notebooks/word_stress_pytorch.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import time\n",
    "import glob\n",
    "import numba\n",
    "import random\n",
    "import itertools\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "from tqdm import tqdm\n",
    "from sklearn.model_selection import train_test_split\n",
    "from torch.nn.utils.rnn import pad_sequence\n",
    "from torch.utils.data import Dataset, DataLoader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'1.7.0+cu110'"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.__version__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_PATH = '../data/'\n",
    "RANDOM_SEED = 42"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def seed_everything(seed=0):\n",
    "    random.seed(seed)\n",
    "    os.environ['PYTHONHASHSEED'] = str(seed)\n",
    "    np.random.seed(seed)\n",
    "    torch.manual_seed(seed)\n",
    "    torch.cuda.manual_seed(seed)\n",
    "    torch.backends.cudnn.deterministic = True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "seed_everything(RANDOM_SEED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @numba.jit\n",
    "def stress_pos(x):\n",
    "    try:\n",
    "        res = np.zeros(len(x)-1)\n",
    "        stress_idx = x.find('+')\n",
    "        res[stress_idx] = 1\n",
    "        return res\n",
    "    except IndexError:\n",
    "        return np.NaN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 34/34 [00:04<00:00,  6.85it/s]\n"
     ]
    }
   ],
   "source": [
    "data = []\n",
    "\n",
    "for file in tqdm(glob.glob(DATA_PATH + \"dictionary/\" + \"*.txt\")):\n",
    "    with open(file, encoding='cp1251') as fil:\n",
    "        lines = fil.readlines()\n",
    "        for line in lines:\n",
    "            try:\n",
    "                word1, word2 = line.strip().split(\" \")\n",
    "                word1, word2 = word1.strip().replace('i', 'і'), word2.strip().replace('i', 'і')\n",
    "                data.append((word1, word2))\n",
    "            except ValueError:\n",
    "                continue"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = pd.DataFrame(data)\n",
    "data.columns = ['word', 'word_with_stress']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 53/53 [01:04<00:00,  1.22s/it]\n"
     ]
    }
   ],
   "source": [
    "exclude_list = [\n",
    "    '\\t', '$', '%', '-', '\\\\', '`', 'a', 'b', 'c', 'd', 'e', 'f', 'g', \n",
    "    'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', \n",
    "    'v', 'w', 'x', 'y', 'z', '~', '\\xa0', '\\xad', '·', 'ђ', 'ј', 'љ', 'њ', 'ћ', \n",
    "    'ў', 'џ', '–', '‘', '’', '“', '”', '•', '№',\n",
    "    'ъ', 'ы', 'э'\n",
    "]\n",
    "\n",
    "for item in tqdm(exclude_list):\n",
    "    data = data.loc[~data.word.str.contains(item, regex=False)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 12 s, sys: 1.29 s, total: 13.3 s\n",
      "Wall time: 13.2 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "data['word_list'] = data.word.map(list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 3.98 s, sys: 115 ms, total: 4.1 s\n",
      "Wall time: 4.1 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "data['word_stress_pos'] = data.word_with_stress.map(stress_pos)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = data.loc[data.word_stress_pos.notna()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "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>word</th>\n",
       "      <th>word_with_stress</th>\n",
       "      <th>word_list</th>\n",
       "      <th>word_stress_pos</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1024316</th>\n",
       "      <td>пивничого</td>\n",
       "      <td>пивн+ичого</td>\n",
       "      <td>[п, и, в, н, и, ч, о, г, о]</td>\n",
       "      <td>[0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1011665</th>\n",
       "      <td>перехитрило</td>\n",
       "      <td>перехитр+ило</td>\n",
       "      <td>[п, е, р, е, х, и, т, р, и, л, о]</td>\n",
       "      <td>[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>224015</th>\n",
       "      <td>відгризатиметься</td>\n",
       "      <td>відгриз+атиметься</td>\n",
       "      <td>[в, і, д, г, р, и, з, а, т, и, м, е, т, ь, с, я]</td>\n",
       "      <td>[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2127636</th>\n",
       "      <td>розбірною</td>\n",
       "      <td>розб+ірною</td>\n",
       "      <td>[р, о, з, б, і, р, н, о, ю]</td>\n",
       "      <td>[0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2792530</th>\n",
       "      <td>ексудат</td>\n",
       "      <td>ексуд+ат</td>\n",
       "      <td>[е, к, с, у, д, а, т]</td>\n",
       "      <td>[0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     word   word_with_stress  \\\n",
       "1024316         пивничого         пивн+ичого   \n",
       "1011665       перехитрило       перехитр+ило   \n",
       "224015   відгризатиметься  відгриз+атиметься   \n",
       "2127636         розбірною         розб+ірною   \n",
       "2792530           ексудат           ексуд+ат   \n",
       "\n",
       "                                                word_list  \\\n",
       "1024316                       [п, и, в, н, и, ч, о, г, о]   \n",
       "1011665                 [п, е, р, е, х, и, т, р, и, л, о]   \n",
       "224015   [в, і, д, г, р, и, з, а, т, и, м, е, т, ь, с, я]   \n",
       "2127636                       [р, о, з, б, і, р, н, о, ю]   \n",
       "2792530                             [е, к, с, у, д, а, т]   \n",
       "\n",
       "                                           word_stress_pos  \n",
       "1024316      [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]  \n",
       "1011665  [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, ...  \n",
       "224015   [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, ...  \n",
       "2127636      [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0]  \n",
       "2792530                [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]  "
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.sample(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "28"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "max_sequence_len = np.max(data.word.str.len())\n",
    "max_sequence_len"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "def flatten(array):\n",
    "    for item in array:\n",
    "        if isinstance(item, list):\n",
    "            yield from flatten(item)\n",
    "        else:\n",
    "            yield item\n",
    "\n",
    "\n",
    "class SequenceTokenizer:\n",
    "    \n",
    "    def __init__(self):\n",
    "        self.word2index = {}\n",
    "        self.index2word = {}\n",
    "        self.oov_token ='<UNK>'\n",
    "        self.oov_token_index = 0\n",
    "        \n",
    "    def fit(self, sequence):\n",
    "        self.index2word = dict(enumerate([self.oov_token] + sorted(set(flatten(sequence))), 1))\n",
    "        self.word2index = {v:k for k,v in self.index2word.items()}\n",
    "        self.oov_token_index = self.word2index.get(self.oov_token)\n",
    "        return self\n",
    "    \n",
    "    def transform(self, X):\n",
    "        res = []\n",
    "        for line in X:\n",
    "            res.append([self.word2index.get(item, self.oov_token_index) for item in line])\n",
    "        return res"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenizer = SequenceTokenizer()\n",
    "tokenizer.fit(data.word_list);"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'<UNK>': 1,\n",
       " 'а': 2,\n",
       " 'б': 3,\n",
       " 'в': 4,\n",
       " 'г': 5,\n",
       " 'д': 6,\n",
       " 'е': 7,\n",
       " 'ж': 8,\n",
       " 'з': 9,\n",
       " 'и': 10,\n",
       " 'й': 11,\n",
       " 'к': 12,\n",
       " 'л': 13,\n",
       " 'м': 14,\n",
       " 'н': 15,\n",
       " 'о': 16,\n",
       " 'п': 17,\n",
       " 'р': 18,\n",
       " 'с': 19,\n",
       " 'т': 20,\n",
       " 'у': 21,\n",
       " 'ф': 22,\n",
       " 'х': 23,\n",
       " 'ц': 24,\n",
       " 'ч': 25,\n",
       " 'ш': 26,\n",
       " 'щ': 27,\n",
       " 'ь': 28,\n",
       " 'ю': 29,\n",
       " 'я': 30,\n",
       " 'є': 31,\n",
       " 'і': 32,\n",
       " 'ї': 33,\n",
       " 'ґ': 34}"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tokenizer.word2index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = tokenizer.transform(data.word_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "def pad_sequence(lst, max_seq=max_sequence_len):\n",
    "    if isinstance(lst[0], list):\n",
    "        return np.array([i + [0]*(max_seq-len(i)) for i in lst])\n",
    "    else:\n",
    "        lst + [0]*(max_seq-len(lst))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "# %%time\n",
    "\n",
    "# input_seq = zip(*itertools.zip_longest(*X, fillvalue=0))\n",
    "# input_seq = list(map(list, input_seq))\n",
    "# input_seq = np.array(input_seq)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 12.1 s, sys: 620 ms, total: 12.7 s\n",
      "Wall time: 12.7 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "input_seq = pad_sequence(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2757242, 28)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[21, 32, 13, ...,  0,  0,  0],\n",
       "       [21, 32, 13, ...,  0,  0,  0],\n",
       "       [21, 32, 13, ...,  0,  0,  0],\n",
       "       ...,\n",
       "       [ 7, 22, 32, ...,  0,  0,  0],\n",
       "       [ 7, 13, 13, ...,  0,  0,  0],\n",
       "       [ 7, 14,  3, ...,  0,  0,  0]])"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(input_seq.shape)\n",
    "input_seq"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "y = data.word_stress_pos.values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "output_seq = zip(*itertools.zip_longest(*y, fillvalue=0))\n",
    "output_seq = list(map(list, output_seq))\n",
    "output_seq = np.array(output_seq).astype(int)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2757242, 28)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[0, 0, 0, ..., 0, 0, 0],\n",
       "       [0, 0, 0, ..., 0, 0, 0],\n",
       "       [0, 0, 0, ..., 0, 0, 0],\n",
       "       ...,\n",
       "       [0, 0, 1, ..., 0, 0, 0],\n",
       "       [0, 1, 0, ..., 0, 0, 0],\n",
       "       [0, 1, 0, ..., 0, 0, 0]])"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(output_seq.shape)\n",
    "output_seq"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "(input_seq_train, input_seq_val, \n",
    " output_seq_train, output_seq_val) = train_test_split(input_seq, \n",
    "                                                      output_seq, \n",
    "                                                      test_size=0.5, \n",
    "                                                      random_state=RANDOM_SEED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_seq_train = torch.tensor(input_seq_train, dtype=torch.long).cuda()\n",
    "input_seq_val = torch.tensor(input_seq_val, dtype=torch.long).cuda()\n",
    "output_seq_train = torch.tensor(output_seq_train, dtype=torch.float).cuda()\n",
    "output_seq_val = torch.tensor(output_seq_val, dtype=torch.float).cuda()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MyDataset(Dataset):\n",
    "    def __init__(self, dataset):\n",
    "        self.dataset = dataset\n",
    "        \n",
    "    def __getitem__(self, index):\n",
    "        data,target = self.dataset[index]\n",
    "        return data, target, index\n",
    "    \n",
    "    def __len__(self):\n",
    "        return len(self.dataset)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "class LSTM_model(nn.Module):\n",
    "\n",
    "    def __init__(self, embedding_dim, hidden_dim, vocab_size, target_size):\n",
    "        super(LSTM_model, self).__init__()\n",
    "        self.hidden_dim = hidden_dim\n",
    "\n",
    "        self.embeddings = nn.Embedding(vocab_size, embedding_dim)\n",
    "\n",
    "        self.lstm = nn.LSTM(input_size=self.embeddings.embedding_dim,\n",
    "                            hidden_size=hidden_dim,\n",
    "                            num_layers=2,\n",
    "                            batch_first=True,\n",
    "                            bidirectional=True)\n",
    "        self.linear = nn.Linear(self.hidden_dim * 4 , 64)\n",
    "        self.batch_norm = nn.BatchNorm1d(self.hidden_dim * 4, affine=False)\n",
    "        self.relu = nn.ReLU()\n",
    "        self.dropout = nn.Dropout(0.2)\n",
    "        self.out = nn.Linear(32, target_size)\n",
    "\n",
    "    def forward(self, x):\n",
    "        h_embeddings = self.embeddings(x)\n",
    "        \n",
    "        h_lstm, _ = self.lstm(h_embeddings)\n",
    "        avg_pool = torch.mean(h_lstm, 1)\n",
    "        max_pool, _ = torch.max(h_lstm, 1)\n",
    "        x = torch.cat((avg_pool, max_pool), 1)\n",
    "        x = self.batch_norm(x)\n",
    "        x = self.linear(x)\n",
    "        x = self.relu(x)\n",
    "        x = self.dropout(x)\n",
    "        x = self.out(x)\n",
    "        y = nn.functional.softmax(x, dim=1)\n",
    "        return y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LSTM_model(\n",
       "  (embeddings): Embedding(35, 64)\n",
       "  (lstm): LSTM(64, 32, num_layers=2, batch_first=True, bidirectional=True)\n",
       "  (linear): Linear(in_features=128, out_features=64, bias=True)\n",
       "  (batch_norm): BatchNorm1d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
       "  (relu): ReLU()\n",
       "  (dropout): Dropout(p=0.2, inplace=False)\n",
       "  (out): Linear(in_features=64, out_features=28, bias=True)\n",
       ")"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = LSTM_model(embedding_dim=64, \n",
    "                   hidden_dim=32, \n",
    "                   vocab_size=len(tokenizer.word2index) + 1, \n",
    "                   target_size=max_sequence_len)\n",
    "model.cuda()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "loss_function = nn.BCEWithLogitsLoss()\n",
    "optimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, model.parameters()), lr=1e-3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "BATCH_SIZE = 256\n",
    "\n",
    "train = MyDataset(torch.utils.data.TensorDataset(input_seq_train, output_seq_train))\n",
    "valid = MyDataset(torch.utils.data.TensorDataset(input_seq_val, output_seq_val))\n",
    "\n",
    "train_loader = torch.utils.data.DataLoader(train, batch_size=BATCH_SIZE, shuffle=True)\n",
    "valid_loader = torch.utils.data.DataLoader(valid, batch_size=BATCH_SIZE, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "# with torch.no_grad():\n",
    "#     (x_sample, y_sample, index) = train[:2]\n",
    "#     preds = model(x_sample.reshape(1, -1))\n",
    "#     print(preds.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### K-Fold training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import KFold"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_folds = 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "kf = KFold(n_splits=n_folds, shuffle=True, random_state=RANDOM_SEED)\n",
    "# kf.get_n_splits(X=input_seq, y=output_seq)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2205793,) (551449,)\n",
      "(2205793,) (551449,)\n",
      "(2205794,) (551448,)\n",
      "(2205794,) (551448,)\n",
      "(2205794,) (551448,)\n"
     ]
    }
   ],
   "source": [
    "for train_index, test_index in kf.split(X=input_seq, y=output_seq):\n",
    "    print(train_index.shape, test_index.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[21, 32, 13, ...,  0,  0,  0],\n",
       "       [21, 32, 13, ...,  0,  0,  0],\n",
       "       [21, 32, 13, ...,  0,  0,  0],\n",
       "       ...,\n",
       "       [ 7, 22, 32, ...,  0,  0,  0],\n",
       "       [ 7, 13, 13, ...,  0,  0,  0],\n",
       "       [ 7, 14,  3, ...,  0,  0,  0]])"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "input_seq"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training loop"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch [1/50]  progress = 100% \t loss=0.6935 \t acc=60.50% \n",
      "Epoch [1/50]  results:\t\t loss=0.6935\t acc=60.50%\t val_loss=0.6902\t val_acc=69.91%\t time=197.03s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [2/50]  progress = 100% \t loss=0.6895 \t acc=72.00% \n",
      "Epoch [2/50]  results:\t\t loss=0.6895\t acc=72.00%\t val_loss=0.6889\t val_acc=73.53%\t time=193.91s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [3/50]  progress = 100% \t loss=0.6887 \t acc=74.30% \n",
      "Epoch [3/50]  results:\t\t loss=0.6887\t acc=74.30%\t val_loss=0.6880\t val_acc=76.46%\t time=197.00s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [4/50]  progress = 100% \t loss=0.6880 \t acc=76.45% \n",
      "Epoch [4/50]  results:\t\t loss=0.6880\t acc=76.45%\t val_loss=0.6877\t val_acc=77.05%\t time=193.29s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [5/50]  progress = 100% \t loss=0.6877 \t acc=77.16% \n",
      "Epoch [5/50]  results:\t\t loss=0.6877\t acc=77.16%\t val_loss=0.6875\t val_acc=77.81%\t time=196.97s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [6/50]  progress = 100% \t loss=0.6875 \t acc=77.78% \n",
      "Epoch [6/50]  results:\t\t loss=0.6875\t acc=77.78%\t val_loss=0.6874\t val_acc=78.00%\t time=195.97s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [7/50]  progress = 100% \t loss=0.6874 \t acc=78.03% \n",
      "Epoch [7/50]  results:\t\t loss=0.6874\t acc=78.03%\t val_loss=0.6872\t val_acc=78.37%\t time=193.68s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [8/50]  progress = 100% \t loss=0.6873 \t acc=78.33% \n",
      "Epoch [8/50]  results:\t\t loss=0.6873\t acc=78.33%\t val_loss=0.6871\t val_acc=78.82%\t time=195.25s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [9/50]  progress = 100% \t loss=0.6871 \t acc=78.66% \n",
      "Epoch [9/50]  results:\t\t loss=0.6871\t acc=78.66%\t val_loss=0.6871\t val_acc=78.88%\t time=193.23s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [10/50]  progress = 100% \t loss=0.6871 \t acc=78.81% \n",
      "Epoch [10/50]  results:\t\t loss=0.6871\t acc=78.81%\t val_loss=0.6870\t val_acc=79.00%\t time=194.78s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [11/50]  progress = 100% \t loss=0.6870 \t acc=79.00% \n",
      "Epoch [11/50]  results:\t\t loss=0.6870\t acc=79.00%\t val_loss=0.6869\t val_acc=79.31%\t time=195.36s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [12/50]  progress = 100% \t loss=0.6870 \t acc=79.04% \n",
      "Epoch [12/50]  results:\t\t loss=0.6870\t acc=79.04%\t val_loss=0.6869\t val_acc=79.25%\t time=192.94s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [13/50]  progress = 100% \t loss=0.6870 \t acc=79.16% \n",
      "Epoch [13/50]  results:\t\t loss=0.6870\t acc=79.16%\t val_loss=0.6870\t val_acc=78.92%\t time=197.68s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [14/50]  progress = 100% \t loss=0.6870 \t acc=79.16% \n",
      "Epoch [14/50]  results:\t\t loss=0.6870\t acc=79.16%\t val_loss=0.6869\t val_acc=79.24%\t time=196.08s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [15/50]  progress = 100% \t loss=0.6870 \t acc=79.09% \n",
      "Epoch [15/50]  results:\t\t loss=0.6870\t acc=79.09%\t val_loss=0.6869\t val_acc=79.44%\t time=199.49s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [16/50]  progress = 100% \t loss=0.6869 \t acc=79.30% \n",
      "Epoch [16/50]  results:\t\t loss=0.6869\t acc=79.30%\t val_loss=0.6868\t val_acc=79.73%\t time=195.46s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [17/50]  progress = 100% \t loss=0.6869 \t acc=79.33% \n",
      "Epoch [17/50]  results:\t\t loss=0.6869\t acc=79.33%\t val_loss=0.6869\t val_acc=79.45%\t time=198.78s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [18/50]  progress = 100% \t loss=0.6868 \t acc=79.59% \n",
      "Epoch [18/50]  results:\t\t loss=0.6868\t acc=79.59%\t val_loss=0.6868\t val_acc=79.71%\t time=197.42s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [19/50]  progress = 100% \t loss=0.6869 \t acc=79.39% \n",
      "Epoch [19/50]  results:\t\t loss=0.6869\t acc=79.39%\t val_loss=0.6869\t val_acc=79.48%\t time=195.85s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [20/50]  progress = 100% \t loss=0.6869 \t acc=79.43% \n",
      "Epoch [20/50]  results:\t\t loss=0.6869\t acc=79.43%\t val_loss=0.6869\t val_acc=79.29%\t time=196.27s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [21/50]  progress = 100% \t loss=0.6869 \t acc=79.47% \n",
      "Epoch [21/50]  results:\t\t loss=0.6869\t acc=79.47%\t val_loss=0.6868\t val_acc=79.56%\t time=198.63s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [22/50]  progress = 100% \t loss=0.6869 \t acc=79.48% \n",
      "Epoch [22/50]  results:\t\t loss=0.6869\t acc=79.48%\t val_loss=0.6870\t val_acc=79.18%\t time=196.65s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [23/50]  progress = 100% \t loss=0.6869 \t acc=79.26% \n",
      "Epoch [23/50]  results:\t\t loss=0.6869\t acc=79.26%\t val_loss=0.6868\t val_acc=79.66%\t time=196.58s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [24/50]  progress = 100% \t loss=0.6868 \t acc=79.61% \n",
      "Epoch [24/50]  results:\t\t loss=0.6868\t acc=79.61%\t val_loss=0.6866\t val_acc=80.26%\t time=197.08s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [25/50]  progress = 100% \t loss=0.6866 \t acc=80.38% \n",
      "Epoch [25/50]  results:\t\t loss=0.6866\t acc=80.38%\t val_loss=0.6864\t val_acc=80.78%\t time=195.12s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [26/50]  progress = 100% \t loss=0.6865 \t acc=80.53% \n",
      "Epoch [26/50]  results:\t\t loss=0.6865\t acc=80.53%\t val_loss=0.6864\t val_acc=80.79%\t time=194.62s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [27/50]  progress = 100% \t loss=0.6865 \t acc=80.41% \n",
      "Epoch [27/50]  results:\t\t loss=0.6865\t acc=80.41%\t val_loss=0.6865\t val_acc=80.65%\t time=196.65s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [28/50]  progress = 100% \t loss=0.6865 \t acc=80.44% \n",
      "Epoch [28/50]  results:\t\t loss=0.6865\t acc=80.44%\t val_loss=0.6865\t val_acc=80.57%\t time=196.46s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [29/50]  progress = 100% \t loss=0.6866 \t acc=80.28% \n",
      "Epoch [29/50]  results:\t\t loss=0.6866\t acc=80.28%\t val_loss=0.6866\t val_acc=80.34%\t time=194.91s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [30/50]  progress = 100% \t loss=0.6866 \t acc=80.21% \n",
      "Epoch [30/50]  results:\t\t loss=0.6866\t acc=80.21%\t val_loss=0.6865\t val_acc=80.45%\t time=195.74s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [31/50]  progress = 100% \t loss=0.6867 \t acc=80.06% \n",
      "Epoch [31/50]  results:\t\t loss=0.6867\t acc=80.06%\t val_loss=0.6867\t val_acc=79.88%\t time=196.98s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [32/50]  progress = 100% \t loss=0.6866 \t acc=80.23% \n",
      "Epoch [32/50]  results:\t\t loss=0.6866\t acc=80.23%\t val_loss=0.6866\t val_acc=80.34%\t time=199.63s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [33/50]  progress = 100% \t loss=0.6866 \t acc=80.19% \n",
      "Epoch [33/50]  results:\t\t loss=0.6866\t acc=80.19%\t val_loss=0.6866\t val_acc=80.18%\t time=196.07s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [34/50]  progress = 100% \t loss=0.6866 \t acc=80.40% \n",
      "Epoch [34/50]  results:\t\t loss=0.6866\t acc=80.40%\t val_loss=0.6863\t val_acc=81.10%\t time=194.77s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [35/50]  progress = 100% \t loss=0.6864 \t acc=80.93% \n",
      "Epoch [35/50]  results:\t\t loss=0.6864\t acc=80.93%\t val_loss=0.6865\t val_acc=80.65%\t time=193.00s\n",
      "------------------------------------------------------------------------------\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch [36/50]  progress = 100% \t loss=0.6865 \t acc=80.59% \n",
      "Epoch [36/50]  results:\t\t loss=0.6865\t acc=80.59%\t val_loss=0.6865\t val_acc=80.71%\t time=194.77s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [37/50]  progress = 100% \t loss=0.6865 \t acc=80.58% \n",
      "Epoch [37/50]  results:\t\t loss=0.6865\t acc=80.58%\t val_loss=0.6863\t val_acc=81.24%\t time=197.09s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [38/50]  progress = 100% \t loss=0.6865 \t acc=80.63% \n",
      "Epoch [38/50]  results:\t\t loss=0.6865\t acc=80.63%\t val_loss=0.6866\t val_acc=80.22%\t time=194.56s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [40/50]  progress = 100% \t loss=0.6865 \t acc=80.53% \n",
      "Epoch [40/50]  results:\t\t loss=0.6865\t acc=80.53%\t val_loss=0.6864\t val_acc=80.92%\t time=192.52s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [41/50]  progress = 100% \t loss=0.6864 \t acc=80.94% \n",
      "Epoch [41/50]  results:\t\t loss=0.6864\t acc=80.94%\t val_loss=0.6863\t val_acc=81.05%\t time=197.55s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [42/50]  progress = 100% \t loss=0.6865 \t acc=80.63% \n",
      "Epoch [42/50]  results:\t\t loss=0.6865\t acc=80.63%\t val_loss=0.6865\t val_acc=80.49%\t time=197.87s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [43/50]  progress = 100% \t loss=0.6866 \t acc=80.43% \n",
      "Epoch [43/50]  results:\t\t loss=0.6866\t acc=80.43%\t val_loss=0.6867\t val_acc=80.11%\t time=197.02s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [44/50]  progress = 100% \t loss=0.6867 \t acc=80.09% \n",
      "Epoch [44/50]  results:\t\t loss=0.6867\t acc=80.09%\t val_loss=0.6868\t val_acc=79.83%\t time=194.02s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [45/50]  progress = 100% \t loss=0.6866 \t acc=80.37% \n",
      "Epoch [45/50]  results:\t\t loss=0.6866\t acc=80.37%\t val_loss=0.6865\t val_acc=80.48%\t time=194.51s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [46/50]  progress = 100% \t loss=0.6865 \t acc=80.65% \n",
      "Epoch [46/50]  results:\t\t loss=0.6865\t acc=80.65%\t val_loss=0.6864\t val_acc=80.87%\t time=194.30s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [47/50]  progress = 100% \t loss=0.6866 \t acc=80.39% \n",
      "Epoch [47/50]  results:\t\t loss=0.6866\t acc=80.39%\t val_loss=0.6867\t val_acc=80.02%\t time=195.24s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [48/50]  progress = 100% \t loss=0.6866 \t acc=80.35% \n",
      "Epoch [48/50]  results:\t\t loss=0.6866\t acc=80.35%\t val_loss=0.6865\t val_acc=80.65%\t time=196.49s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [49/50]  progress = 100% \t loss=0.6867 \t acc=80.14% \n",
      "Epoch [49/50]  results:\t\t loss=0.6867\t acc=80.14%\t val_loss=0.6865\t val_acc=80.57%\t time=201.75s\n",
      "------------------------------------------------------------------------------\n",
      "Epoch [50/50]  progress = 100% \t loss=0.6866 \t acc=80.29% \n",
      "Epoch [50/50]  results:\t\t loss=0.6866\t acc=80.29%\t val_loss=0.6865\t val_acc=80.55%\t time=196.69s\n",
      "------------------------------------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "n_epochs = 1\n",
    "history = {'train': {}, 'val': {}}\n",
    "teacher_forcing_ratio = 0.5\n",
    "\n",
    "for epoch in range(1, n_epochs + 1):\n",
    "    start_time = time.time()\n",
    "    \n",
    "    model.train()\n",
    "\n",
    "    avg_loss, total_loss, avg_acc, total_acc = 0., 0., 0., 0.\n",
    "    for i, (x_batch, y_batch, index) in enumerate(train_loader):\n",
    "        y_pred = model(x_batch)\n",
    "        loss = loss_function(y_pred, y_batch)\n",
    "        \n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        \n",
    "        equal = torch.eq(torch.argmax(y_pred, axis=1), torch.argmax(y_batch, axis=1))\n",
    "        batch_acc = int(equal.sum(-1)) / y_batch.shape[0]\n",
    "        batch_loss = loss.item()\n",
    "        \n",
    "        total_acc += batch_acc\n",
    "        total_loss += batch_loss\n",
    "        print(f\"\\rEpoch [{epoch}/{n_epochs}] \"\n",
    "              f\" progress = {round(i/len(train_loader)*100)}% \"\n",
    "              f\"\\t loss={total_loss / (i + 1):.4f} \"\n",
    "              f\"\\t acc={total_acc / (i + 1) * 100:.2f}% \", end='')\n",
    "    avg_loss = total_loss / len(train_loader)\n",
    "    avg_acc = total_acc / len(train_loader)\n",
    "    history['train']['loss'] = history.get('train', {}).get('loss', []) + [avg_loss]\n",
    "    history['train']['accuracy'] = history.get('train', {}).get('accuracy', []) + [avg_acc]\n",
    "        \n",
    "    model.eval()\n",
    "\n",
    "    \n",
    "    avg_val_loss, total_val_loss, avg_val_acc, total_val_acc = 0., 0., 0., 0.\n",
    "    for i, (x_batch, y_batch, index) in enumerate(valid_loader):\n",
    "        y_pred = model(x_batch).detach()\n",
    "        val_loss = loss_function(y_pred, y_batch)\n",
    "        \n",
    "        equal = torch.eq(torch.argmax(y_pred, axis=1), torch.argmax(y_batch, axis=1))\n",
    "        batch_val_acc = int(equal.sum(-1)) / y_batch.shape[0]\n",
    "        batch_val_loss = val_loss.item()\n",
    "        \n",
    "        total_val_acc += batch_val_acc\n",
    "        total_val_loss += batch_val_loss\n",
    "    avg_val_loss = total_val_loss / len(valid_loader)\n",
    "    avg_val_acc = total_val_acc / len(valid_loader)\n",
    "    history['val']['loss'] = history.get('val', {}).get('loss', []) + [avg_val_loss]\n",
    "    history['val']['accuracy'] = history.get('val', {}).get('accuracy', []) + [avg_val_acc]\n",
    "    \n",
    "    elapsed_time = time.time() - start_time \n",
    "    print(f\"\\nEpoch [{epoch}/{n_epochs}]  results:\"\n",
    "          f\"\\t\\t loss={avg_loss:.4f}\"\n",
    "          f\"\\t acc={avg_acc * 100:.2f}%\"\n",
    "          f\"\\t val_loss={avg_val_loss:.4f}\"\n",
    "          f\"\\t val_acc={avg_val_acc * 100:.2f}%\"\n",
    "          f\"\\t time={elapsed_time:.2f}s\")\n",
    "    print(\"-\"*78)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(12, 8))\n",
    "\n",
    "plt.plot(history['train']['loss'])\n",
    "plt.plot(history['val']['loss'])\n",
    "plt.title('model loss')\n",
    "plt.ylabel('val')\n",
    "plt.xlabel('epoch')\n",
    "plt.xticks(np.arange(len(history['train']['loss'])), np.arange(1, len(history['train']['loss']) + 1))\n",
    "plt.legend(['train', 'val'], loc='upper right')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(12, 8))\n",
    "\n",
    "plt.plot(history['train']['accuracy'])\n",
    "plt.plot(history['val']['accuracy'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.xticks(np.arange(len(history['train']['accuracy'])), np.arange(1, len(history['train']['accuracy']) + 1))\n",
    "plt.legend(['train', 'val'], loc='lower right')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Predict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [],
   "source": [
    "def predict(words, mode='stress'):\n",
    "    tokens = pad_sequence(tokenizer.transform(words))\n",
    "    sequences = torch.tensor(tokens, dtype=torch.long).cuda()\n",
    "    preds = model(sequences)\n",
    "    indeces = torch.argmax(preds, axis=1)\n",
    "    if mode == 'stress':\n",
    "        return [word[:index+1] + chr(769) + word[index+1:] for word, index in zip(words, indeces)]\n",
    "    elif mode == 'asterisk':\n",
    "        return [word[:index+1] + \"*\" + word[index+1:] for word, index in zip(words, indeces)]\n",
    "    else:\n",
    "        raise ValueError(f\"Wrong `mode`={mode}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [],
   "source": [
    "words = [\"словотворення\", \"архаїчний\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['словотво́рення', 'арха́їчний']"
      ]
     },
     "execution_count": 115,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predict(words)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Environment (conda_pytorch_latest_p37)",
   "language": "python",
   "name": "conda_pytorch_latest_p37"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}