u001b[39;00m\n\u001b[1;32m     42\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_call_has_training_arg:\n\u001b[0;32m---> 43\u001b[0m     y_pred \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mx\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtraining\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mTrue\u001b[39;49;00m\u001b[43m)\u001b[49m\n\u001b[1;32m     44\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[1;32m     45\u001b[0m     y_pred \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m(x)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:118\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    116\u001b[0m filtered_tb \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    117\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 118\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    119\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    120\u001b[0m     filtered_tb \u001b[38;5;241m=\u001b[39m _process_traceback_frames(e\u001b[38;5;241m.\u001b[39m__traceback__)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/layers/layer.py:816\u001b[0m, in \u001b[0;36mLayer.__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m    814\u001b[0m         outputs \u001b[38;5;241m=\u001b[39m \u001b[38;5;28msuper\u001b[39m()\u001b[38;5;241m.\u001b[39m\u001b[38;5;21m__call__\u001b[39m(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)\n\u001b[1;32m    815\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m--> 816\u001b[0m     outputs \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__call__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    817\u001b[0m \u001b[38;5;66;03m# Change the layout for the layer output if needed.\u001b[39;00m\n\u001b[1;32m    818\u001b[0m \u001b[38;5;66;03m# This is useful for relayout intermediate tensor in the model\u001b[39;00m\n\u001b[1;32m    819\u001b[0m \u001b[38;5;66;03m# to achieve the optimal performance.\u001b[39;00m\n\u001b[1;32m    820\u001b[0m distribution \u001b[38;5;241m=\u001b[39m distribution_lib\u001b[38;5;241m.\u001b[39mdistribution()\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/torch/nn/modules/module.py:1511\u001b[0m, in \u001b[0;36mModule._wrapped_call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1509\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_compiled_call_impl(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[1;32m   1510\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1511\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_call_impl\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/torch/nn/modules/module.py:1520\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1515\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1516\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1517\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1518\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1519\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1520\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1522\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[1;32m   1523\u001b[0m     result \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/backend/torch/layer.py:26\u001b[0m, in \u001b[0;36mTorchLayer.forward\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m     25\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mforward\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs):\n\u001b[0;32m---> 26\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mOperation\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__call__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:118\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    116\u001b[0m filtered_tb \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    117\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 118\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    119\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    120\u001b[0m     filtered_tb \u001b[38;5;241m=\u001b[39m _process_traceback_frames(e\u001b[38;5;241m.\u001b[39m__traceback__)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/ops/operation.py:42\u001b[0m, in \u001b[0;36mOperation.__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m     37\u001b[0m         call_fn \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcall\n\u001b[1;32m     38\u001b[0m     call_fn \u001b[38;5;241m=\u001b[39m traceback_utils\u001b[38;5;241m.\u001b[39minject_argument_info_in_traceback(\n\u001b[1;32m     39\u001b[0m         call_fn,\n\u001b[1;32m     40\u001b[0m         object_name\u001b[38;5;241m=\u001b[39m(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__class__\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__name__\u001b[39m\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m.call()\u001b[39m\u001b[38;5;124m\"\u001b[39m),\n\u001b[1;32m     41\u001b[0m     )\n\u001b[0;32m---> 42\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mcall_fn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m     44\u001b[0m \u001b[38;5;66;03m# Plain flow.\u001b[39;00m\n\u001b[1;32m     45\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m any_symbolic_tensors(args, kwargs):\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:157\u001b[0m, in \u001b[0;36minject_argument_info_in_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    155\u001b[0m bound_signature \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    156\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 157\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    158\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    159\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mhasattr\u001b[39m(e, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m_keras_call_info_injected\u001b[39m\u001b[38;5;124m\"\u001b[39m):\n\u001b[1;32m    160\u001b[0m         \u001b[38;5;66;03m# Only inject info for the innermost failing call\u001b[39;00m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/models/functional.py:188\u001b[0m, in \u001b[0;36mFunctional.call\u001b[0;34m(self, inputs, training, mask)\u001b[0m\n\u001b[1;32m    186\u001b[0m         \u001b[38;5;28;01mif\u001b[39;00m mask \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[1;32m    187\u001b[0m             x\u001b[38;5;241m.\u001b[39m_keras_mask \u001b[38;5;241m=\u001b[39m mask\n\u001b[0;32m--> 188\u001b[0m outputs \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_run_through_graph\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    189\u001b[0m \u001b[43m    \u001b[49m\u001b[43minputs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43moperation_fn\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mlambda\u001b[39;49;00m\u001b[43m \u001b[49m\u001b[43mop\u001b[49m\u001b[43m:\u001b[49m\u001b[43m \u001b[49m\u001b[43moperation_fn\u001b[49m\u001b[43m(\u001b[49m\u001b[43mop\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtraining\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtraining\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    190\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    191\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m unpack_singleton(outputs)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/ops/function.py:153\u001b[0m, in \u001b[0;36mFunction._run_through_graph\u001b[0;34m(self, inputs, operation_fn)\u001b[0m\n\u001b[1;32m    150\u001b[0m     \u001b[38;5;28;01mcontinue\u001b[39;00m  \u001b[38;5;66;03m# Node is not computable, try skipping.\u001b[39;00m\n\u001b[1;32m    152\u001b[0m args, kwargs \u001b[38;5;241m=\u001b[39m node\u001b[38;5;241m.\u001b[39marguments\u001b[38;5;241m.\u001b[39mfill_in(tensor_dict)\n\u001b[0;32m--> 153\u001b[0m outputs \u001b[38;5;241m=\u001b[39m \u001b[43moperation_fn\u001b[49m\u001b[43m(\u001b[49m\u001b[43mnode\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43moperation\u001b[49m\u001b[43m)\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    155\u001b[0m \u001b[38;5;66;03m# Update tensor_dict.\u001b[39;00m\n\u001b[1;32m    156\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m x, y \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mzip\u001b[39m(node\u001b[38;5;241m.\u001b[39moutputs, tree\u001b[38;5;241m.\u001b[39mflatten(outputs)):\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/models/functional.py:572\u001b[0m, in \u001b[0;36moperation_fn.<locals>.call\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    566\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m (\n\u001b[1;32m    567\u001b[0m     \u001b[38;5;28mhasattr\u001b[39m(operation, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m_call_has_training_arg\u001b[39m\u001b[38;5;124m\"\u001b[39m)\n\u001b[1;32m    568\u001b[0m     \u001b[38;5;129;01mand\u001b[39;00m operation\u001b[38;5;241m.\u001b[39m_call_has_training_arg\n\u001b[1;32m    569\u001b[0m     \u001b[38;5;129;01mand\u001b[39;00m training \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    570\u001b[0m ):\n\u001b[1;32m    571\u001b[0m     kwargs[\u001b[38;5;124m\"\u001b[39m\u001b[38;5;124mtraining\u001b[39m\u001b[38;5;124m\"\u001b[39m] \u001b[38;5;241m=\u001b[39m training\n\u001b[0;32m--> 572\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43moperation\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:118\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    116\u001b[0m filtered_tb \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    117\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 118\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    119\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    120\u001b[0m     filtered_tb \u001b[38;5;241m=\u001b[39m _process_traceback_frames(e\u001b[38;5;241m.\u001b[39m__traceback__)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/layers/layer.py:816\u001b[0m, in \u001b[0;36mLayer.__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m    814\u001b[0m         outputs \u001b[38;5;241m=\u001b[39m \u001b[38;5;28msuper\u001b[39m()\u001b[38;5;241m.\u001b[39m\u001b[38;5;21m__call__\u001b[39m(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)\n\u001b[1;32m    815\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m--> 816\u001b[0m     outputs \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__call__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    817\u001b[0m \u001b[38;5;66;03m# Change the layout for the layer output if needed.\u001b[39;00m\n\u001b[1;32m    818\u001b[0m \u001b[38;5;66;03m# This is useful for relayout intermediate tensor in the model\u001b[39;00m\n\u001b[1;32m    819\u001b[0m \u001b[38;5;66;03m# to achieve the optimal performance.\u001b[39;00m\n\u001b[1;32m    820\u001b[0m distribution \u001b[38;5;241m=\u001b[39m distribution_lib\u001b[38;5;241m.\u001b[39mdistribution()\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/torch/nn/modules/module.py:1511\u001b[0m, in \u001b[0;36mModule._wrapped_call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1509\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_compiled_call_impl(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[1;32m   1510\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1511\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_call_impl\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/torch/nn/modules/module.py:1520\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1515\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1516\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1517\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1518\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1519\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1520\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1522\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[1;32m   1523\u001b[0m     result \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/backend/torch/layer.py:26\u001b[0m, in \u001b[0;36mTorchLayer.forward\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m     25\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mforward\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs):\n\u001b[0;32m---> 26\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mOperation\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__call__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:118\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    116\u001b[0m filtered_tb \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    117\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 118\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    119\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    120\u001b[0m     filtered_tb \u001b[38;5;241m=\u001b[39m _process_traceback_frames(e\u001b[38;5;241m.\u001b[39m__traceback__)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/ops/operation.py:42\u001b[0m, in \u001b[0;36mOperation.__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m     37\u001b[0m         call_fn \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcall\n\u001b[1;32m     38\u001b[0m     call_fn \u001b[38;5;241m=\u001b[39m traceback_utils\u001b[38;5;241m.\u001b[39minject_argument_info_in_traceback(\n\u001b[1;32m     39\u001b[0m         call_fn,\n\u001b[1;32m     40\u001b[0m         object_name\u001b[38;5;241m=\u001b[39m(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__class__\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__name__\u001b[39m\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m.call()\u001b[39m\u001b[38;5;124m\"\u001b[39m),\n\u001b[1;32m     41\u001b[0m     )\n\u001b[0;32m---> 42\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mcall_fn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m     44\u001b[0m \u001b[38;5;66;03m# Plain flow.\u001b[39;00m\n\u001b[1;32m     45\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m any_symbolic_tensors(args, kwargs):\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:157\u001b[0m, in \u001b[0;36minject_argument_info_in_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    155\u001b[0m bound_signature \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    156\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 157\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    158\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    159\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mhasattr\u001b[39m(e, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m_keras_call_info_injected\u001b[39m\u001b[38;5;124m\"\u001b[39m):\n\u001b[1;32m    160\u001b[0m         \u001b[38;5;66;03m# Only inject info for the innermost failing call\u001b[39;00m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/layers/rnn/lstm.py:538\u001b[0m, in \u001b[0;36mLSTM.call\u001b[0;34m(self, sequences, initial_state, mask, training)\u001b[0m\n\u001b[1;32m    537\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mcall\u001b[39m(\u001b[38;5;28mself\u001b[39m, sequences, initial_state\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mNone\u001b[39;00m, mask\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mNone\u001b[39;00m, training\u001b[38;5;241m=\u001b[39m\u001b[38;5;28;01mFalse\u001b[39;00m):\n\u001b[0;32m--> 538\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mcall\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    539\u001b[0m \u001b[43m        \u001b[49m\u001b[43msequences\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mmask\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmask\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtraining\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtraining\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43minitial_state\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43minitial_state\u001b[49m\n\u001b[1;32m    540\u001b[0m \u001b[43m    \u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/layers/rnn/rnn.py:397\u001b[0m, in \u001b[0;36mRNN.call\u001b[0;34m(self, sequences, initial_state, mask, training)\u001b[0m\n\u001b[1;32m    393\u001b[0m \u001b[38;5;66;03m# Prepopulate the dropout state so that the inner_loop is stateless\u001b[39;00m\n\u001b[1;32m    394\u001b[0m \u001b[38;5;66;03m# this is particularly important for JAX backend.\u001b[39;00m\n\u001b[1;32m    395\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_maybe_config_dropout_masks(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcell, sequences, initial_state)\n\u001b[0;32m--> 397\u001b[0m last_output, outputs, states \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43minner_loop\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    398\u001b[0m \u001b[43m    \u001b[49m\u001b[43msequences\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43msequences\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    399\u001b[0m \u001b[43m    \u001b[49m\u001b[43minitial_state\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43minitial_state\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    400\u001b[0m \u001b[43m    \u001b[49m\u001b[43mmask\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmask\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    401\u001b[0m \u001b[43m    \u001b[49m\u001b[43mtraining\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtraining\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    402\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    403\u001b[0m last_output \u001b[38;5;241m=\u001b[39m ops\u001b[38;5;241m.\u001b[39mcast(last_output, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcompute_dtype)\n\u001b[1;32m    404\u001b[0m outputs \u001b[38;5;241m=\u001b[39m ops\u001b[38;5;241m.\u001b[39mcast(outputs, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcompute_dtype)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/layers/rnn/lstm.py:533\u001b[0m, in \u001b[0;36mLSTM.inner_loop\u001b[0;34m(self, sequences, initial_state, mask, training)\u001b[0m\n\u001b[1;32m    531\u001b[0m     \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mNotImplementedError\u001b[39;00m:\n\u001b[1;32m    532\u001b[0m         \u001b[38;5;28;01mpass\u001b[39;00m\n\u001b[0;32m--> 533\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43minner_loop\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    534\u001b[0m \u001b[43m    \u001b[49m\u001b[43msequences\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43minitial_state\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mmask\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmask\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtraining\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtraining\u001b[49m\n\u001b[1;32m    535\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/layers/rnn/rnn.py:339\u001b[0m, in \u001b[0;36mRNN.inner_loop\u001b[0;34m(self, sequences, initial_state, mask, training)\u001b[0m\n\u001b[1;32m    336\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m tree\u001b[38;5;241m.\u001b[39mis_nested(initial_state):\n\u001b[1;32m    337\u001b[0m     initial_state \u001b[38;5;241m=\u001b[39m [initial_state]\n\u001b[0;32m--> 339\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mbackend\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mrnn\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    340\u001b[0m \u001b[43m    \u001b[49m\u001b[43mstep\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    341\u001b[0m \u001b[43m    \u001b[49m\u001b[43msequences\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    342\u001b[0m \u001b[43m    \u001b[49m\u001b[43minitial_state\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    343\u001b[0m \u001b[43m    \u001b[49m\u001b[43mgo_backwards\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mgo_backwards\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    344\u001b[0m \u001b[43m    \u001b[49m\u001b[43mmask\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmask\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    345\u001b[0m \u001b[43m    \u001b[49m\u001b[43munroll\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43munroll\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    346\u001b[0m \u001b[43m    \u001b[49m\u001b[43minput_length\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43msequences\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mshape\u001b[49m\u001b[43m[\u001b[49m\u001b[38;5;241;43m1\u001b[39;49m\u001b[43m]\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    347\u001b[0m \u001b[43m    \u001b[49m\u001b[43mzero_output_for_mask\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mzero_output_for_mask\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    348\u001b[0m \u001b[43m    \u001b[49m\u001b[43mreturn_all_outputs\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mreturn_sequences\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    349\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/backend/torch/rnn.py:346\u001b[0m, in \u001b[0;36mrnn\u001b[0;34m(step_function, inputs, initial_states, go_backwards, mask, constants, unroll, input_length, time_major, zero_output_for_mask, return_all_outputs)\u001b[0m\n\u001b[1;32m    344\u001b[0m new_states \u001b[38;5;241m=\u001b[39m states\n\u001b[1;32m    345\u001b[0m \u001b[38;5;28;01mwhile\u001b[39;00m time \u001b[38;5;241m<\u001b[39m time_steps_t \u001b[38;5;129;01mand\u001b[39;00m it \u001b[38;5;241m<\u001b[39m max_iterations:\n\u001b[0;32m--> 346\u001b[0m     final_outputs \u001b[38;5;241m=\u001b[39m \u001b[43m_step\u001b[49m\u001b[43m(\u001b[49m\u001b[43mtime\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43moutput_ta_t\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mnew_states\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    347\u001b[0m     time, output_ta_t \u001b[38;5;241m=\u001b[39m final_outputs[:\u001b[38;5;241m2\u001b[39m]\n\u001b[1;32m    348\u001b[0m     new_states \u001b[38;5;241m=\u001b[39m final_outputs[\u001b[38;5;241m2\u001b[39m:]\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/backend/torch/rnn.py:329\u001b[0m, in \u001b[0;36mrnn.<locals>._step\u001b[0;34m(time, output_ta_t, *states)\u001b[0m\n\u001b[1;32m    327\u001b[0m current_input \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mtuple\u001b[39m(ta[time] \u001b[38;5;28;01mfor\u001b[39;00m ta \u001b[38;5;129;01min\u001b[39;00m input_ta)\n\u001b[1;32m    328\u001b[0m current_input \u001b[38;5;241m=\u001b[39m pack_sequence_as(inputs, current_input)\n\u001b[0;32m--> 329\u001b[0m output, new_states \u001b[38;5;241m=\u001b[39m \u001b[43mstep_function\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    330\u001b[0m \u001b[43m    \u001b[49m\u001b[43mcurrent_input\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43mtuple\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mstates\u001b[49m\u001b[43m)\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m+\u001b[39;49m\u001b[43m \u001b[49m\u001b[38;5;28;43mtuple\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43mconstants\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    331\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    332\u001b[0m flat_new_state \u001b[38;5;241m=\u001b[39m tree\u001b[38;5;241m.\u001b[39mflatten(new_states)\n\u001b[1;32m    334\u001b[0m flat_output \u001b[38;5;241m=\u001b[39m tree\u001b[38;5;241m.\u001b[39mflatten(output)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/layers/rnn/rnn.py:331\u001b[0m, in \u001b[0;36mRNN.inner_loop.<locals>.step\u001b[0;34m(inputs, states)\u001b[0m\n\u001b[1;32m    330\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mstep\u001b[39m(inputs, states):\n\u001b[0;32m--> 331\u001b[0m     output, new_states \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mcell\u001b[49m\u001b[43m(\u001b[49m\u001b[43minputs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mstates\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mcell_kwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    332\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m tree\u001b[38;5;241m.\u001b[39mis_nested(new_states):\n\u001b[1;32m    333\u001b[0m         new_states \u001b[38;5;241m=\u001b[39m [new_states]\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:118\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    116\u001b[0m filtered_tb \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    117\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 118\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    119\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    120\u001b[0m     filtered_tb \u001b[38;5;241m=\u001b[39m _process_traceback_frames(e\u001b[38;5;241m.\u001b[39m__traceback__)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/layers/layer.py:816\u001b[0m, in \u001b[0;36mLayer.__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m    814\u001b[0m         outputs \u001b[38;5;241m=\u001b[39m \u001b[38;5;28msuper\u001b[39m()\u001b[38;5;241m.\u001b[39m\u001b[38;5;21m__call__\u001b[39m(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)\n\u001b[1;32m    815\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m--> 816\u001b[0m     outputs \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__call__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    817\u001b[0m \u001b[38;5;66;03m# Change the layout for the layer output if needed.\u001b[39;00m\n\u001b[1;32m    818\u001b[0m \u001b[38;5;66;03m# This is useful for relayout intermediate tensor in the model\u001b[39;00m\n\u001b[1;32m    819\u001b[0m \u001b[38;5;66;03m# to achieve the optimal performance.\u001b[39;00m\n\u001b[1;32m    820\u001b[0m distribution \u001b[38;5;241m=\u001b[39m distribution_lib\u001b[38;5;241m.\u001b[39mdistribution()\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/torch/nn/modules/module.py:1511\u001b[0m, in \u001b[0;36mModule._wrapped_call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1509\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_compiled_call_impl(\u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[1;32m   1510\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1511\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_call_impl\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/torch/nn/modules/module.py:1520\u001b[0m, in \u001b[0;36mModule._call_impl\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m   1515\u001b[0m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[1;32m   1516\u001b[0m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[1;32m   1517\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_forward_pre_hooks\n\u001b[1;32m   1518\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[1;32m   1519\u001b[0m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[0;32m-> 1520\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mforward_call\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m   1522\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[1;32m   1523\u001b[0m     result \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/backend/torch/layer.py:26\u001b[0m, in \u001b[0;36mTorchLayer.forward\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m     25\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mforward\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;241m*\u001b[39margs, \u001b[38;5;241m*\u001b[39m\u001b[38;5;241m*\u001b[39mkwargs):\n\u001b[0;32m---> 26\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mOperation\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[38;5;21;43m__call__\u001b[39;49m\u001b[43m(\u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:118\u001b[0m, in \u001b[0;36mfilter_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    116\u001b[0m filtered_tb \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    117\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 118\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    119\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    120\u001b[0m     filtered_tb \u001b[38;5;241m=\u001b[39m _process_traceback_frames(e\u001b[38;5;241m.\u001b[39m__traceback__)\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/ops/operation.py:42\u001b[0m, in \u001b[0;36mOperation.__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m     37\u001b[0m         call_fn \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcall\n\u001b[1;32m     38\u001b[0m     call_fn \u001b[38;5;241m=\u001b[39m traceback_utils\u001b[38;5;241m.\u001b[39minject_argument_info_in_traceback(\n\u001b[1;32m     39\u001b[0m         call_fn,\n\u001b[1;32m     40\u001b[0m         object_name\u001b[38;5;241m=\u001b[39m(\u001b[38;5;124mf\u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00m\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__class__\u001b[39m\u001b[38;5;241m.\u001b[39m\u001b[38;5;18m__name__\u001b[39m\u001b[38;5;132;01m}\u001b[39;00m\u001b[38;5;124m.call()\u001b[39m\u001b[38;5;124m\"\u001b[39m),\n\u001b[1;32m     41\u001b[0m     )\n\u001b[0;32m---> 42\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mcall_fn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m     44\u001b[0m \u001b[38;5;66;03m# Plain flow.\u001b[39;00m\n\u001b[1;32m     45\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m any_symbolic_tensors(args, kwargs):\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/utils/traceback_utils.py:157\u001b[0m, in \u001b[0;36minject_argument_info_in_traceback.<locals>.error_handler\u001b[0;34m(*args, **kwargs)\u001b[0m\n\u001b[1;32m    155\u001b[0m bound_signature \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[1;32m    156\u001b[0m \u001b[38;5;28;01mtry\u001b[39;00m:\n\u001b[0;32m--> 157\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mfn\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    158\u001b[0m \u001b[38;5;28;01mexcept\u001b[39;00m \u001b[38;5;167;01mException\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m e:\n\u001b[1;32m    159\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mhasattr\u001b[39m(e, \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m_keras_call_info_injected\u001b[39m\u001b[38;5;124m\"\u001b[39m):\n\u001b[1;32m    160\u001b[0m         \u001b[38;5;66;03m# Only inject info for the innermost failing call\u001b[39;00m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/layers/rnn/lstm.py:267\u001b[0m, in \u001b[0;36mLSTMCell.call\u001b[0;34m(self, inputs, states, training)\u001b[0m\n\u001b[1;32m    264\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[1;32m    265\u001b[0m     z \u001b[38;5;241m=\u001b[39m ops\u001b[38;5;241m.\u001b[39mmatmul(inputs, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mkernel)\n\u001b[0;32m--> 267\u001b[0m     z \u001b[38;5;241m+\u001b[39m\u001b[38;5;241m=\u001b[39m \u001b[43mops\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mmatmul\u001b[49m\u001b[43m(\u001b[49m\u001b[43mh_tm1\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mrecurrent_kernel\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    268\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39muse_bias:\n\u001b[1;32m    269\u001b[0m         z \u001b[38;5;241m+\u001b[39m\u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mbias\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/ops/numpy.py:3547\u001b[0m, in \u001b[0;36mmatmul\u001b[0;34m(x1, x2)\u001b[0m\n\u001b[1;32m   3545\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m any_symbolic_tensors((x1, x2)):\n\u001b[1;32m   3546\u001b[0m     \u001b[38;5;28;01mreturn\u001b[39;00m Matmul()\u001b[38;5;241m.\u001b[39msymbolic_call(x1, x2)\n\u001b[0;32m-> 3547\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43mbackend\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mnumpy\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mmatmul\u001b[49m\u001b[43m(\u001b[49m\u001b[43mx1\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mx2\u001b[49m\u001b[43m)\u001b[49m\n",
      "File \u001b[0;32m~/miniforge3/lib/python3.9/site-packages/keras/src/backend/torch/numpy.py:72\u001b[0m, in \u001b[0;36mmatmul\u001b[0;34m(x1, x2)\u001b[0m\n\u001b[1;32m     70\u001b[0m x1 \u001b[38;5;241m=\u001b[39m cast(x1, compute_dtype)\n\u001b[1;32m     71\u001b[0m x2 \u001b[38;5;241m=\u001b[39m cast(x2, compute_dtype)\n\u001b[0;32m---> 72\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m cast(\u001b[43mtorch\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mmatmul\u001b[49m\u001b[43m(\u001b[49m\u001b[43mx1\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mx2\u001b[49m\u001b[43m)\u001b[49m, result_dtype)\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "model.fit(X_train, y_train, \n",
    "          validation_data=(X_valid, y_valid),\n",
    "          batch_size=200,\n",
    "          epochs=15)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(model.history.history['rec@prec'])\n",
    "plt.plot(model.history.history['val_rec@prec'])\n",
    "plt.title('model metrics')\n",
    "plt.ylabel('recall at precision')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Рекурретную сеть можно применить в прямом и обратном порядке и объединить состояния в одно (половина вектора будет про инфу до, половина про инфу после). Это назвается Bidirectional и делается в keras очень просто. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = keras.layers.Input(shape=(MAX_LEN,))\n",
    "embeddings = keras.layers.Embedding(input_dim=len(word2id), output_dim=100)(inputs, )\n",
    "\n",
    "lstm = keras.layers.Bidirectional(keras.layers.LSTM(128, return_sequences=False))(embeddings)\n",
    "\n",
    "outputs = keras.layers.Dense(len(label2id), activation='softmax')(lstm)\n",
    "\n",
    "model = keras.Model(inputs=inputs, outputs=outputs)\n",
    "optimizer = keras.optimizers.Adam()\n",
    "model.compile(optimizer=optimizer,\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=[keras.metrics.RecallAtPrecision(0.8, name='rec@prec')])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Схема задачи, чтобы было понятнее."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![bilstm classification](./image_bilstm_clf.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"functional_5\"</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1mModel: \"functional_5\"\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\"> Layer (type)                    </span>┃<span style=\"font-weight: bold\"> Output Shape           </span>┃<span style=\"font-weight: bold\">       Param # </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ input_layer_3 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)      │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">200</span>)            │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ embedding_3 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Embedding</span>)         │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">200</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">100</span>)       │     <span style=\"color: #00af00; text-decoration-color: #00af00\">2,409,300</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ bidirectional (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Bidirectional</span>)   │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">256</span>)            │       <span style=\"color: #00af00; text-decoration-color: #00af00\">234,496</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_3 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                 │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">19</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">4,883</span> │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape          \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ input_layer_3 (\u001b[38;5;33mInputLayer\u001b[0m)      │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m200\u001b[0m)            │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ embedding_3 (\u001b[38;5;33mEmbedding\u001b[0m)         │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m200\u001b[0m, \u001b[38;5;34m100\u001b[0m)       │     \u001b[38;5;34m2,409,300\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ bidirectional (\u001b[38;5;33mBidirectional\u001b[0m)   │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m)            │       \u001b[38;5;34m234,496\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_3 (\u001b[38;5;33mDense\u001b[0m)                 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m19\u001b[0m)             │         \u001b[38;5;34m4,883\u001b[0m │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">2,648,679</span> (10.10 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m2,648,679\u001b[0m (10.10 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">2,648,679</span> (10.10 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m2,648,679\u001b[0m (10.10 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model.summary()\n",
    "# обратите внимание на размер output в biderectional\n",
    "# он равен units * 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/5\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m124s\u001b[0m 586ms/step - loss: 2.2686 - rec@prec: 0.0062 - val_loss: 1.9336 - val_rec@prec: 0.0162\n",
      "Epoch 2/5\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m125s\u001b[0m 594ms/step - loss: 1.7359 - rec@prec: 0.0495 - val_loss: 1.8858 - val_rec@prec: 0.1041\n",
      "Epoch 3/5\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m237s\u001b[0m 1s/step - loss: 1.7391 - rec@prec: 0.0959 - val_loss: 1.6266 - val_rec@prec: 0.0771\n",
      "Epoch 4/5\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m164s\u001b[0m 778ms/step - loss: 1.4073 - rec@prec: 0.2681 - val_loss: 1.7428 - val_rec@prec: 0.1677\n",
      "Epoch 5/5\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m178s\u001b[0m 842ms/step - loss: 1.3729 - rec@prec: 0.3524 - val_loss: 1.3247 - val_rec@prec: 0.4125\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.src.callbacks.history.History at 0x2abe54c40>"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train, y_train, \n",
    "          validation_data=(X_valid, y_valid),\n",
    "          batch_size=200,\n",
    "         epochs=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAjcAAAHHCAYAAABDUnkqAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjYuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy89olMNAAAACXBIWXMAAA9hAAAPYQGoP6dpAABxpElEQVR4nO3dd1xWdf/H8dfFBhEQUXCggJorR4I7V65KTW3o3XJV2tZs3NrQzDut7oYNK63UrPzZ1LK6XeQeufcWtwJOpjKu6/z+OEnhBAUO18X7+Xjw6JzDuc55Hy7p+nDOd9gMwzAQERERcRFuVgcQERERKUgqbkRERMSlqLgRERERl6LiRkRERFyKihsRERFxKSpuRERExKWouBERERGXouJGREREXIqKGxEREXEpKm5EpEjs378fm83GlClT8v3ahQsXYrPZWLhwYYHnKkqvvvoqNpvN6hgiLk/FjYjINRgzZgwzZ860OoaIXIKKGxGRa3Atxc3LL7/M2bNnCyeQiORQcSMiUsjS0tIA8PDwwMfHx+I0Iq5PxY1ICXG+vceuXbt44IEHCAwMpFy5crzyyisYhsGhQ4fo3r07AQEBhIWF8c4771x0jMTERB566CFCQ0Px8fGhQYMGfPnllxftd+bMGfr160dgYCBBQUH07duXM2fOXDLXjh07uPvuuwkODsbHx4eYmBh++eUXy64xIyODkSNHUr16dby9vQkPD+eFF14gIyMjZx+bzUZaWhpffvklNpsNm81Gv379cmXYtm0b9913H2XKlOHmm2/O9b0Lff311zRp0gQ/Pz/KlClD69atmTt3bs7316xZQ+fOnQkJCcHX15fIyEgGDBhwTT8jkZLAw+oAIlK0evfuTe3atXnjjTf47bff+M9//kNwcDATJkzglltu4c033+Sbb77hueeeo3HjxrRu3RqAs2fP0rZtW/bs2cOTTz5JZGQk33//Pf369ePMmTMMHjwYAMMw6N69O0uXLuXRRx+ldu3azJgxg759+16UZevWrbRs2ZJKlSoxbNgwSpUqxXfffUePHj348ccf6dmzZ5Feo8Ph4I477mDp0qUMHDiQ2rVrs3nzZt577z127dqV8xjqq6++4uGHH6ZJkyYMHDgQgGrVquXKcM8991CjRg3GjBmDYRiXzTpq1CheffVVWrRowWuvvYaXlxd//vknf/zxB506dSIxMZFOnTpRrlw5hg0bRlBQEPv37+enn366pp+NSIlgiEiJMHLkSAMwBg4cmLMtOzvbqFy5smGz2Yw33ngjZ/vp06cNX19fo2/fvjnbxo0bZwDG119/nbMtMzPTaN68ueHv728kJycbhmEYM2fONADjrbfeynWeVq1aGYAxefLknO3t27c36tWrZ5w7dy5nm8PhMFq0aGHUqFEjZ9uCBQsMwFiwYEGhXuNXX31luLm5GUuWLMl13E8//dQAjGXLluVsK1WqVK7XXpjh3nvvvez3ztu9e7fh5uZm9OzZ07Db7bn2dTgchmEYxowZMwzAWL169RWvXUT+psdSIiXMww8/nLPs7u5OTEwMhmHw0EMP5WwPCgqiZs2axMXF5Wz7/fffCQsL4957783Z5unpydNPP01qaiqLFi3K2c/Dw4PHHnss13meeuqpXDlOnTrFH3/8Qa9evUhJSeHEiROcOHGCkydP0rlzZ3bv3s2RI0eK9Bq///57ateuTa1atXLynDhxgltuuQWABQsW5DnDo48+etV9Zs6cicPhYMSIEbi55f7f8fnHV0FBQQD8+uuvZGVl5fn8IiWZHkuJlDBVqlTJtR4YGIiPjw8hISEXbT958mTO+oEDB6hRo8ZFH8K1a9fO+f75/1aoUAF/f/9c+9WsWTPX+p49ezAMg1deeYVXXnnlklkTExOpVKlSPq7OdK3XuHv3brZv3065cuUumyevIiMjr7rP3r17cXNzo06dOpfdp02bNtx1112MGjWK9957j7Zt29KjRw/uu+8+vL2985xHpCRRcSNSwri7u+dpG3DFtiLXy+FwAPDcc8/RuXPnS+5TvXr1azr2tV6jw+GgXr16vPvuu5fcNzw8PM8ZfH1987zvldhsNn744QdWrlzJrFmzmDNnDgMGDOCdd95h5cqVFxWRIqLiRkTyqGrVqmzatAmHw5Hr7s2OHTtyvn/+v7GxsaSmpub64N25c2eu40VFRQHmo60OHToUdvw8qVatGhs3bqR9+/ZXHUm4IEYarlatGg6Hg23bttGwYcMr7tusWTOaNWvG66+/zrRp07j//vuZPn16rkdwImJSmxsRyZPbb7+d+Ph4vv3225xt2dnZfPjhh/j7+9OmTZuc/bKzs/nkk09y9rPb7Xz44Ye5jle+fHnatm3LhAkTOHbs2EXnO378eCFdyeX16tWLI0eO8Nlnn130vbNnz+aMVwNQqlSpy3Zvz6sePXrg5ubGa6+9lnMn67zzd5ROnz590R2084XQP7uni8jfdOdGRPJk4MCBTJgwgX79+rF27VoiIiL44YcfWLZsGePGjaN06dIAdOvWjZYtWzJs2DD2799PnTp1+Omnn0hKSrromOPHj+fmm2+mXr16PPLII0RFRZGQkMCKFSs4fPgwGzduLNJrfPDBB/nuu+949NFHWbBgAS1btsRut7Njxw6+++475syZQ0xMDADR0dHMnz+fd999l4oVKxIZGUnTpk3zdb7q1avz0ksvMXr0aFq1asWdd96Jt7c3q1evpmLFiowdO5Yvv/ySjz/+mJ49e1KtWjVSUlL47LPPCAgI4Pbbby+MH4OI01NxIyJ54uvry8KFCxk2bBhffvklycnJ1KxZk8mTJ+cMYAfg5ubGL7/8wpAhQ/j666+x2WzccccdvPPOO9x00025jlmnTh3WrFnDqFGjmDJlCidPnqR8+fLcdNNNjBgxooiv0Mw+c+ZM3nvvPaZOncqMGTPw8/MjKiqKwYMHc8MNN+Ts++677zJw4MCcKRX69u2b7+IG4LXXXiMyMpIPP/yQl156CT8/P+rXr8+DDz4ImA2KV61axfTp00lISCAwMJAmTZrwzTff5KnRskhJZDMKs8WgiIiISBFTmxsRERFxKSpuRERExKWouBERERGXouJGREREXIqKGxEREXEpKm5ERETEpZS4cW4cDgdHjx6ldOnSBTJ8uoiIiBQ+wzBISUmhYsWKF03ge6ESV9wcPXo0X5PfiYiISPFx6NAhKleufMV9Slxxc36I+EOHDhEQEGBxGhEREcmL5ORkwsPDcz7Hr6TEFTfnH0UFBASouBEREXEyeWlSogbFIiIi4lJU3IiIiIhLUXEjIiIiLqXEtbnJK7vdTlZWltUxnJKnpyfu7u5WxxARkRJKxc0FDMMgPj6eM2fOWB3FqQUFBREWFqaxhEREpMipuLnA+cKmfPny+Pn56cM5nwzDID09ncTERAAqVKhgcSIRESlpVNz8g91uzylsypYta3Ucp+Xr6wtAYmIi5cuX1yMqEREpUmpQ/A/n29j4+flZnMT5nf8Zqt2SiIgUNRU3l6BHUddPP0MREbGKihsRERFxKSpu5CIRERGMGzfO6hgiIiLXRA2KXUTbtm1p2LBhgRQlq1evplSpUtcfSkRExAK6c1NCGIZBdnZ2nvYtV66cGlWLiMi12TMf7Hn7vCksKm5cQL9+/Vi0aBHvv/8+NpsNm83GlClTsNls/O9//yM6Ohpvb2+WLl3K3r176d69O6Ghofj7+9O4cWPmz5+f63gXPpay2Wx8/vnn9OzZEz8/P2rUqMEvv/xSxFcpIiLF3sGV8PVdMLENZGdYFkPFzVUYhkF6ZrYlX4Zh5Cnj+++/T/PmzXnkkUc4duwYx44dIzw8HIBhw4bxxhtvsH37durXr09qaiq33347sbGxrF+/nltvvZVu3bpx8ODBK55j1KhR9OrVi02bNnH77bdz//33c+rUqev++YqIiItwOGD2cHOxYjR4eFsWRW1uruJslp06I+ZYcu5tr3XGz+vqb1FgYCBeXl74+fkRFhYGwI4dOwB47bXX6NixY86+wcHBNGjQIGd99OjRzJgxg19++YUnn3zysufo168f9957LwBjxozhgw8+YNWqVdx6663XdG0iIuJitvwAR9eRavjw3LHbeC/Tjq+XNYO46s6Ni4uJicm1npqaynPPPUft2rUJCgrC39+f7du3X/XOTf369XOWS5UqRUBAQM4UCyIiUsJlppM1ZwQAH2d3x7dMBXw8rSsxdOfmKnw93dn2WmfLzn29Luz19NxzzzFv3jzefvttqlevjq+vL3fffTeZmZlXPI6np2eudZvNhsPhuO58IiLi/NIWjqNU2jEOGyFsrHw/k+6qZ+lgripursJms+Xp0ZDVvLy8sNvtV91v2bJl9OvXj549ewLmnZz9+/cXcjoREXFV504dxn35OAC+8OnHR32a4+1h7ZyCeizlIiIiIvjzzz/Zv38/J06cuOxdlRo1avDTTz+xYcMGNm7cyH333ac7MCIick0cDoP1U57Fhww2cAMPPjSEMqW8rI5VPIqb8ePHExERgY+PD02bNmXVqlV5et306dOx2Wz06NGjcAM6geeeew53d3fq1KlDuXLlLtuG5t1336VMmTK0aNGCbt260blzZxo1alTEaUVExBV8PeMXmiaZnW48bnuDqPKlLU5kshl57W9cSL799lv69OnDp59+StOmTRk3bhzff/89O3fupHz58pd93f79+7n55puJiooiODiYmTNn5ul8ycnJBAYGkpSUREBAQK7vnTt3jn379hEZGYmPj8/1XFaJp5+liIhr+371QcJn9aKZ23YOVupClUemFer5rvT5fSHL79y8++67PPLII/Tv3586derw6aef4ufnx6RJky77Grvdzv3338+oUaOIiooqwrQiIiKyYu9J/vh5Ms3ctpNl86bKPW9aHSkXS4ubzMxM1q5dS4cOHXK2ubm50aFDB1asWHHZ17322muUL1+ehx566KrnyMjIIDk5OdeXiIiIXJu9x1N58quVvOD2DQAeLZ+CoHCLU+VmaXFz4sQJ7HY7oaGhubaHhoYSHx9/ydcsXbqUL774gs8++yxP5xg7diyBgYE5X+dH7hUREZH8OZWWyYApq+mR9RuRbgkYpUKxtXrG6lgXsfyxVH6kpKTw4IMP8tlnnxESEpKn1wwfPpykpKScr0OHDhVyShEREdeTkW1n0FdrSD4ZzxDPGQDY2r8C3v4WJ7uYpQO4hISE4O7uTkJCQq7tCQkJOdMI/NPevXvZv38/3bp1y9l2vhuzh4cHO3fupFq1arle4+3tjbe3dfNbiIiIODvDMBj242ZW7z/NGO+ZlCYdwupBw/usjnZJlt658fLyIjo6mtjY2JxtDoeD2NhYmjdvftH+tWrVYvPmzWzYsCHn64477qBdu3Zs2LBBj5xEREQKwQexe5ix/gg3uB3lXrd55sbOY8DN2sH6LsfyoXeHDh1K3759iYmJoUmTJowbN460tDT69+8PQJ8+fahUqRJjx47Fx8eHG2+8Mdfrg4KCAC7aLiIiItfv5w1HeG/+LgAmV/oZ23E71OwCka0tTnZ5lhc3vXv35vjx44wYMYL4+HgaNmzI7NmzcxoZHzx4EDc3p2oaJCIi4hLW7D/F899vAmBsg+NU2rkE3Dyg02iLk12Z5YP4FTUN4lc09LMUEXFuB06m0fPj5ZxKy+TW2iF8kjYEW+I2aPY43Dq2yPM41SB+UjxEREQwbtw4q2OIiEgxkJSeRf8pqzmVlkm9SoF8UHOzWdj4loE2L1gd76pU3IiIiEiOzGwHj369lrjjaVQI9GHSv27Aa/Ffd2raDjcLnGJOxY2IiIgAZpfvl2duZkXcSUp5uTOpX2PKbRgP6SegbA2IGWB1xDxRceMCJk6cSMWKFXPG/Dmve/fuDBgwgL1799K9e3dCQ0Px9/encePGzJ8/36K0IiJSXH26KI7v1hzGzQYf3deI2j6nYOXH5jc7/QfcPa0NmEcqbq7GMCAzzZqvPLb1vueeezh58iQLFizI2Xbq1Clmz57N/fffT2pqKrfffjuxsbGsX7+eW2+9lW7dunHw4MHC+qmJiIiT+X3zMd6cvQOAkd3q0q5WeZg3EuyZENUWbuhsbcB8sLwreLGXlQ5jKlpz7hePglepq+5WpkwZbrvtNqZNm0b79u0B+OGHHwgJCaFdu3a4ubnRoEGDnP1Hjx7NjBkz+OWXX3jyyScLLb6IiDiH9QdP88y3GwDo1yKCvi0i4OBK2DYTbG7mgH02m5UR80V3blzE/fffz48//khGRgYA33zzDf/6179wc3MjNTWV5557jtq1axMUFIS/vz/bt2/XnRsREeHQqXQembqGjGwHt9Qqzytd64DDAbOHmzvc9CCE1rU2ZD7pzs3VePqZd1CsOncedevWDcMw+O2332jcuDFLlizhvffeA+C5555j3rx5vP3221SvXh1fX1/uvvtuMjMzCyu5iIg4geRzWTz05WpOpGZSu0IAH9x7E+5uNtj4HRxdB16l4ZaXrY6ZbypursZmy9OjIav5+Phw55138s0337Bnzx5q1qxJo0aNAFi2bBn9+vWjZ8+eAKSmprJ//34L04qIiNWy7Q6e+GYduxJSKV/am0n9YvD39oDMdIgdZe7Uaij4l7c26DVQceNC7r//frp27crWrVt54IEHcrbXqFGDn376iW7dumGz2XjllVcu6lklIiIlh2EYjPxlK0t2n8DX050v+jamQqCv+c0VH0HyEQisYo5G7ITU5saF3HLLLQQHB7Nz507uu+/vaejfffddypQpQ4sWLejWrRudO3fOuasjIiIlzxdL9/HNnwex2eD9fzWkXuVA8xvJx2Cp2aSBjq+Cp3NOn6M7Ny7Ezc2No0cvbh8UERHBH3/8kWvbE088kWtdj6lEREqGuVvjef337QC8dHttOtUN+/ubf4w2ewmHN4W6d1qU8Prpzo2IiEgJseVIEoOnb8Aw4L6mVXjo5si/v3l0A2yYZi47WdfvC6m4ERERKQGOJZ3loS9XczbLTqsaIYy6oy628wWMYcCclwAD6t0DlWMszXq9VNyIiIi4uLSMbAZMWUNCcgY3hPoz/v5GeLr/owTY8SscWAoePtB+pHVBC4iKGxERERdmdxg8/X/r2X4smRB/L77o25gAn3/MEZWdAXNfMZdbPAVB4dYELUAqbi7ByOOcTnJ5+hmKiBQP//ltG7E7EvH2cOOzPjGEB18wQOyqz+D0PvAPhZZDLMlY0FTc/IOnp1nJpqenW5zE+Z3/GZ7/mYqISNGbumI/k5ftB+DdXg25qUqZ3DuknYRFb5nLt7wC3v5FG7CQqCv4P7i7uxMUFERiYiIAfn5+fze2kjwxDIP09HQSExMJCgrC3d3d6kgiIiXSgh2JvPrLVgCe71yTLvUrXLzTwrGQkQRh9aDhfRd/30mpuLlAWJjZ3/98gSPXJigoKOdnKSIiRWv7sWSenLYOhwH3RFfm8bbVLt7p+E5YM8lc7jwG3Fznj1EVNxew2WxUqFCB8uXLk5WVZXUcp+Tp6ak7NiIiFklMPsdDU1aTlmmneVRZXu9Z79JPIea+DIYdanaByNZFH7QQqbi5DHd3d31Ai4iIU0nPzObhqWs4mnSOqHKl+PSBaLw8LtG8dk8s7J4Lbp7QaXTRBy1kalAsIiLiAhwOg2e+3cCmw0mU8fNkcr/GBPpdolOHPfuvAfuAJgOh7CUeWTk5FTciIiIu4M3ZO5izNQEvdzcm9omhatlSl95x/VQ4vh18y0Cb54s2ZBFRcSMiIuLk/m/VQSYsjgPgv/fUp3FE8KV3PJcEf7xuLrcdbhY4LkjFjYiIiBNbuvsEL8/cAsCQDjXo3rDS5Xde8g6kn4CQGyBmQBElLHoqbkRERJzU7oQUHvtmLXaHQc+bKjG4fY3L73x6P6z8xFzu9B9wd91BVlXciIiIOKETqRn0n7KalHPZNI4owxt3XabL93nzRoI9E6LaQo1ORZbTCipuREREnMy5LDuPTF3D4dNnqVrWjwkPxuDtcYXhSw6sgG0zweZmDtjn4qPvq7gRERFxIg6HwXPfb2T9wTME+noyqV9jgkt5XekFMGe4udyoD4TWLZqgFlJxIyIi4kTem7+LXzcdw8PNxqcPRFOt3FUmu9z8PRxdD16lod1LRRPSYipuREREnMQPaw/z4R97ABh7Zz2aVyt75RdkpkPsKHO51VDwL1/ICYsHFTciIiJOYGXcSYb/tAmAJ9pV456Y8Ku/aPmHkHwEgqpAs8cLOWHxoeJGRESkmIs7nsqgr9aSZTfoUq8Cz3asefUXJR+DZePM5Q6jwNOnUDMWJypuREREirHTaZkMmLKapLNZNAwP4p1eDXBzy0Nvpz9GQ1Y6hDeFuj0LP2gxouJGRESkmMrItjPoq7XsP5lO5TK+fNYnBh/PK3T5Pu/oBtgwzVzuPNblu35fSMWNiIhIMWQYBsN/3Myq/aco7e3BpH6NKVfaOy8vhDkvAgbU6wWVows9a3Gj4kZERKQY+vCPPfy0/gjubjY+fqARN4SWztsLd/wKB5aBhw+0H1G4IYspFTciIiLFzM8bjvDuvF0AjO5+I61qlMvbC7MzYO4r5nKLpyAoDz2qXJCKGxERkWJk7YFTPP+D2eX7kVaR3Ne0St5fvGoinN4H/mHQckjhBHQCKm5ERESKiYMn03lk6loysx10qhPKsNtq5/3FaSdh0X/N5favgPdVRi52YSpuREREioGk9Cz6T1nFqbRM6lUKZNy/GuKely7f5y0cCxlJEFYPGtxbeEGdgIobERERi2XZHTz2zVr2Hk+jQqAPn/eNwc/LI+8HSNwBayaZy53Hglseuou7MBU3IiIiFjIMg5dnbGH53pOU8nLni76NCQ3I52jCc18Gww61ukJkq8IJ6kRU3IiIiFhowuI4vl1zCDcbfHjfTdSpGJC/A+yZD3vmgZsndHytcEI6GRU3IiIiFvnf5mO88b8dAIzoWodbaoXm7wD2bJjzsrncZCCUrVbACZ2TihsRERELbDh0hiHfbgCgX4sI+rWMzP9B1n0Jx7eDbzC0eb5gAzoxFTciIiJF7PDpdB7+cg0Z2Q7a1SzHy13y0eX7vHNJsGCMudx2OPiWKdiQTkzFjYiISBFKOZfFQ1PWcCI1g1phpfnwvkZ4uF/Dx/GSdyD9BITcADH9Cz6oE1NxIyIiUkSy7Q6enLaenQkplC/tzaR+jfH3zkeX7/NO7YOVn5jLnf4D7p4FG9TJqbgREREpAoZh8OqsrSzadRxfT7PLd8Ug32s72PyRYM+EqHZQo1PBBnUBKm5ERESKwKRl+/l65UFsNhj3r4bUqxx4bQc6sAK2/Qw2N+j8OtjyMYpxCaHiRkREpJDN35bAf37bBsCLt9Wmc92wazuQwwFzhpvLjfpAaN0CSuhaVNyIiIgUoi1Hknh6+noMA+5tUoWHW11Dl+/zNn8HR9eDV2lo91LBhXQxKm5EREQKSXzSOR76cjXpmXZa1Qjhte51sV3rY6TMdJg/ylxuNRT8yxdcUBej4kZERKQQpGVk89CXq0lIzqBGeX/G398Iz2vp8n3e8g8h5SgEVYFmjxdcUBek4kZERKSA2R0Gg6evZ+vRZEL8vZjUrzEBPtfRXTv5KCwbZy53GAWe+ZxYs4RRcSMiIlLAXv9tO/O3J+Lt4cbEPjGEB/td3wFjR0NWOoQ3hbo9CyakC1NxIyIiUoC+WrGfScv2AfBOrwY0qnKd0yIcXQ8bp5nLnceq63ceqLgREREpIAt3JvLqLLPL9/Oda9K1fsXrO6BhwJy/ekXV6wWVo68zYcmg4kZERKQA7IhP5slp67E7DO6Orszjbatd/0G3z4IDy8DDBzqMvP7jlRAqbkRERK5TYso5HpqyhtSMbJpFBTOmZ71r7/J9XnYGzBthLrd4CgIrX3/QEkLFjYiIyHU4m2nnkS/XcOTMWaJCSvHpA9F4eRTAx+uqiXB6H/iHQcsh13+8EkTFjYiIyDVyOAye+XYDGw8nUcbPk0n9GhPk53X9B047AYv+ay63fwW8/a//mCWIihsREZFr9OacHczeGo+XuxsTHowhIqRUwRx44VjISIKw+tDgvoI5Zgmi4kZEROQaTF91kAmL4gB48+56NIkMLpgDJ+6ANZPN5c5jwE0f1fmln5iIiEg+LdtzgpdnbgFgcPsa9LypABv7zn0ZDDvU6gqRrQruuCWIihsREZF82JOYwqNfryXbYdC9YUWGdKhRgAefD3vmgZsndHyt4I5bwqi4ERERyaMTqRn0n7KalHPZxFQtw5t31b/+Lt/n2bNhzsvmcpOBULYAxskpoVTciIiI5MG5LDsDp67h0KmzVAn2Y8KD0fh4uhfcCdZ9Cce3g28wtHm+4I5bAhWL4mb8+PFERETg4+ND06ZNWbVq1WX3/emnn4iJiSEoKIhSpUrRsGFDvvrqqyJMKyIiJY3DYfD8D5tYd/AMAT4eTOrXmLL+3gV3gnNJsGCMudx2OPhe53xUJZzlxc23337L0KFDGTlyJOvWraNBgwZ07tyZxMTES+4fHBzMSy+9xIoVK9i0aRP9+/enf//+zJkzp4iTi4hISTFu/i5mbTyKh5uNTx+Ipnr5Ah53ZvHbkH4CQm6AmP4Fe+wSyGYYhmFlgKZNm9K4cWM++ugjABwOB+Hh4Tz11FMMGzYsT8do1KgRXbp0YfTo0VfdNzk5mcDAQJKSkggICLiu7CIi4vp+XHuYZ7/fCMBbd9enV0x4wZ7g1D4Y3wTsmXDfd3BD54I9vovIz+e3pXduMjMzWbt2LR06dMjZ5ubmRocOHVixYsVVX28YBrGxsezcuZPWrVsXZlQRESmB/ow7ybCfNgHwWNtqBV/YAMwfaRY2Ue2gRqeCP34J5GHlyU+cOIHdbic0NDTX9tDQUHbs2HHZ1yUlJVGpUiUyMjJwd3fn448/pmPHjpfcNyMjg4yMjJz15OTkggkvIiIubd+JNAZ9vZYsu8Ht9cJ4vlPNgj/JgeWw7WewuUHn16Ggel6VcJYWN9eqdOnSbNiwgdTUVGJjYxk6dChRUVG0bdv2on3Hjh3LqFGjij6kiIg4rdNpmQyYspoz6Vk0CA/i3V4NcXMr4MLD4YA5L5rLjfpAaN2CPX4JZmlxExISgru7OwkJCbm2JyQkEBYWdtnXubm5Ub16dQAaNmzI9u3bGTt27CWLm+HDhzN06NCc9eTkZMLDC+G2ooiIuISMbDuDvl7LvhNpVAry5fM+MQXb5fu8zd/B0fXgVRravVTwxy/BLG1z4+XlRXR0NLGxsTnbHA4HsbGxNG/ePM/HcTgcuR49/ZO3tzcBAQG5vkRERC7FMAyG/7SZVftOUdrb7PJdrnQBdvk+LzMN5v/1VKH1s+BfvuDPUYJZ/lhq6NCh9O3bl5iYGJo0acK4ceNIS0ujf3+zK1yfPn2oVKkSY8eOBczHTDExMVSrVo2MjAx+//13vvrqKz755BMrL0NERFzAR3/s4ad1R3B3szH+/kbUDCtdOCda/iGkHIWgKtD0scI5RwlmeXHTu3dvjh8/zogRI4iPj6dhw4bMnj07p5HxwYMHcfvHjKhpaWk8/vjjHD58GF9fX2rVqsXXX39N7969rboEERFxAb9sPMo783YB8Fr3urS+oVzhnCj5KCx731zuMAo8fQrnPCWY5ePcFDWNcyMiIhdae+AU9372J5nZDh6+OZKXu9YpvJPNeAw2ToPwZjBgtnpI5ZHTjHMjIiJitYMn0xk4dS2Z2Q461gll+O21C+9kR9ebhQ1A5zEqbAqJihsRESmxks5m0X/KKk6mZXJjpQDe/1dD3Au6y/d5hgFz/uoVVa8XVI4unPOIihsRESmZsuwOHv9mLXuPpxEW4MMXfRvj51WITVG3z4IDy8DDFzqMLLzziIobEREpeQzD4JWZW1i25yR+Xu580S+G0IBCbNibnQHzXjGXWzwFgZUL71yi4kZEREqeiYvjmL76EG42+Oi+m6hbMbBwT/jnBDi9H/zDoOXgwj2XqLgREZGSZfaWY7wx25y/8JWudbilVuhVXnGd0k7A4v+ay+1fAW//wj2fqLgREZGSY+OhMwz5dgOGAX2bV6V/y8jCP+nCsZCRDGH1ocF9hX8+UXEjIiIlw5EzZ3l46hrOZTloW7McrxTmWDbnJe6ANZPN5c5jwE0fu0VBP2UREXF5KeeyeGjKao6nZFArrDQf3dcID/ci+Aic+zIYdqjVFSJbFf75BFBxIyIiLi7b7uDJaevZEZ9CudLefNGvMf7eRTD70J75sGceuHlCx9cK/3yS45re3TNnzrBq1SoSExNxOBy5vtenT58CCSYiInK9DMNg1KxtLNp1HB9PN77oG0OlIN/CP7E9++8B+5oOgrLVCv+ckiPfxc2sWbO4//77SU1NJSAgANs/ho622WwqbkREpNiYvGw/X608gM0G43rfRP3KQUVz4nVT4PgO8A2G1s8XzTklR74fSz377LMMGDCA1NRUzpw5w+nTp3O+Tp06VRgZRURE8m3+tgRG/7YNgOG31eLWG8OK5sTnkmDBGHO57XDwDSqa80qOfBc3R44c4emnn8bPz68w8oiIiFy3LUeSeHr6egwD7m0SziOtooru5IvfhvSTEHIDxPQvuvNKjnwXN507d2bNmjWFkUVEROS6xSed4+Ev15Ceaefm6iG81v3GXE0oCtWpffDnp+Zyp9fB3bNoziu55LvNTZcuXXj++efZtm0b9erVw9Mz9xt3xx13FFg4ERGR/EjLyOahL1cTn3yOGuX9GX9/IzyLosv3efNHgj0TotpBjY5Fd17JxWYYhpGfF7hdYQAim82G3W6/7lCFKTk5mcDAQJKSkggICLA6joiIFBC7w2DQV2uZvz2BsqW8mPlES8KDi7AJxYHlMPk2sLnBo8sgtAgGCSxB8vP5ne87Nxd2/RYRESkOxvy+nfnbE/DycGNin5iiLWwcDpg93Fxu1FeFjcU0iJ+IiDi9r1Ye4Iul+wB4554GRFctU7QBNn0LxzaAV2lo91LRnlsuck3FzaJFi+jWrRvVq1enevXq3HHHHSxZsqSgs4mIiFzVol3HefWXrQA81+kGujWoWLQBMtMg9q8RiFs/C/7livb8cpF8Fzdff/01HTp0wM/Pj6effpqnn34aX19f2rdvz7Rp0wojo4iIyCXtjE/hiW/WYXcY3NWoMk+0q170IZZ/CClHIagKNH2s6M8vF8l3g+LatWszcOBAnnnmmVzb3333XT777DO2b99eoAELmhoUi4i4hsSUc/Qcv5wjZ87SNDKYrx5qipdHEbe2SD4KH0ZDVjrcMwXq9iza85cg+fn8zve/gri4OLp163bR9jvuuIN9+/bl93AiIiL5djbTziNT13LkzFkiQ0ox4cHooi9sAGJHm4VNeDOo06Pozy+XlO9/CeHh4cTGxl60ff78+YSHhxdIKBERkctxOAyGfreBjYfOEOTnyaR+jQny8yr6IEfXw8a/mmPcOgaKaqBAuap8dwV/9tlnefrpp9mwYQMtWrQAYNmyZUyZMoX333+/wAOKiIj803/n7uR/W+LxdLcx8cEYIkNKFX0Iw4DZL5rL9XtDpeiizyCXle/i5rHHHiMsLIx33nmH7777DjDb4Xz77bd07969wAOKiIic993qQ3yycC8Ab91dnyaRwdYE2f4LHFwOHr7QfoQ1GeSy8l3cAPTs2ZOePdVoSkREis7yPSd4ccZmAJ5uX4OeN1W2Jkh2Bsz7q6Bp8RQEWpRDLkuD+ImISLG3JzGVR79eS7bD4I4GFXmmQw3rwvw5AU7vB/8waDnYuhxyWXm6cxMcHMyuXbsICQmhTJkyV5xd9dSpUwUWTkRE5GRqBgOmrCb5XDbRVcvw1t31i26W7wulnYDF/zWX248Ab39rcsgV5am4ee+99yhdunTOsmX/qEREpEQ5l2Vn4FdrOXgqnSrBfkx8MBofT3frAi0cCxnJEFYfGtxrXQ65onwP4ufsNIifiIhzMAyDwdM38MvGo5T28WDG4y2pXt7COyWJO+CTFmDYod9vEHGzdVlKoEIdxG/dunVs3rw5Z/3nn3+mR48evPjii2RmZuY/rYiIyCW8N383v2w8ioebjQkPRFtb2ADMfcksbGp1VWFTzOW7uBk0aBC7du0CzNGKe/fujZ+fH99//z0vvPBCgQcUEZGSZ8b6w3wQuxuA13veSIvqIdYG2j0f9swHN0/o+Jq1WeSq8l3c7Nq1i4YNGwLw/fff06ZNG6ZNm8aUKVP48ccfCzqfiIiUMKv2neLfP5hPCB5tU43ejatYG8iebd61AWg6CMpWszaPXFW+ixvDMHA4HIA55cLtt98OmNMynDhxomDTiYhIibL/RBoDv1pDpt3BbTeG8ULnmlZHgnVT4PgO8A2G1s9bnUbyIN/FTUxMDP/5z3/46quvWLRoEV26dAFg3759hIaGFnhAEREpGQ6eTKfv5FWcSc+iQXgQ7/ZqiJubxb1zz56BBWPM5XYvgm+QlWkkj/Jd3IwbN45169bx5JNP8tJLL1G9enUAfvjhh5y5pkRERPJj/cHT9Px4GQdOplO5jC+f9YnG18vCLt/nLXkH0k9CyA0Q3c/qNJJHBdYV/Ny5c7i7u+Pp6VkQhys06gouIlK8zN0az9PT13Muy0HdigFM6teY0AAfq2PBqX0wvgnYM+G+7+GGTlYnKtHy8/l9TXNLXYqPTzH4hygiIk7ly+X7eXXWVgwD2txQjvH3N8Lfu8A+mq7PvBFmYVPtFqjR0eo0kg+afkFERIqcw2Ew9n/b+WzJPgDubRLO6O434uFeTKY8PLDcnPnb5gadXgeNzO9UNP2CiIgUqXNZdoZ+t4HfN8cD8Hznmjzetlrx+WxxOGD2cHO5UV8IrWNtHsk3Tb8gIiJF5lRaJo9MXcPaA6fxdLfx9j0N6N6wktWxctvwfzDzUfAOgKfWgX85qxMJhdzm5vfff8fd3Z3OnTvn2j537lzsdju33XZbfg8pIiIlwIGTafSbvJp9J9II8PFgwoMxNK9W1upYuWWmQewoc7nVsypsnFS+H24OGzYMu91+0XaHw8GwYcMKJJSIiLiWdQdP0/Pj5ew7kUalIF9+fKxF8StsAJZ/CCnHIKgKNH3U6jRyjfJ952b37t3UqXPx88datWqxZ8+eAgklIiKuY/aWeAZPX09GtoMbK5ldvcuXLoY9bJOPwrL3zeWOr4FnMcwoeZLvOzeBgYHExcVdtH3Pnj2UKlWqQEKJiIhrmLR0H499s5aMbAe31CrPtwObF8/CBiD2NchKhyrNoU4Pq9PIdch3cdO9e3eGDBnC3r17c7bt2bOHZ599ljvuuKNAw4mIiHOyOwxem7WN137dhmHA/U2rMPHBaEoVlzFsLnRkHWz8P3O5s7p+O7t8FzdvvfUWpUqVolatWkRGRhIZGUnt2rUpW7Ysb7/9dmFkFBERJ3Iuy84T36xj0jJzDJtht9XiPz2K0Rg2FzIMmPPXrN/1e0OlaGvzyHXLdwkdGBjI8uXLmTdvHhs3bsTX15f69evTunXrwsgnIiJO5GRqBg9PXcP6g2fwcnfj7V4NuKNBRatjXdn2X+DgcvDwhfYjrU4jBeCa7g/abDY6depE69at8fb2Lj4DL4mIiGX2nUij3+RVHDiZTqCvJxMfjKZpVDHsEfVP2RnmNAsALZ+GwGI25o5ck3zfI3Q4HIwePZpKlSrh7+/Pvn3mbcdXXnmFL774osADiohI8bf2wCnu/Mes3j8+1qL4FzYAf06A0/vBPwxaPG11Gikg+S5u/vOf/zBlyhTeeustvLy8crbfeOONfP755wUaTkREir//bT7GvZ/9yen0LOpXDmTG4y2pXt7f6lhXl3YCFv/XXG4/ArydILPkSb6Lm6lTpzJx4kTuv/9+3N3dc7Y3aNCAHTt2FGg4EREpvgzD4PMlcTw+bR2Z2Q7a1yrP9IHNKFfa2+poebNgDGQkQ4UG0OBeq9NIAcp3m5sjR45QvXr1i7Y7HA6ysrIKJJSIiBRvdofB6F+3MWX5fgAeaFaFV7vVLb49oi6UuB3WTjaXO48BNyfJLXmS7+KmTp06LFmyhKpVq+ba/sMPP3DTTTcVWDARESmezmbaGTx9PXO3JQAw/LZaDGwd5VydS+a+DIYDanWFiJutTiMFLN/FzYgRI+jbty9HjhzB4XDw008/sXPnTqZOncqvv/5aGBlFRKSYOJGawcNfrmHDIbOr9zu9GtCtuHf1vtDu+bBnPrh5mtMsiMu5phGKZ82axfz58ylVqhQjRoxg+/btzJo1i44dOxZGRhERKQbijqdy58fL2XDoDIG+nnz9cFPnK2zs2TD3rwH7mg6CstWszSOFIl93brKzsxkzZgwDBgxg3rx5hZVJRESKmTX7T/Hw1DWcSc8iPNiXKf2bUK2cE/YuWjcFju8A32Bo/bzVaaSQ5OvOjYeHB2+99RbZ2dmFlUdERIqZ3zYd477P/+RMehYNKgfy02MtnbOwOXvG7CEF0O5F8A2yMo0Uonw/lmrfvj2LFi0qjCwiIlKMGIbBZ4vjeOKvrt4daofyf87U1ftCS96G9JMQUhOi+1udRgpRvhsU33bbbQwbNozNmzcTHR1NqVKlcn1fM4OLiDg/c1bvrXy54gAAfZtXZUS3uri7OVGPqH86FWeORgzQ6T/gXkxnJ5cCYTMMw8jPC9yuMBaAzWbDbrdfd6jClJycTGBgIElJSQQEBFgdR0Sk2EnPzObp/9vA/O1mV++Xu9TmoZsjnaur94W+fdCcILPaLfDAT+DM11JC5efzO9+lq8PhuOZgIiJSvB1PyeDhL1ez8XASXh5ujOvdkNvrVbA61vXZv8wsbGxu0Ol1FTYlgO7LiYgIAHuPp9Jv8ioOnTpLGT9PPusTQ0xEsNWxro/DAXNeNJej+0FoHUvjSNG4pvGmY2Nj6dq1K9WqVaNatWp07dqV+fPnF3Q2EREpIqv2neLOj5dz6NRZqgT78eNjLZy/sAHY9C0c2wDeAdD2RavTSBHJd3Hz8ccfc+utt1K6dGkGDx7M4MGDCQgI4Pbbb2f8+PGFkVFERArRrI1HeeDzP0k6m0XD8CBmPN6CKGfs6n2hzDSIHWUut3oW/MtZm0eKTL4bFFeuXJlhw4bx5JNP5to+fvx4xowZw5EjRwo0YEFTg2IREZNhGExYHMcb/9sBQOe6oYzrfRO+Xu4WJysgC8bCojcgqCo8sQo8faxOJNchP5/f+b5zc+bMGW699daLtnfq1ImkpKT8Hk5ERCyQbXfwys9bcgqb/i0j+Pj+aNcpbJKPwrL3zeWOo1TYlDD5Lm7uuOMOZsyYcdH2n3/+ma5duxZIKBERKTzpmdkM+motX688iM0Gr3Stw0hnHsPmUmJfg+yzUKU51OlhdRopYvnuLVWnTh1ef/11Fi5cSPPmzQFYuXIly5Yt49lnn+WDDz7I2ffpp58uuKQiInLdElPO8fCXa9h0OAnvv7p63+bsXb0vdGQdbPw/c7mzun6XRPlucxMZGZm3A9tsxMXFXVOowqQ2NyJSUu1JTKHf5NUcPm129f68b2Oiq5axOlbBMgyYfBscXAH1/wV3TrA6kRSQQh3Eb9++fdcc7HLGjx/Pf//7X+Lj42nQoAEffvghTZo0ueS+n332GVOnTmXLli0AREdHM2bMmMvuLyIi8GfcSR6Zuobkc9lElPVjcv8mRIaUuvoLnc32X8zCxsMX2o+wOo1Y5JrGuSlI3377LUOHDmXkyJGsW7eOBg0a0LlzZxITEy+5/8KFC7n33ntZsGABK1asIDw8nE6dOhX7XloiIlb5ecMRHvxiFcnnsrmpShA/PtbCNQub7AyY91dB0/JpCKxkbR6xTL4fSxW0pk2b0rhxYz766CPAnN4hPDycp556imHDhl319Xa7nTJlyvDRRx/Rp0+fq+6vx1IiUlIYhsGni+J4c7bZI+rWumGM+1dDfDxdpEfUhZa9bxY3/mHw9DrwcsECrgQr1MdSBSkzM5O1a9cyfPjwnG1ubm506NCBFStW5OkY6enpZGVlERx86ZE0MzIyyMjIyFlPTk6+vtAiIk4g2+5gxC9bmfbnQQAeujmSF2+v7Vo9ov4p7QQsfttcbj9ChU0JZ+ljqRMnTmC32wkNDc21PTQ0lPj4+Dwd49///jcVK1akQ4cOl/z+2LFjCQwMzPkKDw+/7twiIsVZWkY2j0xdw7Q/za7eI7vV4ZWudVy3sAFYMAYykqFCA2hwr9VpxGL5Lm4OHjzIpZ5kGYbBwYMHCyRUXr3xxhtMnz6dGTNm4ONz6QGahg8fTlJSUs7XoUOHijSjiEhRSkw+R++JK1iw8zjeHm58cn80/VvmrZer00rcDmsnm8udx4Kb5c1JxWL5fiwVGRnJsWPHKF++fK7tp06dIjIyErvdnudjhYSE4O7uTkJCQq7tCQkJhIWFXfG1b7/9Nm+88Qbz58+nfv36l93P29sbb2/vPGcSEXFWuxPMrt5HzpwluJQXn/eNoVEVF+vqfSlzXgLDAbW7QURLq9NIMZDv8tYwDGyXGBApNTX1sndPLsfLy4vo6GhiY2NztjkcDmJjY3MGCLyUt956i9GjRzN79mxiYmLydU4REVe0Yu9J7vxkOUfOnCUypBQzHm9RMgqb3fNhbyy4eUKHUVankWIiz3duhg4dCpiD873yyiv4+fnlfM9ut/Pnn3/SsGHDfAcYOnQoffv2JSYmhiZNmjBu3DjS0tLo378/AH369KFSpUqMHTsWgDfffJMRI0Ywbdo0IiIictrm+Pv74+/vArPYiojk08z1R3j+h41k2Q2iq5bhsz4xBJfysjpW4bNnw9yXzOWmg6BsNWvzSLGR5+Jm/fr1gHnnZvPmzXh5/f2L4+XlRYMGDXjuuefyHaB3794cP36cESNGEB8fT8OGDZk9e3ZOI+ODBw/i9o/np5988gmZmZncfffduY4zcuRIXn311XyfX0TEWRmGwccL9/LfOTsBuL1eGO/2cuGu3hdaOxmO7wDfYGj9vNVppBjJ9zg3/fv35/3333faMWI0zo2IuILzs3r/3yqzk8TA1lEMu7UWbq7cI+qfzp6BDxtB+km4/W1o8ojViaSQFeo4N5MnT77mYCIicv1SM7J54pt1LNp1HDcbjOxWl74tIqyOVbSWvG0WNiE1Ibq/1WmkmLmmQfzWrFnDd999x8GDB8nMzMz1vZ9++qlAgomIyMUSks/Rf/Jqth1LxsfTjQ/vbUTHOqFXf6ErORUHKz81lzu/Du6WjkcrxVC+e0tNnz6dFi1asH37dmbMmEFWVhZbt27ljz/+IDAwsDAyiogIsCshhZ7jl7HtWDIh/l5MH9i85BU2YE6x4MiCau2hRker00gxlO/iZsyYMbz33nvMmjULLy8v3n//fXbs2EGvXr2oUqVKYWQUESnxlu85wV2fLOdo0jmiQkrx02MtaRgeZHWsord/GWyfBTY36PQfq9NIMZXv4mbv3r106dIFMHtJpaWlYbPZeOaZZ5g4cWKBBxQRKelmrD9M38mrSDmXTeOIMvz4WAuqlPW7+gtdjcMBc140l6P7QWgdS+NI8ZXv4qZMmTKkpKQAUKlSJbZs2QLAmTNnSE9PL9h0IiIlmGEYfBi7m2e+Ncew6VK/Al891JQyJWEMm0vZNB2ObQDvAGj7otVppBjLdyus1q1bM2/ePOrVq8c999zD4MGD+eOPP5g3bx7t27cvjIwiIiVOlt3ByzO28O0as6v3oNZR/LskdfW+UGYaxL5mLrd6FvzLWZtHirV8FzcfffQR586dA+Cll17C09OT5cuXc9ddd/Hyyy8XeEARkZIm5VwWT0xbz+K/unqPuqMuDzaPsDqWtZZ9ACnHIKgqNHvM6jRSzOW7uAkODs5ZdnNzY9iwYQUaSESkJItPOkf/KavZfiwZX093Prz3JjqUxB5R/5R0BJa9by53fA08NBmyXJkGBxARKSZ2xCfTf/JqjiWdI8Tfi0n9GlO/cpDVsawX+xpkn4UqzaFOd6vTiBNQcSMiUgws23OCR79aS0pGNtXKlWJK/yaEB5fAHlEXOrLObEgM5oB9thLa5kjyRcWNiIjFflh7mGE/biLbYdAkMpiJD0YT5FdCe0T9k2H83fW7/r+gUrS1ecRpqLgREbGIYRh8ELuH9+bvAqBbg4q8fU99vD1KyKzeV7PtZzi4Ajx8of0Iq9OIE1FxIyJigSy7g+E/beaHtYcBeKxtNZ7vVLPkdvW+UHaGOc0CQMunIbCStXnEqeSpuLnzzjvzfEBNnCkicmUp57J4/Jt1LNl9AjcbjO5xI/c3rWp1rOLlz0/hzAEoXQFaDrY6jTiZPBU3mhBTRKRgHEs6S//Jq9kRn4Kflzsf3XcTt9Qq4V29L5R6HBa/bS63HwFepazNI04nT8XN5MmTCzuHiIjL237M7Oodn3yOcqW9mdS3MfUq64/HiywcAxnJUKGB2ZBYJJ/U5kZEpAgs2X2cx75eR2pGNtXL+zOlf2Mql1FX74skboe1U8zlzmPBLd9TIIrkrbi56aabsOVxbIF169ZdVyAREVfz3ZpDvPjTZrIdBs2igpnwQAyBfp5Wxyp+MlLh9+fBcEDtbhDR0upE4qTyVNz06NGjkGOIiLgewzAYN38378fuBqB7w4q8dbe6el/S3j/gl8GQdBDcvcxpFkSuUZ6Km5EjRxZ2DhERl5KZbXb1/nGd2dX7iXbVeK5TzTzfBS8xzp6BuS/B+q/N9cAq0GM8BEdZGkucm9rciIgUsORzWTz29VqW7TmJu5uN0d1v5L6mVayOVfzs+B1+G2rO9g3QZJDZO8rb39pc4vTyXdzY7Xbee+89vvvuOw4ePEhmZmau7586darAwomIOJujZ8yu3jsTzK7e4+9vRLua5a2OVbyknYT/vQBbfjDXg6tB94+gagtrc4nLyHcz9FGjRvHuu+/Su3dvkpKSGDp0KHfeeSdubm68+uqrhRBRRMQ5bD2aRM+Pl7EzIYVypb35blBzFTb/ZBiw5ScY38QsbGxu5gB9jy1TYSMFymYYhpGfF1SrVo0PPviALl26ULp0aTZs2JCzbeXKlUybNq2wshaI5ORkAgMDSUpKIiAgwOo4IuIiFu06zuNfryUt006N8v5MVlfv3FLi4bdnYcev5nr5OubdGk2GKXmUn8/vfD+Wio+Pp169egD4+/uTlJQEQNeuXXnllVeuIa6IiHP7dvVBXpyxBbvDoHlUWT59MJpAX3X1Bsy7NRumwZzhcC4J3Dyg1XPQ6lnw0MznUjjyXdxUrlyZY8eOUaVKFapVq8bcuXNp1KgRq1evxtvbuzAyiogUS4Zh8O68XXz4xx4Aet5UiTfvqo+XhwaeA+DMIZg1GPbGmusVGkL38RB2o6WxxPXlu7jp2bMnsbGxNG3alKeeeooHHniAL774goMHD/LMM88URkYRkWInM9vBsB838dP6IwA8dUt1hna8QV29ARwOWDsJ5o2EzFRw94Z2w6H5U+CuTrpS+PLd5uZCK1euZPny5dSoUYNu3boVVK5CozY3InK9ks5m8ehXa1kRZ3b1HtPzRno3VldvAE7uhV+ehgNLzfXwpubdmpAa1uYSp1eobW4u1KxZM5o1a3a9hxERcQpHzpyl/+RV7EpIpZSXOx8/EE2bG8pZHct6Djus/AT++A9knwVPP2g/Epo8Am4akVmKVr6Lm7FjxxIaGsqAAQNybZ80aRLHjx/n3//+d4GFExEpTrYcSWLAlNUkpmQQGuDNpH6NqVtRs3qTuAN+fgKOrDHXI1tDtw8gONLaXFJi5bvV24QJE6hVq9ZF2+vWrcunn35aIKFERIqbBTsT6TVhBYkpGdQMLc2Mx1uqsLFnweL/woRWZmHjHQDd3oc+v6iwEUtdU1fwChUqXLS9XLlyHDt2rEBCiYgUJ/+36iAvzzS7eresXpZPHogmwKeEd/U+tgl+fhziN5vrNTpD1/cgsJK1uUS4huImPDycZcuWERmZuypftmwZFStWLLBgIiJWMwyDt+fuZPyCvQDc1agyY++sV7K7emdnwKK3YNk4cGSDbxm47S2odw+op5gUE/kubh555BGGDBlCVlYWt9xyCwCxsbG88MILPPvsswUeUETEChnZdv79wyZmbjgKwOD2NRjSoUbJ7up9eI3Ztub4DnO9Tne4/W3w1xQTUrzku7h5/vnnOXnyJI8//njOpJk+Pj78+9//Zvjw4QUeUESkqCWlZzHo6zWsjDuFh5uNMXfWo1dMuNWxrJOZDgteh5Ufg+GAUuWhy9tmcSNSDF3zODepqals374dX19fatSo4TSjE2ucGxG5ksOn0+k3eTV7ElPx9/bg4/sb0bokd/XevxR+fhJO7zPXG9wLnceAX7C1uaTEKZJxbuLj4zl16hStW7fG29sbwzBK9u1aEXF6mw8nMeDL1RxPySAswIdJ/RpTp2IJ/SMoI8UcYXjNF+Z6QCXoOg5u6GRpLJG8yHdxc/LkSXr16sWCBQuw2Wzs3r2bqKgoHnroIcqUKcM777xTGDlFRArVgh2JPDFtHemZdmqFlWZy/8ZUCPS1OpY19syHWUMg6ZC5Ht0POr4GPiW867s4jXw3+X/mmWfw9PTk4MGD+Pn55Wzv3bs3s2fPLtBwIiJF4Zs/D/DQl6tJz7Rzc/UQvnu0ecksbM6ehpmPw9d3mYVNUFVzzJpu76uwEaeS7zs3c+fOZc6cOVSuXDnX9ho1anDgwIECCyYiUtgcDoP/zt3JJwvNrt53R5tdvT3dS2BX7+2/wm9DITUBsEHTR6H9K+BVyupkIvmW7+ImLS0t1x2b806dOuU0jYpFRDKy7Tz3/SZmbTS7ej/T4Qaebl+95LUdTDsBvz8PW38y18vWMCe6rNLU2lwi1yHff560atWKqVOn5qzbbDYcDgdvvfUW7dq1K9BwIiKF4Ux6Jg9+sYpZG4/i4Wbj7XsaMLikjWFjGLD5B/iosVnY2Nzh5mfg0aUqbMTp5fvOzVtvvUX79u1Zs2YNmZmZvPDCC2zdupVTp06xbNmywsgoIlJgDp1Kp9/kVew9nkZpbw8+eSCam2uEWB2raCUfMx9B7fzdXA+9Ebp/BBVvsjaXSAHJd3Fz4403smvXLj766CNKly5Namoqd955J0888cQl55wSESkuNh0+w4ApqzmRmkmFQB8m929MrbAS1NXbMGD91zDnJchIAjdPaPMCtBwCHl5WpxMpMPkqbrKysrj11lv59NNPeemllwork4hIgZu/LYGn/m89Z7Ps1K4QwOR+jQkL9LE6VtE5fQBmDYa4BeZ6xUZm25rQOtbmEikE+SpuPD092bRpU2FlEREpFF+t2M/IX7biMKD1DeUYf99NlC4ps3o7HOZAfPNGQlYaePhAu5eg2ePgfs3juIoUa/luUPzAAw/wxRdfFEYWEZEC5XAYjP3fdl752SxseseE80XfmJJT2JzYA1O6wO/PmYVNlRbw6DJo+bQKG3Fp+f7XnZ2dzaRJk5g/fz7R0dGUKpV7DIR33323wMKJiFyrc1l2nvt+I79uOgbAsx1v4MlbSkhXb3s2rBwPC8ZA9jnwLAUdR0HMQ+BWAsfwkRIn38XNli1baNSoEQC7du3K9b0S8T8NESn2zqRn8sjUNazefxpPdxtv3lWfOxtVvvoLXUHCNvj5CTi6zlyPameOMFymqrW5RIpQvoubBQsWFEYOEZECcfBkOv2mrCLueBqlfTyY8EA0LaqXgK7e9ixY+h4segscWeAdCJ1fh5seAP3hKSWMHrqKiEswDIOle07wzLcbOJGaScVAH6YMaMINoaWtjlb4jm6An5+EhM3m+g23Qdf3IEDDc0jJpOJGRJxatt3B/7bEM3FxHJuPJAFQt2IAk/o1JjTAxbt6Z52DRW/CsvfBsINvMNz+X7jxLt2tkRJNxY2IOKX0zGy+W32Iz5fu4/DpswB4e7jRu3E4L9xaC39vF//f28E/4Zcn4cRfbR/r3gm3vQX+5azNJVIMuPhvv4i4muMpGUxdsZ+vVh7gTHoWAMGlvOjTvCoPNqtKWX8Xn8A3Mw3++A+s/AQwwD8UurwLtbtanUyk2FBxIyJOYe/xVD5fso8f1x0mM9sBQNWyfjzcKoq7G1XG18vd4oRFIG4RzHoaTu831xvebzYa9i1jaSyR4kbFjYgUa2v2n2LC4jjmb0/AMMxtDcODGNQ6ik51w3B3KwFtS84lwbwRsHaKuR5Q2ezeXaODpbFEiisVNyJS7DgcBnO3JTBx8V7WHTyTs71D7fIMbF2NxhFlSs64Wrvmwq9DIPmIuR7zEHR4FXxK0ISfIvmk4kZEio1zWXZ+XHeYz5fsY9+JNAC83N3oeVMlHmkdSfXyJaBb93npp2D2cNg03VwvEwl3fAiRrazNJeIEVNyIiOVOp2Xy1coDfLl8PyfTMgEI8PHggWZV6dcigvKu3qX7Qtt+ht+eg7REwAbNnzAnu/TyszqZiFNQcSMiljl0Kp3Pl8Tx3ZrDnM2yA1ApyJcBN0fSu3G463fnvlBqojnJ5bafzfWQmtB9PIQ3tjaXiJMpYf/nEJHiYNPhM0xYHMf/Nh/D8Vcj4ToVAhjUJorb61XA072ETe5oGLDpO5j9bzh7GmzucPMz0OYF8HDxru0ihUDFjYgUCcMwWLjzOBMW72Vl3Kmc7a1vKMeg1lG0qFa25DQS/qekI/DrM7B7jrkeVs+8W1OhgbW5RJyYihsRKVSZ2Q5+3nCEz5bEsSshFQAPNxt3NKjII62jqF2hhPb6MQxY9yXMfQUyksHdy7xT03IIuHtanU7Eqam4EZFCkXwui2l/HmTysn0kJGcA4O/twb1NwunfMpKKQb4WJ7TQ6f3wy9Owb5G5XinGvFtTvpalsURchYobESlQR8+cZdLSfUxffYjUjGwAQgO86d8ykvuaViHApwTflXA4YNVEiB0FWeng4Qu3vAzNHgO3EjDCskgRUXEjIgVi+7FkJi6OY9bGo2T/1Ur4hlB/HmkVRfeGlfDyKGGNhC90Yjf8/CQcWmmuV70Z7vgAylazNpeIC1JxIyLXzDAMlu05yYTFe1my+0TO9mZRwQxqXY22NcuVzEbC/2TPhhUfwoKxYM8AL3/o+BpE9we3El7wiRQSFTcikm9Zdge/bz7GxMVxbD2aDICbDW6rV4FBraOoXznI2oDFRcJWmPk4HNtgrldrb84JFRRuaSwRV6fiRkTyLC0jm+mrDzFp6T6OnDkLgK+nO70bhzOgZSRVymoEXQCyM2HJO+aXIwt8AuHWN6DBvVDS72SJFAHL74mOHz+eiIgIfHx8aNq0KatWrbrsvlu3buWuu+4iIiICm83GuHHjii6ouDZ7Niz/CD5qDJNvN+f02TgdEreb3yvhEpPP8dbsHTQfG8voX7dx5MxZQvy9eLbjDSwfdguv3lFXhc15R9bBxLaw6A2zsKnVFZ5YBQ3vU2EjUkQsvXPz7bffMnToUD799FOaNm3KuHHj6Ny5Mzt37qR8+fIX7Z+enk5UVBT33HMPzzzzjAWJxSUd22h2yz3/6ODELjiw7O/ve/hC2I3moGrnv8rVBg8vS+IWpT2JKXy2eB8z1h8h0+4AICqkFA+3iuLORpXw8VQPnxxZZ2HhWFj+IRgO8AuB2/8LdXuqqBEpYjbDMAyrTt60aVMaN27MRx99BIDD4SA8PJynnnqKYcOGXfG1ERERDBkyhCFDhuTrnMnJyQQGBpKUlERAQAkdPExMmenmh9GK8WDYzUcHt7wCXqXMgufYRji2CbLSLn6tuxeUr/1XsdPQ/AqtA57OP3aLYRis3n+aiYv3Mn97Ys726KplGNg6io61Q3Fz04d1LgdWwC9Pwsk95vqNd8Ntb0KpEGtzibiQ/Hx+W3bnJjMzk7Vr1zJ8+PCcbW5ubnTo0IEVK1ZYFUtKir1/wKwhcOaAuV63J9z6JpQONdcb3mf+12GHU3F/FTob/i56ziX9vcxUc1+bO5SrlfsOT9iN4F26iC/u2tgdBnO2xjNxcRwbDp0BzBsOHWuHMqhNFNFVg60NWBxlpELsa+bYNRjgHwZd34Nat1udTKREs6y4OXHiBHa7ndDQ0FzbQ0ND2bFjR4GdJyMjg4yMjJz15OTkAju2OKG0kzDnRdg03VwPqAxd3oGat156fzd3CKlhftW729xmGGZRlHN3ZyMc3QDpJyBxq/m1cdpfB7BB2eq5C54K9cG3TGFfaZ6dzbTzw9pDfL50HwdOpgPg5eHGXY0q83CrSKqV87c4YTG1dwHMehrOHDTXb3oAOr0OvkGWxhKREtBbauzYsYwaNcrqGGI1w4BN35oNhc+eAmzQdJA5Omx+76zYbFAmwvyq0/3v46cc+7vQOV/0pByFk7vNry0//H2MoKr/KHYamv/1L1cgl5pXJ1MzmLriAF+tPMCptEwzlp8nDzarSp/mEZQrrdmoL+lcEsx9Gdb9dccusAp0GwfV21saS0T+ZllxExISgru7OwkJCbm2JyQkEBYWVmDnGT58OEOHDs1ZT05OJjxcY0yUKKf2mbMuxy0w18vXhTs+hMrRBXcOmw0CKppfNW/7e3tqotlu55+PtM4c+Ptr+y9/71u64gV3eBqYxyvgxqj7T6Tx+dI4vl9zmIxss5Fw5TK+PHxzJL0ah+Pn5fJ/81y7nbPh1yFmIQvQ+BHoMNJpHj2KlBSW/V/My8uL6OhoYmNj6dGjB2A2KI6NjeXJJ58ssPN4e3vj7a2/QEskezasHG+ODJt9Fty9oe2/ocXTRTfrsn95qNHB/Drv7Om/Cp5/PNY6uce8y5NyFHb97+99S5W7uOAJqnpNBc/6g6eZuDiO2VvjOd+NoF6lQAa2juK2G8PwcLd8ZIjiK+0kzB4Gm78z14Oj4I6PIKKltblE5JIs/RNt6NCh9O3bl5iYGJo0acK4ceNIS0ujf//+APTp04dKlSoxduxYwGyEvG3btpzlI0eOsGHDBvz9/alevbpl1yHF0NH18MtTEL/ZXI9sDV3HFY95fHzLQFQb8+u8jBSI35K74Dm+A9KOw5755td5PoEXP9IKrnbJofwdDoM/diQycXEcq/afytnermY5BrauRrOoYE2PcCWGAdtmwm/PmW2qbG7Q/Elo96JL9IwTcVWWdgUH+Oijj/jvf/9LfHw8DRs25IMPPqBp06YAtG3bloiICKZMmQLA/v37iYyMvOgYbdq0YeHChXk6n7qCu7jMNFgwBlZ+bI414hMEncc45wBqWWchYVvuR1qJ28CeefG+Xv4QVi+n6MksX4+Zh0oxYelB9h43u7J7utu4o0ElBraOomaYHqNcVUoC/DYUdvxqrperDd3HF+zjTBHJs/x8flte3BQ1FTcubPd8s21N0l+9V2682xzyvogb6haq7Ew4vj33HZ74LeZjtwucMzzZYVRhpy2K0lHRNGl+CyFRDcFDj2mvyDDM0alnD4NzZ8DNA1o9a37pZydiGRU3V6DixgWlHoc5w2Hz9+Z6YBXo+i7U6GhtrqJizyZ+32ZWLI0led9aahlx1LEdoLTt4oIHN48LBh9sAKE3gpemTgDgzCGzwfD5x4AVGph3a8LqWRpLRFTcXJGKGxdiGLBhGsx9yWyka3ODZo9D2+HgXTLGZtlyJImJi+P4bfMx7A7zV7lWWGkGtoqga3gGXokXNFw+e/rig9jcIOSG3AVPWD3wKUG/Hw4HrJsCc0dAZspfjc+H/dX4XL3HRIoDFTdXoOLGRZzca/6FvW+xuR5WD7p9AJUaWRqrKBiGweLdJ5i4eC/L9pzM2d6yelkGtq5G6xohl24kbBiQdOjiwQfTEi/eF8xGyhf21PJzwVGKT8WZc4vtX2KuV25i3q0pd4O1uUQkFxU3V6DixsnZs8yJCRe9CdnnzEkt2w0379gUVfdui2TZHczaeJSJi+PYEZ8CgLubjS71KjCwdRQ3Vgq8tgMnH8td8BzbCMmHL71vYBVzhOXzd3gqNPh7ygpn47DDn59C7GizzZKnH7QfAU0GmiNTi0ixouLmClTcOLHDa83h7hO2mOtRbc15fIKjLI1V2FLOZTF91SEmLdvHsaRzAPh5udO7cTgDWkYSHlwI7WXSTlxc8Jzed+l9/cMuvsMTWLl49047vhN+fgIOrzbXI1qZAzsGX9wbU0SKBxU3V6DixgllpMAfr5t/ZWOAbzDcOhbq9y7eH6DXKSH5HJOW7WPanwdJOZcNQIi/N/1bRvBA06oE+hXxnaqzZ8xxg/5Z8JzYBVzifyG+wRcXPMFR1r9f9ixY/gEsfMPsUu9VGjqNhuh+1mcTkStScXMFKm6czK458OvQvx+T1P8XdH4dSoVYm6sQ7UpIYeLiOH7ecIQsu/nrWa1cKQa2jqJ7w0r4eBajRyYZqZCw9YLBB7eDI/vifb0D/3qk9Y+Cp2z1onsEdGyTebcmfpO5XqOTeecvsHLRnF9ErouKmytQceMkUhJg9r9h6wxzPaiq+UHkopMTGobByrhTTFy8lwU7j+dsbxIRzMDWUdxSqzxubk5yZyHrnDnYYE7Bs8EsgC41+KCnX67BB6nQAMrVKtj2U9kZsPhtWPquWXT5BMFtb7r8nT8RV5Ofz2/1cZTixTBg/VfmrMvnksDmDs2fMLvlepWyOl2By7Y7mL01nomL49h0OAkwP29vrRvGwNZR3FSljMUJr4Gnj9lr7Z891+xZ5nQSuQYf3AxZ6XDoT/PrPHcvCK2bu+ApX9c8bn4dXmPerTm+w1yvfQfc/rbzNoIWkTzRnRspPk7sgVmD4cBSc71CA7ORZ4UG1uYqBOmZ2Xy/5jCfL43j0ClzsD1vDzfuianMwzdHERHieoXcRRx2c8LQCxsuZyRfvK/N/R+DDzb4e/DBy41nlJkOC8fAivHmNBylyplFTd0ehXpJIlJ49FjqClTcFEPZmbD8fVj0X7BnmI8q2r0ITR9zuQHUTqRm8OXy/Xy18gBn0rMAKOPnSZ/mEfRpXpWy/iV8eH+HA87sv3gsnrOnLrGzDUJq5C54wuqbj8B+edIcvwbMx0+3vuGaY/SIlCAqbq5AxU0xc2i12b070ZztnWrtzakTykRYGqugxR1P5bMl+/hx3WEysx0AVC3rx8M3R3J3dDi+XsWokXBxYxiQfOTiOzwpx678utIVzXZaNW8tmpwiUqjU5kaKv3PJEPsarP4cMMCvrPnXdb17XKqR59oDp5iwKI552xM4/2dEg/AgBrWOonPdMNydpZGwlWw2s0dTYGWo1eXv7SkJZs+noxv+mjl909+Tpjbqa3bx9rnGgQ1FxKmpuJGit+N3+O1ZSDlqrje4z+ze7SKPDRwOg3nbE5i4OI61B/6ey6l9rfIMbB1Fk8jgS0+PIPlTOhRKd8w9QWr6KbNXVukw63KJiOVU3EjRSYmH35+H7b+Y62UioOs4qNbOylQF5lyWnZ/WHeHzJXHEnUgDwMvdjZ43VeLhVpHUCC1tccISwEUKZBG5PipupPA5HLDuS5g3EjL+6t7d4ilo82/wKoSpA4rY6bRMvl55gC9X7OdEqjmWS4CPBw80q0q/FhGUD7iGLswiInLNVNxI4Tq+y+zefXC5uV6xEdzxgTlwm5M7dCqdL5bu49vVhzibZQegUpAvA26OpHfjcPy99eslImIF/d9XCkd2Jix9D5a8bbaB8CwFt7wMTQc5/YzLmw6fYeLiOH7ffAzHX42E61QIYFCbKG6vVwFPdzdrA4qIlHAqbqTgHVxp3q05PypsjU7Q5R0IqmJtrutgGAYLdx5nwuK9rIz7e8yVVjVCGNS6Gi2rl1UjYRGRYkLFjRScc0kwfxSs+cJcL1XOnMOn7p1O2707M9vBzxuO8NmSOHYlpALg4WajW4OKPNIqijoVNVaSiEhxo+JGCsb2WWZPqPMDq930AHQc7bS9V5LPZTHtz4NMXraPhOQMAEp5uXNvkyoMuDmSikG+FicUEZHLUXEj1yf5qFnU7PjVXA+uBt3GQWRrS2Ndq2NJZ5m0dB//t+oQqRnZAJQv7U3/lpHc17QKgb4FOFu1iIgUChU3cm0cDlg7yXwMlZEMbh7QcjC0fh48ne+uxo74ZCYujuOXDUfJ/quVcI3y/jzSOoruDSvi7eHcjaBFREoSFTeSf4k7zPmgDv1prleKMbt3h9a1Nlc+GYbByrhTTFi8l4U7j+dsbxoZzKA2UbS9oTxumh5BRMTpqLiRvMvOgCXvwJJ3wZEFXv7QfiQ0fsipunfbHQZztsYzYdFeNh5OAsDNBrfdWIGBraNoEB5kbUAREbkuKm4kbw4sh1+ehpO7zfUbboMub5uTGTqJc1l2flh7mM+WxHHgZDoA3h5u3BNTmYdvjiIipJTFCUVEpCCouJErO3sG5o+EtVPMdf9QuO0tqNPdabp3n07L5KuVB/hy+X5OppnTIwT5edKneQR9mlclxN/b4oQiIlKQVNzIpRkGbPsZ/vcCpCaY2xr1hY6jwLeMtdny6HLTIzzSKpJejcPx89I/fxERV6T/u8vFkg6b3bt3/m6ul60B3d6HiJbW5sqjrUeTmLg4jl83HcP+V8+n89MjdKlXAQ9NjyAi4tJU3MjfHHZY/QXEjoLMVHDzhFZD4eah4Fm8Z7Y2DINle04yYfFeluw+kbP95uohDGoTxc3VQzQ9gohICaHiRkwJW835oA6vNtfDm5p3a8rXtjbXVWTbHfy+xez5tPVoMgDubja61DN7Pt1YKdDihCIiUtRU3JR0Wedg8X9h2ThwZINXaej4KkQPALfi+/gmPTOb71Yf4vOl+zh8+iwAvp7u9G4czkM3RxIe7GdxQhERsYqKm5Js3xLzbs2pveZ6ra5w+38hoKK1ua7gZGoGX644wNQV+zmTngVAcCkv+rWI4MFmVSlTysvihCIiYjUVNyVR+imYNwLWf2Wu+4eZY9bU7mZtris4cDKNz5fs47s1h8jIdgBQJdiPR1pHcXejyvh6Oc8ggiIiUrhU3JQkhgFbf4L//RvS/ppuIOYh6DASfIpn25RNh88wYVEc/9tyjL86PlG/ciCDWlfj1hvDcNf0CCIicgEVNyXFmUPw27Owe465HlLTnA+qSjNrc12CYRgs2nWcCYviWBF3Mmd725rlGNS6Gs2igtXzSURELkvFjatz2GHVRIgdDVlp4O4FrZ6Dm4eAR/EamTfL7mDWxqNMXBzHjvgUADzcbNzRoCKPtI6idoUAixOKiIgzUHHjyuI3m/NBHV1nrldpbnbvLlfT2lwXSM3IZvqqg0xauo+jSecAKOXlzr1NqjDg5kgqBvlanFBERJyJihtXlHUWFr0Jyz4Aww7egea0CY36Fqvu3Ykp5/hy+X6+WnGA5HPZAIT4e9O/ZQQPNK1KoJ+nxQlFRMQZqbhxNXELYdYQOL3PXK99h9m9u3SYlalyiTueymdL9vHjusNk/tXzKSqkFI+0jqLnTZXw8VTPJxERuXYqblxF+imY8xJsnGaul65odu+u1cXaXP+w7uBpJizay9xtCRh/9Xy6qUoQg1pXo2OdUPV8EhGRAqHixtkZBmz+AWYPg/QTgA0aPwztR4CP9Q1wHQ6DP3YkMnFxHKv2n8rZ3qF2eQa1qUZM1TLq+SQiIgVKxY0zO30AfhsKe+ab6+Vqm927w5tYmwvIyLbz8waz59OexFQAPN1t9LypEo+0iqJGaGmLE4qIiKtSceOM7Nnw5yewYAxkpZvdu1u/AC0Hg4e10w8kn8vi//48yKRl+0hIzgCgtLcH9zWrQv8WkYQFFu/ZxUVExPmpuHE2xzaa3buPbTDXq94M3cZBSA0rU5GQfI5Jy/YxbeVBUjLMnk+hAd4MaBnJvU2rEOCjnk8iIlI0VNw4i8x0WDgGVnxsdu/2CYSOo+GmBy3t3r07IYWJi+OYueEIWXazlXD18v4MbB1F94YV8fZQzycRESlaKm6cwZ5Y+PUZOHPAXK/bE259E0qHWhLHMAzWHDB7Ps3fnpizvUlEMIPaRNGuZnnc1PNJREQsouKmOEs7CXOGw6ZvzfWAytDlHah5qyVxHA6DudsSmLB4L+sPngHAZoNOdUIZ2Loa0VXLWJJLRETkn1TcFEeGYRY0s4fD2VOADZoOglteBu+i72V0LsvOjPVH+GxxHHEn0gDw8nDjrkaVeaRVJFHl/Is8k4iIyOWouCluTu0zH0HFLTDXy9eFOz6EytFFHiUpPYuv/zzA5GX7OZFq9nwK8PHgweZV6dsigvKl1fNJRESKHxU3xYU9G1aOhwVjIfssuHtD239Di6fBvWh7Gh05c5ZJS/cxfdVB0jLtAFQI9OGhmyP5V5Mq+Hvrn42IiBRf+pQqDo6uh1+eMmfxBohsDV3HQdlqRRpjR3wyExfF8cvGo2Q7zJ5PtcJKM7B1FN0aVMTTvfhMuikiInI5Km6slJlmDsS38mMwHOATBJ3HQMP7zJa6RcAwDFbEnWTi4jgW7jyes715VFkGtYmizQ3lND2CiIg4FRU3Vtk932xbk3TQXL/xbrj1DfAvVySntzsMZm+JZ8LivWw6nASAmw1uq1eBQa2jqF85qEhyiIiIFDQVN0Ut9bjZvXvz9+Z6YBXo+i7U6Fgkpz+XZef7tYf5fEkcB06mA+Dt4UavmHAebhVJ1bKliiSHiIhIYVFxU1QMAzZMg7kvwdnTYHODZo9D2+HgXfhdqU+nZTJ1xQGmrtjPybRMAIL8POnTPIK+zatS1t+70DOIiIgUBRU3ReHkXvh1COxbbK6H1YNuH0ClRoV+6kOn0vli6T6+XX2Is1lmz6dKQb480iqSXo3D8fPSPwEREXEt+mQrTPYsWP4hLHoTss+Bhy+0G27esSnk7t1bjiQxcXEcv20+hv2vnk91KwYwqE01br8xDA/1fBIRERel4qawHF4Ls56GhC3melRb6PoeBEcV2ikNw2DpnhNMWBTH0j0ncra3qhHCoNbVaFm9rHo+iYiIy1NxU9AyUuCP1+HPTwEDfIPh1rFQv3ehde/Otjv4bfMxJiyKY9uxZADc3Wx0rV+BR1pFcWOlwEI5r4iISHGk4qYg7ZoDvw6F5MPmev1/QefXoVRIoZwuPTObb1cf4oul+zh8+iwAvp7u9G4czkM3RxIe7Fco5xURESnOVNwUlA3TYOZj5nJQVfMRVPX2hXKqE6kZTF2+n6krD3AmPQuA4FJe9GsRwYPNqlKmlFehnFdERMQZqLgpKLW7wcI3oE53aDsMvAp+vJj9J9L4fGkc3685TEa2A4CqZf14pFUUd0dXxsfTvcDPKSIi4mxU3BQU79LwxJ/g6Vvgh9546AwTF8fxvy3H+KvjEw0qBzKoTTU61w3D3U2NhEVERM5TcVOQCrCwMQyDhbuOM2HRXlbGncrZ3q5mOQa2rkazqGD1fBIREbkEFTfFTGa2g1kbj/LZkjh2xKcA4OFm446GFRnYOopaYQEWJxQRESneVNwUE6kZ2UxfdZAvlu7jWNI5AEp5uXNvkyoMuDmSikEF/7hLRETEFam4sVhiyjmmLNvPVysPkHIuG4AQf2/6t4zggaZVCfQr3JGMRUREXI2KG4vsPZ7K50vi+HHtETLtZs+nqJBSDGwdRY+bKqnnk4iIyDVScVPE1h44zYRFe5m3PQHjr55PjaoEMahNNTrWDsVNPZ9ERESui4qbIuBwGPyxI5EJi/eyev/pnO0daofyaJsoYiKCLUwnIiLiWorF1NDjx48nIiICHx8fmjZtyqpVq664//fff0+tWrXw8fGhXr16/P7770WUNH8ysu18t/oQHd9bxMNT17B6/2k83W30iqnM/KGt+bxvjAobERGRAmb5nZtvv/2WoUOH8umnn9K0aVPGjRtH586d2blzJ+XLl79o/+XLl3PvvfcyduxYunbtyrRp0+jRowfr1q3jxhtvtOAKLpZ8Lotpfx5k0tJ9JKZkAFDa24P7mlVhQMtIQgN8LE4oIiLiumyGcb7lhzWaNm1K48aN+eijjwBwOByEh4fz1FNPMWzYsIv27927N2lpafz6668525o1a0bDhg359NNPr3q+5ORkAgMDSUpKIiCgYMeMiU86x+Rl+/jmz4OkZpg9n0IDvHno5kj+1aQKAT7q+SQiInIt8vP5bemdm8zMTNauXcvw4cNztrm5udGhQwdWrFhxydesWLGCoUOH5trWuXNnZs6cecn9MzIyyMjIyFlPTk6+/uCX8L/Nx3h6+nqy7GatWKO8PwNbR9G9YSW8PIrF0z8REZESwdLi5sSJE9jtdkJDQ3NtDw0NZceOHZd8TXx8/CX3j4+Pv+T+Y8eOZdSoUQUT+AqiI8pgs9loElmGQa2jaFezvHo+iYiIWMDyNjeFbfjw4bnu9CQnJxMeHl7g5ylf2ofYoW0ID/Yr8GOLiIhI3lla3ISEhODu7k5CQkKu7QkJCYSFhV3yNWFhYfna39vbG29v74IJfBUqbERERKxnaWMQLy8voqOjiY2NzdnmcDiIjY2lefPml3xN8+bNc+0PMG/evMvuLyIiIiWL5Y+lhg4dSt++fYmJiaFJkyaMGzeOtLQ0+vfvD0CfPn2oVKkSY8eOBWDw4MG0adOGd955hy5dujB9+nTWrFnDxIkTrbwMERERKSYsL2569+7N8ePHGTFiBPHx8TRs2JDZs2fnNBo+ePAgbm5/32Bq0aIF06ZN4+WXX+bFF1+kRo0azJw5s9iMcSMiIiLWsnycm6JWmOPciIiISOHIz+e3BmARERERl6LiRkRERFyKihsRERFxKSpuRERExKWouBERERGXouJGREREXIqKGxEREXEpKm5ERETEpai4EREREZdi+fQLRe38gMzJyckWJxEREZG8Ov+5nZeJFUpccZOSkgJAeHi4xUlEREQkv1JSUggMDLziPiVubimHw8HRo0cpXbo0NputQI+dnJxMeHg4hw4dcsl5q1z9+sD1r1HX5/xc/Rp1fc6vsK7RMAxSUlKoWLFirgm1L6XE3blxc3OjcuXKhXqOgIAAl/1HC65/feD616jrc36ufo26PudXGNd4tTs256lBsYiIiLgUFTciIiLiUlTcFCBvb29GjhyJt7e31VEKhatfH7j+Ner6nJ+rX6Ouz/kVh2sscQ2KRURExLXpzo2IiIi4FBU3IiIi4lJU3IiIiIhLUXEjIiIiLkXFTT6NHz+eiIgIfHx8aNq0KatWrbri/t9//z21atXCx8eHevXq8fvvvxdR0muTn+ubMmUKNpst15ePj08Rps2fxYsX061bNypWrIjNZmPmzJlXfc3ChQtp1KgR3t7eVK9enSlTphR6zmuV3+tbuHDhRe+fzWYjPj6+aALn09ixY2ncuDGlS5emfPny9OjRg507d171dc70O3gt1+hMv4effPIJ9evXzxncrXnz5vzvf/+74muc6f3L7/U503t3KW+88QY2m40hQ4ZccT8r3kMVN/nw7bffMnToUEaOHMm6deto0KABnTt3JjEx8ZL7L1++nHvvvZeHHnqI9evX06NHD3r06MGWLVuKOHne5Pf6wByB8tixYzlfBw4cKMLE+ZOWlkaDBg0YP358nvbft28fXbp0oV27dmzYsIEhQ4bw8MMPM2fOnEJOem3ye33n7dy5M9d7WL58+UJKeH0WLVrEE088wcqVK5k3bx5ZWVl06tSJtLS0y77G2X4Hr+UawXl+DytXrswbb7zB2rVrWbNmDbfccgvdu3dn69atl9zf2d6//F4fOM97d6HVq1czYcIE6tevf8X9LHsPDcmzJk2aGE888UTOut1uNypWrGiMHTv2kvv36tXL6NKlS65tTZs2NQYNGlSoOa9Vfq9v8uTJRmBgYBGlK1iAMWPGjCvu88ILLxh169bNta13795G586dCzFZwcjL9S1YsMAAjNOnTxdJpoKWmJhoAMaiRYsuu4+z/Q5eKC/X6My/h4ZhGGXKlDE+//zzS37P2d8/w7jy9Tnre5eSkmLUqFHDmDdvntGmTRtj8ODBl93XqvdQd27yKDMzk7Vr19KhQ4ecbW5ubnTo0IEVK1Zc8jUrVqzItT9A586dL7u/la7l+gBSU1OpWrUq4eHhV/0Lxdk40/t3PRo2bEiFChXo2LEjy5YtszpOniUlJQEQHBx82X2c/T3MyzWCc/4e2u12pk+fTlpaGs2bN7/kPs78/uXl+sA537snnniCLl26XPTeXIpV76GKmzw6ceIEdrud0NDQXNtDQ0Mv20YhPj4+X/tb6Vqur2bNmkyaNImff/6Zr7/+GofDQYsWLTh8+HBRRC50l3v/kpOTOXv2rEWpCk6FChX49NNP+fHHH/nxxx8JDw+nbdu2rFu3zupoV+VwOBgyZAgtW7bkxhtvvOx+zvQ7eKG8XqOz/R5u3rwZf39/vL29efTRR5kxYwZ16tS55L7O+P7l5/qc7b0DmD59OuvWrWPs2LF52t+q97DEzQouBad58+a5/iJp0aIFtWvXZsKECYwePdrCZJIXNWvWpGbNmjnrLVq0YO/evbz33nt89dVXFia7uieeeIItW7awdOlSq6MUmrxeo7P9HtasWZMNGzaQlJTEDz/8QN++fVm0aNFlCwBnk5/rc7b37tChQwwePJh58+YV+4bPKm7yKCQkBHd3dxISEnJtT0hIICws7JKvCQsLy9f+VrqW67uQp6cnN910E3v27CmMiEXucu9fQEAAvr6+FqUqXE2aNCn2BcOTTz7Jr7/+yuLFi6lcufIV93Wm38F/ys81Xqi4/x56eXlRvXp1AKKjo1m9ejXvv/8+EyZMuGhfZ3z/8nN9Fyru793atWtJTEykUaNGOdvsdjuLFy/mo48+IiMjA3d391yvseo91GOpPPLy8iI6OprY2NicbQ6Hg9jY2Ms+T23evHmu/QHmzZt3xeevVrmW67uQ3W5n8+bNVKhQobBiFilnev8KyoYNG4rt+2cYBk8++SQzZszgjz/+IDIy8qqvcbb38Fqu8ULO9nvocDjIyMi45Pec7f27lCtd34WK+3vXvn17Nm/ezIYNG3K+YmJiuP/++9mwYcNFhQ1Y+B4WanNlFzN9+nTD29vbmDJlirFt2zZj4MCBRlBQkBEfH28YhmE8+OCDxrBhw3L2X7ZsmeHh4WG8/fbbxvbt242RI0canp6exubNm626hCvK7/WNGjXKmDNnjrF3715j7dq1xr/+9S/Dx8fH2Lp1q1WXcEUpKSnG+vXrjfXr1xuA8e677xrr1683Dhw4YBiGYQwbNsx48MEHc/aPi4sz/Pz8jOeff97Yvn27MX78eMPd3d2YPXu2VZdwRfm9vvfee8+YOXOmsXv3bmPz5s3G4MGDDTc3N2P+/PlWXcIVPfbYY0ZgYKCxcOFC49ixYzlf6enpOfs4++/gtVyjM/0eDhs2zFi0aJGxb98+Y9OmTcawYcMMm81mzJ071zAM53//8nt9zvTeXc6FvaWKy3uo4iafPvzwQ6NKlSqGl5eX0aRJE2PlypU532vTpo3Rt2/fXPt/9913xg033GB4eXkZdevWNX777bciTpw/+bm+IUOG5OwbGhpq3H777ca6dessSJ0357s+X/h1/pr69u1rtGnT5qLXNGzY0PDy8jKioqKMyZMnF3nuvMrv9b355ptGtWrVDB8fHyM4ONho27at8ccff1gTPg8udW1ArvfE2X8Hr+Uanen3cMCAAUbVqlUNLy8vo1y5ckb79u1zPvgNw/nfv/xenzO9d5dzYXFTXN5Dm2EYRuHeGxIREREpOmpzIyIiIi5FxY2IiIi4FBU3IiIi4lJU3IiIiIhLUXEjIiIiLkXFjYiIiLgUFTciIiLiUlTciEiJt3DhQmw2G2fOnLE6iogUABU3IiIi4lJU3IiIiIhLUXEjIpZzOByMHTuWyMhIfH19adCgAT/88APw9yOj3377jfr16+Pj40OzZs3YsmVLrmP8+OOP1K1bF29vbyIiInjnnXdyfT8jI4N///vfhIeH4+3tTfXq1fniiy9y7bN27VpiYmLw8/OjRYsW7Ny5s3AvXEQKhYobEbHc2LFjmTp1Kp9++ilbt27lmWee4YEHHmDRokU5+zz//PO88847rF69mnLlytGtWzeysrIAsyjp1asX//rXv9i8eTOvvvoqr7zyClOmTMl5fZ8+ffi///s/PvjgA7Zv386ECRPw9/fPleOll17inXfeYc2aNXh4eDBgwIAiuX4RKViaOFNELJWRkUFwcDDz58+nefPmOdsffvhh0tPTGThwIO3atWP69On07t0bgFOnTlG5cmWmTJlCr169uP/++zl+/Dhz587Nef0LL7zAb7/9xtatW9m1axc1a9Zk3rx5dOjQ4aIMCxcupF27dsyfP5/27dsD8Pvvv9OlSxfOnj2Lj49PIf8URKQg6c6NiFhqz549pKen07FjR/z9/XO+pk6dyt69e3P2+2fhExwcTM2aNdm+fTsA27dvp2XLlrmO27JlS3bv3o3dbmfDhg24u7vTpk2bK2apX79+znKFChUASExMvO5rFJGi5WF1ABEp2VJTUwH47bffqFSpUq7veXt75ypwrpWvr2+e9vP09MxZttlsgNkeSESci+7ciIil6tSpg7e3NwcPHqR69eq5vsLDw3P2W7lyZc7y6dOn2bVrF7Vr1wagdu3aLFu2LNdxly1bxg033IC7uzv16tXD4XDkasMjIq5Ld25ExFKlS5fmueee45lnnsHhcHDzzTeTlJTEsmXLCAgIoGrVqgC89tprlC1bltDQUF566SVCQkLo0aMHAM8++yyNGzdm9OjR9O7dmxUrVvDRRx/x8ccfAxAREUHfvn0ZMGAAH3zwAQ0aNODAgQMkJibSq1cvqy5dRAqJihsRsdzo0aMpV64cY8eOJS4ujqCgIBo1asSLL76Y81jojTfeYPDgwezevZuGDRsya9YsvLy8AGjUqBHfffcdI0aMYPTo0VSoUIHXXnuNfv365Zzjk08+4cUXX+Txxx/n5MmTVKlShRdffNGKyxWRQqbeUiJSrJ3vyXT69GmCgoKsjiMiTkBtbkRERMSlqLgRERERl6LHUiIiIuJSdOdGREREXIqKGxEREXEpKm5ERETEpai4EREREZei4kZERERcioobERERcSkqbkRERMSlqLgRERERl6LiRkRERFzK/wMqq01TmkOaQgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(model.history.history['rec@prec'])\n",
    "plt.plot(model.history.history['val_rec@prec'])\n",
    "plt.title('model metrics')\n",
    "plt.ylabel('recall at precision')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "NB: когда мы оборачиваем LSTM в Bidirectional, то этот слой применяется и напрямую и обратно. Кажется, что логичнее иметь один слой для прямого прогона и один для обратного.\n",
    "В Bidirectional можно также указать отдельный backward_layer, чтобы использовался новый слой."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = keras.layers.Input(shape=(MAX_LEN,))\n",
    "embeddings = keras.layers.Embedding(input_dim=len(word2id), output_dim=100)(inputs, )\n",
    "\n",
    "# 1 слой на 2 прогона вручную\n",
    "# lstm = keras.layers.LSTM(128, return_sequences=False)\n",
    "# lstm_forward = lstm(embeddings)\n",
    "# lstm_backward = lstm(embeddings[::-1])\n",
    "\n",
    "# 2 слоя вручную\n",
    "lstm_forward =  keras.layers.LSTM(128, return_sequences=False)(embeddings)\n",
    "lstm_backward =  keras.layers.LSTM(128, return_sequences=False, go_backwards=True)(embeddings)\n",
    "\n",
    "hidden = keras.layers.concatenate([lstm_forward, lstm_backward])\n",
    "\n",
    "# 2 слоя через Bidirectional\n",
    "# hidden = keras.layers.Bidirectional(\n",
    "#                                        keras.layers.LSTM(128, return_sequences=False),\n",
    "#                         backward_layer=keras.layers.LSTM(128, return_sequences=False, \n",
    "#                                                             go_backwards=True))(embeddings)\n",
    "\n",
    "outputs = keras.layers.Dense(len(label2id), activation='softmax')(hidden)\n",
    "\n",
    "model = keras.Model(inputs=inputs, outputs=outputs)\n",
    "\n",
    "model.compile(optimizer='rmsprop',\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=[keras.metrics.RecallAtPrecision(0.8, name='rec@prec')])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"functional_7\"</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1mModel: \"functional_7\"\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\"> Layer (type)        </span>┃<span style=\"font-weight: bold\"> Output Shape      </span>┃<span style=\"font-weight: bold\">    Param # </span>┃<span style=\"font-weight: bold\"> Connected to      </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩\n",
       "│ input_layer_4       │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">200</span>)       │          <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │ -                 │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)        │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_4         │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">200</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">100</span>)  │  <span style=\"color: #00af00; text-decoration-color: #00af00\">2,409,300</span> │ input_layer_4[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]… │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Embedding</span>)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ lstm_3 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">LSTM</span>)       │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)       │    <span style=\"color: #00af00; text-decoration-color: #00af00\">117,248</span> │ embedding_4[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>] │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ lstm_4 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">LSTM</span>)       │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)       │    <span style=\"color: #00af00; text-decoration-color: #00af00\">117,248</span> │ embedding_4[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>] │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ concatenate         │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">256</span>)       │          <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │ lstm_3[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>],     │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Concatenate</span>)       │                   │            │ lstm_4[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]      │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ dense_4 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">19</span>)        │      <span style=\"color: #00af00; text-decoration-color: #00af00\">4,883</span> │ concatenate[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>] │\n",
       "└─────────────────────┴───────────────────┴────────────┴───────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape     \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m   Param #\u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mConnected to     \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩\n",
       "│ input_layer_4       │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m200\u001b[0m)       │          \u001b[38;5;34m0\u001b[0m │ -                 │\n",
       "│ (\u001b[38;5;33mInputLayer\u001b[0m)        │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_4         │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m200\u001b[0m, \u001b[38;5;34m100\u001b[0m)  │  \u001b[38;5;34m2,409,300\u001b[0m │ input_layer_4[\u001b[38;5;34m0\u001b[0m]… │\n",
       "│ (\u001b[38;5;33mEmbedding\u001b[0m)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ lstm_3 (\u001b[38;5;33mLSTM\u001b[0m)       │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m128\u001b[0m)       │    \u001b[38;5;34m117,248\u001b[0m │ embedding_4[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ lstm_4 (\u001b[38;5;33mLSTM\u001b[0m)       │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m128\u001b[0m)       │    \u001b[38;5;34m117,248\u001b[0m │ embedding_4[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ concatenate         │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m)       │          \u001b[38;5;34m0\u001b[0m │ lstm_3[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m],     │\n",
       "│ (\u001b[38;5;33mConcatenate\u001b[0m)       │                   │            │ lstm_4[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m]      │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ dense_4 (\u001b[38;5;33mDense\u001b[0m)     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m19\u001b[0m)        │      \u001b[38;5;34m4,883\u001b[0m │ concatenate[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n",
       "└─────────────────────┴───────────────────┴────────────┴───────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">2,648,679</span> (10.10 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m2,648,679\u001b[0m (10.10 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">2,648,679</span> (10.10 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m2,648,679\u001b[0m (10.10 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model.summary()\n",
    "# обратите внимание на размер output в biderectional\n",
    "# он равен units * 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m158s\u001b[0m 747ms/step - loss: 2.3474 - rec@prec: 8.1717e-06 - val_loss: 1.9513 - val_rec@prec: 0.0090\n",
      "Epoch 2/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m147s\u001b[0m 697ms/step - loss: 1.9130 - rec@prec: 0.0029 - val_loss: 1.8059 - val_rec@prec: 0.0334\n",
      "Epoch 3/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m153s\u001b[0m 722ms/step - loss: 1.6005 - rec@prec: 0.1076 - val_loss: 2.2274 - val_rec@prec: 0.0419\n",
      "Epoch 4/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m141s\u001b[0m 668ms/step - loss: 1.4400 - rec@prec: 0.1686 - val_loss: 1.3343 - val_rec@prec: 0.3084\n",
      "Epoch 5/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m161s\u001b[0m 765ms/step - loss: 1.1950 - rec@prec: 0.4057 - val_loss: 1.1789 - val_rec@prec: 0.4784\n",
      "Epoch 6/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m149s\u001b[0m 707ms/step - loss: 1.0621 - rec@prec: 0.5585 - val_loss: 1.1777 - val_rec@prec: 0.4901\n",
      "Epoch 7/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m164s\u001b[0m 776ms/step - loss: 0.9190 - rec@prec: 0.6704 - val_loss: 1.0938 - val_rec@prec: 0.5753\n",
      "Epoch 8/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m194s\u001b[0m 919ms/step - loss: 0.8104 - rec@prec: 0.7420 - val_loss: 1.0677 - val_rec@prec: 0.6249\n",
      "Epoch 9/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m169s\u001b[0m 799ms/step - loss: 0.7079 - rec@prec: 0.8021 - val_loss: 1.0145 - val_rec@prec: 0.6312\n",
      "Epoch 10/10\n",
      "\u001b[1m211/211\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m161s\u001b[0m 761ms/step - loss: 0.6056 - rec@prec: 0.8504 - val_loss: 1.0441 - val_rec@prec: 0.6285\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.src.callbacks.history.History at 0x30c283fd0>"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train, y_train, \n",
    "          validation_data=(X_valid, y_valid),\n",
    "          batch_size=200,\n",
    "         epochs=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(model.history.history['rec@prec'])\n",
    "plt.plot(model.history.history['val_rec@prec'])\n",
    "plt.title('model metrics')\n",
    "plt.ylabel('recall at precision')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Sequence labelling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "На выходе из рекурректной сети получается по вектору на каждый элемент последовательности. Мы можем использовать только последний, если нам нужно просто классифицировать весь текст (на последнем состоянии по идее должна была накопиться информация о всей последовательности).\n",
    "\n",
    "Кроме этого, мы можем классифицировать каждый элемент последовательности. Это называется sequence labelling (разметка последовательности). Под этот паттерн попадает предсказание части речи, определение именованных сущностей, определение опечаток и другие задачи, где длина обоих последовательностей совпадает.\n",
    "\n",
    "Давайте попробуем выделять именованные сущности."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Данные возьмем из бибиотеки datasets (часть huggingface)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# !pip install datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import load_dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "В этой библиотеке несколько датасетов для именованных сущностей. Мы возьмем датасет, собранный по википедии. Он мультиязычный, но для семинара мы возьмем только русскую часть."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "184f1b21d07542efab1636ead02be479",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading builder script:   0%|          | 0.00/5.16k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d202c2a39ae6428f9885debcb1b16cfa",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading readme:   0%|          | 0.00/12.7k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading and preparing dataset wikiann/ru to /Users/mnefedov/.cache/huggingface/datasets/tner___wikiann/ru/1.1.0/39367cc2fcf8467e7d7d81fdd2e3b5277c3b0c003bbe7f3e5e4895a41a141507...\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "73cc8c3c82474ec6a6744772958348e4",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading data files:   0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "275db003a081432bb7b25cfa2d7336e7",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading data:   0%|          | 0.00/2.64M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "9c7bb85c2da34d9c92ff9f337a3692c7",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading data:   0%|          | 0.00/5.26M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "91e112be30be4b93b9d31d5392450da0",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Downloading data:   0%|          | 0.00/2.63M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a535b0e53a934e66b11d1209645226e8",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Extracting data files:   0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Generating train split: 0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Generating validation split: 0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Generating test split: 0 examples [00:00, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset wikiann downloaded and prepared to /Users/mnefedov/.cache/huggingface/datasets/tner___wikiann/ru/1.1.0/39367cc2fcf8467e7d7d81fdd2e3b5277c3b0c003bbe7f3e5e4895a41a141507. Subsequent calls will reuse this data.\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1fca5d14f2ec42b88395a032760523e3",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "  0%|          | 0/3 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "dataset = load_dataset(\"tner/wikiann\", 'ru')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Датасет уже заранее разбит на трейн, валидацию и тест."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    train: Dataset({\n",
       "        features: ['tokens', 'tags'],\n",
       "        num_rows: 20000\n",
       "    })\n",
       "    validation: Dataset({\n",
       "        features: ['tokens', 'tags'],\n",
       "        num_rows: 10000\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['tokens', 'tags'],\n",
       "        num_rows: 10000\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Пример данных"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(['Илизаров', ',', 'Гавриил', 'Абрамович'], [2, 5, 5, 5])"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset['train']['tokens'][0], dataset['train']['tags'][0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Индексы в разметке соответствуют вот таким классам:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['O', 'B-PER', 'I-PER', 'B-ORG', 'I-ORG', 'B-LOC', 'I-LOC']"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "['O', 'B-PER', 'I-PER', 'B-ORG', 'I-ORG',  'B-LOC', 'I-LOC']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Что они значат?\n",
    "![](https://editor.analyticsvidhya.com/uploads/19617Intro%20image.jpg)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**PER, ORG, LOC** - это типы именованных сущностей, Person, Organization и Location. Это стандартные типы сущностей, которые чаще всего приходится выделять. Оценочные датасеты обычно размечены этими типами. Другие популярные сущности - номера телефонов, числа, имейлы, дата, время, но их обычно определяют правилами (регулярками например).\n",
    "\n",
    "**I, B** - эта часть тэга показывает в какой части именованной сущности находится токен. **B - beggining** (начало), **I - inside** (внутри, т.е. после начала).\n",
    "\n",
    "**O** - это отрицательный класс (outside или other), т.е. несущность.\n",
    "\n",
    "Такую разметку назвают IOB, она помогает точнее определять сущности и различать рядом стоящие сущности одного типа. Есть и другие типы (например, BILOU - L означает last, последний токен в сущности, а U - unit, сущность из 1 токена). "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Именованной сущностью в принципе можно считать любую последовательность токенов, которую нужно выделить из текста. В биоинформатике, например, сущностями считаются белки, гены, препараты, заболевания, а при разборе научных статей можно выделять цитаты, ключевые метрики, методы (используемые алгоритмы и модели). \n",
    "\n",
    "Для того, чтобы извлекать специфичные типы сущностей вам скорее всего придется либо а) самому размечать данные, либо б) создавать словари, правила и грамматики. А скорее всего и то, и другое. Сущности, размеченные моделью обычно еще нужно привести к стандартному виду и делается это в большинстве случаев маппингом на словарь стандартных форм (например, Вышка, ВШЭ, вше, НИУВШЭ, Высшая школа экономики -> НИУ ВШЭ). Маппинг можно делать с помощью простого словаря или с помощью какой-нибудь функции близости (например, расстояния редактирования).\n",
    "\n",
    "С помощью правил также можно размечать данные и обучать на этом более сложные модели (в надежде, что сложные модели запомнят разметку и дополнительно научатся обобщать). Такой подход называется weak labelling/weak supervision. Для этого есть такие библиотеки: [Snorkel](https://www.snorkel.org/) , [Rubrix](https://github.com/recognai/rubrix)\n",
    "\n",
    "Для создания грамматик на русском языке можно использовать [Tomita Parser](https://yandex.ru/dev/tomita/), [Yargy](https://github.com/natasha/yargy). Томита быстрее, но сложнее в использовании (внутри там майстем). Yargy написан на питоне и его проще освоить, внутри там пайморфи. В тетрадке NER_rule_based.ipynb (лежит в этой же папке) есть пример написания грамматик на Yargy. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Давайте посмотрим, как обучать модели, когда разметка уже есть"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "# словарь\n",
    "vocab = Counter()\n",
    "\n",
    "for sent in dataset['train']['tokens']:\n",
    "    vocab.update([x.lower() for x in sent])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "36015"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# тут словарь не очень большой и тут нужно предсказывать тэги для всех слов\n",
    "# поэтому не будем фильтровать словарь\n",
    "len(vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [],
   "source": [
    "# индексируем слова\n",
    "word2id = {'PAD':0, 'UNK':1}\n",
    "\n",
    "for word in vocab:\n",
    "    word2id[word] = len(word2id)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "id2word = {i:word for word, i in word2id.items()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "# переводим тексты в последовательности индексов\n",
    "X = []\n",
    "\n",
    "for sent in dataset['train']['tokens']:\n",
    "    tokens = [w.lower() for w in sent]\n",
    "    ids = [word2id.get(token, 1) for token in tokens]\n",
    "    X.append(ids)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "# переводим тексты в последовательности индексов\n",
    "X_test = []\n",
    "\n",
    "for sent in dataset['test']['tokens']:\n",
    "    tokens = [w.lower() for w in sent]\n",
    "    ids = [word2id.get(token, 1) for token in tokens]\n",
    "    X_test.append(ids)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "MAX_LEN = max(len(x) for x in X)\n",
    "\n",
    "# паддинг\n",
    "X = keras.preprocessing.sequence.pad_sequences(X, maxlen=MAX_LEN, padding='post')\n",
    "X_test = keras.preprocessing.sequence.pad_sequences(X_test, maxlen=MAX_LEN, padding='post')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Теги уже переведены в индексы, но обратный маппинг мы тоже сделаем, чтобы дальше удобнее было проверять работу модели. \n",
    "Длина последовательностей тегов должна совпадать с длиной последовательностей токенов, поэтому тут тоже нужно сделать паддинг, но в этом случае мы не можем паддить нулями, т.к. ноль у нас уже забронирован тэгом O. Добавим дополнительный тэг пад и будем использовать его индекс при паддинге (value=7)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "id2labels = {0:'O', 1:'B-PER', 2:'I-PER', 3:'B-ORG', 4:'I-ORG', 5: 'B-LOC', 6:'I-LOC', 7:'PAD'}\n",
    "label2id = {v:k for k,v in id2labels.items()} "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "y = keras.preprocessing.sequence.pad_sequences(dataset['train']['tags'], value=7,\n",
    "                                                  maxlen=MAX_LEN,  padding='post')\n",
    "y_test = keras.preprocessing.sequence.pad_sequences(dataset['test']['tags'], value=7,\n",
    "                                                       maxlen=MAX_LEN,  padding='post')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Схема задачи, чтобы было понятнее, что мы предсказываем:\n",
    "![](image_seq_labelling.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((20000, 54), (10000, 54))"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X.shape, X_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((20000, 54), (10000, 54))"
      ]
     },
     "execution_count": 70,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y.shape, y_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = keras.layers.Input(shape=(MAX_LEN,))\n",
    "embeddings = keras.layers.Embedding(input_dim=len(word2id), output_dim=100)(inputs)\n",
    "\n",
    "lstm_1 = keras.layers.Bidirectional(keras.layers.LSTM(128, return_sequences=True))(embeddings)\n",
    "\n",
    "outputs = keras.layers.Dense(len(label2id), activation='softmax')(lstm_1)\n",
    "\n",
    "model = keras.Model(inputs=inputs, outputs=outputs)\n",
    "model.compile(optimizer='adam',\n",
    "              loss='sparse_categorical_crossentropy', \n",
    "             metrics=['accuracy'])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/5\n",
      "\u001b[1m157/157\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m37s\u001b[0m 234ms/step - accuracy: 0.8877 - loss: 0.5748 - val_accuracy: 0.9490 - val_loss: 0.1677\n",
      "Epoch 2/5\n",
      "\u001b[1m157/157\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m44s\u001b[0m 280ms/step - accuracy: 0.9547 - loss: 0.1456 - val_accuracy: 0.9719 - val_loss: 0.0934\n",
      "Epoch 3/5\n",
      "\u001b[1m157/157\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m66s\u001b[0m 424ms/step - accuracy: 0.9790 - loss: 0.0700 - val_accuracy: 0.9806 - val_loss: 0.0627\n",
      "Epoch 4/5\n",
      "\u001b[1m157/157\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m44s\u001b[0m 280ms/step - accuracy: 0.9913 - loss: 0.0331 - val_accuracy: 0.9830 - val_loss: 0.0546\n",
      "Epoch 5/5\n",
      "\u001b[1m157/157\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m33s\u001b[0m 209ms/step - accuracy: 0.9961 - loss: 0.0165 - val_accuracy: 0.9833 - val_loss: 0.0539\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.src.callbacks.history.History at 0x2c2f6f4c0>"
      ]
     },
     "execution_count": 72,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X, y, \n",
    "          validation_data=(X_test, y_test),\n",
    "          batch_size=128,\n",
    "         epochs=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['accuracy', 'loss', 'val_accuracy', 'val_loss'])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(model.history.history.keys())\n",
    "plt.plot(model.history.history['accuracy'])\n",
    "plt.plot(model.history.history['val_accuracy'])\n",
    "plt.title('model acc')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Оценивать предсказания последовательностей нужно отдельно. Лучше конечно еще оценивать правильность всей последовательности, но пока сделаем только поэлементную оценку."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import classification_report"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m313/313\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m12s\u001b[0m 37ms/step\n"
     ]
    }
   ],
   "source": [
    "pred = model.predict(X_test).argmax(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[6, 6, 1, ..., 7, 7, 7],\n",
       "       [6, 6, 4, ..., 7, 7, 7],\n",
       "       [6, 6, 6, ..., 7, 7, 7],\n",
       "       ...,\n",
       "       [2, 5, 5, ..., 7, 7, 7],\n",
       "       [1, 4, 4, ..., 7, 7, 7],\n",
       "       [6, 6, 6, ..., 7, 7, 7]])"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           O       0.85      0.64      0.73      4560\n",
      "       B-PER       0.60      0.77      0.68      4074\n",
      "       I-PER       0.82      0.83      0.83      3542\n",
      "       B-ORG       0.88      0.72      0.79      3060\n",
      "       I-ORG       0.77      0.77      0.77      8008\n",
      "       B-LOC       0.85      0.89      0.87      7544\n",
      "       I-LOC       0.94      0.94      0.94     40480\n",
      "         PAD       1.00      1.00      1.00    468732\n",
      "\n",
      "    accuracy                           0.98    540000\n",
      "   macro avg       0.84      0.82      0.83    540000\n",
      "weighted avg       0.98      0.98      0.98    540000\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(classification_report(y_test.reshape(-1), pred.reshape(-1), labels=list(id2labels.keys()),\n",
    "                                                                     target_names=list(id2labels.values()),\n",
    "                                                                     zero_division=0))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Предсказание/Inference"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "После оценки часто хочется посмотреть как будет работать модель на каких-то отдельных примерах. Для этого нужно воспрозвести весь пайплайн и прогонять через него нужный текст."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [],
   "source": [
    "import re\n",
    "\n",
    "def tokenize(text, word2id):\n",
    "    # токенизирует и переводит в индексы\n",
    "    tokens = re.findall('\\w+|[^\\w\\s]+', text)\n",
    "    ids = [word2id.get(token.lower(), 1) for token in tokens]\n",
    "    return tokens, ids\n",
    "\n",
    "def pred2tags(pred, id2label, length):\n",
    "    # декодирует индексы в части речи\n",
    "    # length нужно чтобы откидывать паддинги или некорректные предсказания\n",
    "    pred = pred.argmax(2)[0, :length]\n",
    "    labels = [id2label[l] for l in pred]\n",
    "    return labels\n",
    "\n",
    "def label_seq(text, word2id, id2label, max_len, model):\n",
    "    tokens, ids = tokenize(text, word2id)\n",
    "    pred = model.predict(keras.preprocessing.sequence.pad_sequences([ids], \n",
    "                                                                       maxlen=max_len, \n",
    "                                                                       padding='post'))\n",
    "    labels = pred2tags(pred, id2label, len(ids))\n",
    "    \n",
    "    return list(zip(tokens, labels))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 37ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[('Алексей', 'I-PER'),\n",
       " ('сказал', 'I-LOC'),\n",
       " ('Светлане', 'B-LOC'),\n",
       " (',', 'I-LOC'),\n",
       " ('чтобы', 'I-LOC'),\n",
       " ('она', 'I-LOC'),\n",
       " ('собиралась', 'I-LOC'),\n",
       " ('на', 'I-LOC'),\n",
       " ('поезд', 'I-LOC'),\n",
       " ('в', 'I-LOC'),\n",
       " ('Москву', 'O'),\n",
       " ('.', 'I-LOC')]"
      ]
     },
     "execution_count": 79,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "label_seq('Алексей сказал Светлане, чтобы она собиралась на поезд в Москву.', word2id, id2labels, MAX_LEN, model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 36ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[('Я', 'I-LOC'),\n",
       " ('поступил', 'I-LOC'),\n",
       " ('в', 'I-LOC'),\n",
       " ('ВШЭ', 'I-LOC'),\n",
       " ('в', 'I-LOC'),\n",
       " ('Питер', 'B-LOC')]"
      ]
     },
     "execution_count": 80,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "label_seq('Я поступил в ВШЭ в Питер', word2id, id2labels, MAX_LEN, model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 46ms/step\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[('Виктор', 'I-PER'),\n",
       " (',', 'B-LOC'),\n",
       " ('Семен', 'B-LOC'),\n",
       " (',', 'B-LOC'),\n",
       " ('Петр', 'B-LOC'),\n",
       " (',', 'I-LOC'),\n",
       " ('Арагорн', 'I-PER'),\n",
       " (',', 'I-LOC'),\n",
       " ('Осел', 'I-LOC')]"
      ]
     },
     "execution_count": 82,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "label_seq('Виктор, Семен, Петр, Арагорн, Осел', word2id, id2labels, MAX_LEN, model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.16"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}