version_3.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tłumacz polsko-bułgarski oparty o architekturę transformera"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Biblioteki i ustawienia"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import warnings\n",
    "\n",
    "from torchtext.data import Field, BucketIterator\n",
    "from tqdm import tqdm\n",
    "import matplotlib.pyplot as plt\n",
    "import spacy\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "\n",
    "from utils import get_dataset_splits, random_eval, get_batch_bleu, print_dataset_info, get_subset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "warnings.filterwarnings(\"ignore\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Python 3.12.3\n"
     ]
    }
   ],
   "source": [
    "!python --version"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "device(type='cpu')"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "DEVICE = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "DEVICE"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# training hyperparameters\n",
    "LEARNING_RATE = 1e-4\n",
    "NUM_EPOCHS = 50\n",
    "BATCH_SIZE = 32\n",
    "\n",
    "# model hyperparameters\n",
    "EMBEDDING_SIZE = 128\n",
    "NUM_ENC_LAYERS = 8\n",
    "NUM_DEC_LAYERS = 8\n",
    "TOKEN_LIMIT = 32\n",
    "DIM_FEEDFORWARD = 512\n",
    "NUM_HEADS = 16\n",
    "DROPOUT = 0.1\n",
    "\n",
    "# other\n",
    "ENABLE_TRAINING = True\n",
    "SUBSET_RATIO = 0.05\n",
    "CORPUS_PATH = \"./raw_data/\"\n",
    "TRAIN_DATASET_PATH = f\"./datasets/train_dataset_{TOKEN_LIMIT}.pkl\"\n",
    "TEST_DATASET_PATH = f\"./datasets/test_dataset_{TOKEN_LIMIT}.pkl\"\n",
    "MODEL_PATH = f\"./serialized_models/model_v3\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Tokenizatory"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Zaprojektowałem dwa tokenizatory, osobno dla każdego języka. Są one oparte o modele `pl_core_news_lg` i `uk_core_news_lg` z biblioteki `spacy`. Jak widać dla języka bułgarskiego użyty został model przeznaczony dla języka ukraińskiego. Jest to spowodowane brakiem modelu dla języka bułgarskiego w bibliotece `spacy` oraz wysokim podobieństwem tych języków. Zakładam, że nie powinno to negatywnie wpłynąć na jakość tłumaczenia."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "nlp_pl = spacy.load(\"pl_core_news_lg\")\n",
    "nlp_bg = spacy.load(\"uk_core_news_lg\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tokenize_pl(text):\n",
    "    return [token.text for token in nlp_pl.tokenizer(text)]\n",
    "\n",
    "def tokenize_bg(text):\n",
    "    return [token.text for token in nlp_bg.tokenizer(text)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "polish = Field(\n",
    "    tokenize=tokenize_pl, lower=True,\n",
    "    init_token=\"<bos>\", eos_token=\"<eos>\"\n",
    ")\n",
    "\n",
    "bulgarian = Field(\n",
    "    tokenize=tokenize_bg, lower=True,\n",
    "    init_token=\"<bos>\", eos_token=\"<eos>\"\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Wczytanie danych"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading dataset from: ./datasets/train_dataset_32.pkl\n",
      "Loading dataset from: ./datasets/test_dataset_32.pkl\n",
      "Done.\n",
      "\n",
      "<|TRAIN SET|>\n",
      "number of examples: 217228\n",
      "number of source tokens: 755338\n",
      "number of target tokens: 700961\n",
      "\n",
      "\n",
      "<|TEST SET|>\n",
      "number of examples: 24080\n",
      "number of source tokens: 83504\n",
      "number of target tokens: 77626\n",
      "\n"
     ]
    }
   ],
   "source": [
    "fields = [(\"src\", polish), (\"tgt\", bulgarian)]\n",
    "train_dataset, test_dataset = get_dataset_splits(\n",
    "    train_dataset_path=TRAIN_DATASET_PATH,\n",
    "    test_dataset_path=TEST_DATASET_PATH,\n",
    "    corpus_path=CORPUS_PATH,\n",
    "    max_len=TOKEN_LIMIT,\n",
    "    fields=fields,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "<|TRAIN SET|>\n",
      "number of examples: 10861\n",
      "number of source tokens: 37635\n",
      "number of target tokens: 35038\n",
      "\n",
      "\n",
      "<|TEST SET|>\n",
      "number of examples: 1204\n",
      "number of source tokens: 4273\n",
      "number of target tokens: 3942\n",
      "\n"
     ]
    }
   ],
   "source": [
    "train_dataset = get_subset(train_dataset, SUBSET_RATIO)\n",
    "test_dataset = get_subset(test_dataset, SUBSET_RATIO)\n",
    "print_dataset_info(train_dataset, \"train\")\n",
    "print_dataset_info(test_dataset, \"test\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "polish.build_vocab(train_dataset, max_size=10000, min_freq=2)\n",
    "bulgarian.build_vocab(train_dataset, max_size=10000, min_freq=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_iterator, test_iterator = BucketIterator.splits(\n",
    "    (train_dataset, test_dataset),\n",
    "    batch_size=BATCH_SIZE,\n",
    "    sort_within_batch=True,\n",
    "    sort_key = lambda x: len(x.src),\n",
    "    device=DEVICE\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Architektura modelu"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "class PL2BG(nn.Module):\n",
    "    def __init__(\n",
    "        self,\n",
    "        embedding_size,\n",
    "        src_vocab_size,\n",
    "        tgt_vocab_size,\n",
    "        src_pad_idx,\n",
    "        num_heads,\n",
    "        num_encoder_layers,\n",
    "        num_decoder_layers,\n",
    "        dim_feedforward,\n",
    "        dropout,\n",
    "        token_limit,\n",
    "        device,\n",
    "    ):\n",
    "        super(PL2BG, self).__init__()\n",
    "        self.src_word_embedding = nn.Embedding(src_vocab_size, embedding_size)\n",
    "        self.tgt_word_embedding = nn.Embedding(tgt_vocab_size, embedding_size)\n",
    "        self.position_embedding = nn.Embedding(token_limit, embedding_size)\n",
    "        self.device = device\n",
    "        self.transformer = nn.Transformer(\n",
    "            d_model=embedding_size,\n",
    "            nhead=num_heads,\n",
    "            num_encoder_layers=num_encoder_layers,\n",
    "            num_decoder_layers=num_decoder_layers,\n",
    "            dim_feedforward=dim_feedforward,\n",
    "            dropout=dropout,\n",
    "        )\n",
    "        self.fc_1 = nn.Linear(embedding_size, 512)\n",
    "        self.fc_2 = nn.Linear(512, 512)\n",
    "        self.fc_3 = nn.Linear(512, 1024)\n",
    "        self.fc_4 = nn.Linear(1024, embedding_size)\n",
    "        self.unembedding = nn.Linear(embedding_size, tgt_vocab_size)\n",
    "        self.dropout = nn.Dropout(dropout)\n",
    "        self.src_pad_idx = src_pad_idx\n",
    "\n",
    "    def make_src_mask(self, src):\n",
    "        return src.transpose(0, 1) == self.src_pad_idx\n",
    "\n",
    "    def forward(self, src, tgt):\n",
    "        src_seq_len, n = src.shape\n",
    "        tgt_seq_len = tgt.shape[0]\n",
    "        src_positions = torch.arange(0, src_seq_len) \\\n",
    "            .unsqueeze(1).expand(src_seq_len, n).to(self.device)\n",
    "        tgt_positions = torch.arange(0, tgt_seq_len) \\\n",
    "            .unsqueeze(1).expand(tgt_seq_len, n).to(self.device)\n",
    "        embedded_src = self.dropout(\n",
    "            self.src_word_embedding(src)\n",
    "            + self.position_embedding(src_positions)\n",
    "        )\n",
    "        embedded_tgt = self.dropout(\n",
    "            self.tgt_word_embedding(tgt)\n",
    "            + self.position_embedding(tgt_positions)\n",
    "        )\n",
    "        src_padding_mask = self.make_src_mask(src)\n",
    "        tgt_mask = self.transformer \\\n",
    "            .generate_square_subsequent_mask(tgt_seq_len).to(self.device)\n",
    "        output = self.transformer(\n",
    "            embedded_src,\n",
    "            embedded_tgt,\n",
    "            src_key_padding_mask=src_padding_mask,\n",
    "            tgt_mask=tgt_mask\n",
    "        )\n",
    "        output = F.relu(self.dropout(self.fc_1(output)))\n",
    "        output = F.relu(self.dropout(self.fc_2(output)))\n",
    "        output = F.relu(self.dropout(self.fc_3(output)))\n",
    "        output = F.relu(self.dropout(self.fc_4(output)))\n",
    "        return self.unembedding(output)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Inicjalizacja i trening"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "tgt_pad_idx = bulgarian.vocab.stoi[\"<pad>\"]\n",
    "src_pad_idx = polish.vocab.stoi[\"<pad>\"]\n",
    "tgt_vocab_size = len(bulgarian.vocab)\n",
    "src_vocab_size = len(polish.vocab)\n",
    "\n",
    "model = PL2BG(\n",
    "    num_encoder_layers=NUM_ENC_LAYERS,\n",
    "    num_decoder_layers=NUM_DEC_LAYERS,\n",
    "    dim_feedforward=DIM_FEEDFORWARD,\n",
    "    embedding_size=EMBEDDING_SIZE,\n",
    "    src_vocab_size=src_vocab_size,\n",
    "    tgt_vocab_size=tgt_vocab_size,\n",
    "    src_pad_idx=src_pad_idx,\n",
    "    token_limit=TOKEN_LIMIT,\n",
    "    num_heads=NUM_HEADS,\n",
    "    dropout=DROPOUT,\n",
    "    device=DEVICE,\n",
    ").to(DEVICE)\n",
    "\n",
    "optimizer = optim.Adam(model.parameters(), lr=LEARNING_RATE)\n",
    "criterion = nn.CrossEntropyLoss(ignore_index=tgt_pad_idx)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "layers = [layer.view(1, -1).squeeze().shape[0] for layer in model.parameters()]\n",
    "\n",
    "model_metadata = {\n",
    "    \"training_hyperparameters\": {\n",
    "        \"learning_rate\": LEARNING_RATE,\n",
    "        \"num_epochs\": NUM_EPOCHS,\n",
    "        \"batch_size\": BATCH_SIZE,\n",
    "    },\n",
    "    \"model_hyperparameters\": {\n",
    "        \"embedding_size\": EMBEDDING_SIZE,\n",
    "        \"num_enc_layers\": NUM_ENC_LAYERS,\n",
    "        \"num_dec_layers\": NUM_DEC_LAYERS,\n",
    "        \"dim_feedforward\": DIM_FEEDFORWARD,\n",
    "        \"token_limit\": TOKEN_LIMIT,\n",
    "        \"num_heads\": NUM_HEADS,\n",
    "        \"dropout\": DROPOUT,\n",
    "        \"device\": str(DEVICE),\n",
    "    },\n",
    "    \"model_architecture\": {\n",
    "        \"num_params\": sum(layers),\n",
    "        \"num_layers\": len(layers),\n",
    "    },\n",
    "    \"other\": {\n",
    "        \"subset_ratio\": SUBSET_RATIO,\n",
    "        \"model_path\": MODEL_PATH,\n",
    "        \"train_dataset_path\": TRAIN_DATASET_PATH,\n",
    "        \"test_dataset_path\": TEST_DATASET_PATH,\n",
    "        \"corpus_path\": CORPUS_PATH,\n",
    "    }\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "EPOCH [01 / 50]: 100%|██████████| 340/340 [03:00<00:00,  1.89it/s, loss=4.783747\ttrain_bleu=19.0721\ttest_bleu=0.0000]\n",
      "EPOCH [02 / 50]: 100%|██████████| 340/340 [04:01<00:00,  1.41it/s, loss=4.691220\ttrain_bleu=16.3461\ttest_bleu=12.5000]\n",
      "EPOCH [03 / 50]: 100%|██████████| 340/340 [05:16<00:00,  1.07it/s, loss=3.259279\ttrain_bleu=24.9667\ttest_bleu=20.7841]\n",
      "EPOCH [04 / 50]: 100%|██████████| 340/340 [06:11<00:00,  1.09s/it, loss=4.215424\ttrain_bleu=14.0160\ttest_bleu=27.3812]\n",
      "EPOCH [05 / 50]: 100%|██████████| 340/340 [05:51<00:00,  1.03s/it, loss=2.091084\ttrain_bleu=39.7392\ttest_bleu=17.0904]\n",
      "EPOCH [06 / 50]: 100%|██████████| 340/340 [05:25<00:00,  1.04it/s, loss=3.818819\ttrain_bleu=19.6245\ttest_bleu=45.1787]\n",
      "EPOCH [07 / 50]: 100%|██████████| 340/340 [05:11<00:00,  1.09it/s, loss=2.609610\ttrain_bleu=45.4479\ttest_bleu=26.5926]\n",
      "EPOCH [08 / 50]: 100%|██████████| 340/340 [05:17<00:00,  1.07it/s, loss=1.619935\ttrain_bleu=52.4976\ttest_bleu=37.5785]\n",
      "EPOCH [09 / 50]: 100%|██████████| 340/340 [05:28<00:00,  1.04it/s, loss=3.098684\ttrain_bleu=16.6526\ttest_bleu=37.2558]\n",
      "EPOCH [10 / 50]: 100%|██████████| 340/340 [05:13<00:00,  1.09it/s, loss=3.462913\ttrain_bleu=29.4168\ttest_bleu=49.0096]\n",
      "EPOCH [11 / 50]: 100%|██████████| 340/340 [04:45<00:00,  1.19it/s, loss=1.466907\ttrain_bleu=65.1701\ttest_bleu=26.8823]\n",
      "EPOCH [12 / 50]: 100%|██████████| 340/340 [04:49<00:00,  1.17it/s, loss=1.523578\ttrain_bleu=59.0624\ttest_bleu=29.7266]\n",
      "EPOCH [13 / 50]: 100%|██████████| 340/340 [04:45<00:00,  1.19it/s, loss=1.490764\ttrain_bleu=61.0359\ttest_bleu=25.3540]\n",
      "EPOCH [14 / 50]: 100%|██████████| 340/340 [04:46<00:00,  1.19it/s, loss=1.243860\ttrain_bleu=72.9842\ttest_bleu=47.8174]\n",
      "EPOCH [15 / 50]: 100%|██████████| 340/340 [04:45<00:00,  1.19it/s, loss=0.690396\ttrain_bleu=78.3302\ttest_bleu=31.3700]\n",
      "EPOCH [16 / 50]: 100%|██████████| 340/340 [04:45<00:00,  1.19it/s, loss=1.417640\ttrain_bleu=56.3714\ttest_bleu=36.5836]\n",
      "EPOCH [17 / 50]: 100%|██████████| 340/340 [04:44<00:00,  1.19it/s, loss=1.007938\ttrain_bleu=68.7500\ttest_bleu=36.6049]\n",
      "EPOCH [18 / 50]: 100%|██████████| 340/340 [04:44<00:00,  1.19it/s, loss=1.942695\ttrain_bleu=27.7825\ttest_bleu=29.2124]\n",
      "EPOCH [19 / 50]: 100%|██████████| 340/340 [04:48<00:00,  1.18it/s, loss=1.046522\ttrain_bleu=68.6164\ttest_bleu=39.8246]\n",
      "EPOCH [20 / 50]: 100%|██████████| 340/340 [04:46<00:00,  1.19it/s, loss=0.924893\ttrain_bleu=77.5860\ttest_bleu=47.0404]\n",
      "EPOCH [21 / 50]: 100%|██████████| 340/340 [04:51<00:00,  1.17it/s, loss=1.434916\ttrain_bleu=48.1875\ttest_bleu=55.0194]\n",
      "EPOCH [22 / 50]: 100%|██████████| 340/340 [04:48<00:00,  1.18it/s, loss=1.653855\ttrain_bleu=26.1692\ttest_bleu=49.2809]\n",
      "EPOCH [23 / 50]: 100%|██████████| 340/340 [04:49<00:00,  1.18it/s, loss=0.373079\ttrain_bleu=71.8991\ttest_bleu=31.0150]\n",
      "EPOCH [24 / 50]: 100%|██████████| 340/340 [04:51<00:00,  1.17it/s, loss=0.943853\ttrain_bleu=30.4919\ttest_bleu=44.9551]\n",
      "EPOCH [25 / 50]: 100%|██████████| 340/340 [04:51<00:00,  1.17it/s, loss=0.582896\ttrain_bleu=66.3925\ttest_bleu=52.2704]\n",
      "EPOCH [26 / 50]: 100%|██████████| 340/340 [04:50<00:00,  1.17it/s, loss=0.803698\ttrain_bleu=63.7918\ttest_bleu=55.3284]\n",
      "EPOCH [27 / 50]: 100%|██████████| 340/340 [04:54<00:00,  1.15it/s, loss=0.793914\ttrain_bleu=67.6454\ttest_bleu=41.8504]\n",
      "EPOCH [28 / 50]: 100%|██████████| 340/340 [04:50<00:00,  1.17it/s, loss=0.502682\ttrain_bleu=82.3996\ttest_bleu=60.5229]\n",
      "EPOCH [29 / 50]: 100%|██████████| 340/340 [04:51<00:00,  1.17it/s, loss=0.941663\ttrain_bleu=54.7161\ttest_bleu=37.5858] \n",
      "EPOCH [30 / 50]: 100%|██████████| 340/340 [04:51<00:00,  1.16it/s, loss=1.184362\ttrain_bleu=46.3796\ttest_bleu=49.7780] \n",
      "EPOCH [31 / 50]: 100%|██████████| 340/340 [04:53<00:00,  1.16it/s, loss=0.431991\ttrain_bleu=83.3842\ttest_bleu=49.1408]\n",
      "EPOCH [32 / 50]: 100%|██████████| 340/340 [04:54<00:00,  1.16it/s, loss=0.421074\ttrain_bleu=81.4623\ttest_bleu=43.4694]\n",
      "EPOCH [33 / 50]: 100%|██████████| 340/340 [04:50<00:00,  1.17it/s, loss=0.369121\ttrain_bleu=90.5340\ttest_bleu=31.8209] \n",
      "EPOCH [34 / 50]: 100%|██████████| 340/340 [04:51<00:00,  1.17it/s, loss=0.619773\ttrain_bleu=78.4219\ttest_bleu=31.0960] \n",
      "EPOCH [35 / 50]: 100%|██████████| 340/340 [04:53<00:00,  1.16it/s, loss=0.406057\ttrain_bleu=85.3192\ttest_bleu=38.6718]\n",
      "EPOCH [36 / 50]: 100%|██████████| 340/340 [04:52<00:00,  1.16it/s, loss=0.497652\ttrain_bleu=84.4628\ttest_bleu=39.3415]\n",
      "EPOCH [37 / 50]: 100%|██████████| 340/340 [04:51<00:00,  1.17it/s, loss=0.700633\ttrain_bleu=61.9412\ttest_bleu=52.3306]\n",
      "EPOCH [38 / 50]: 100%|██████████| 340/340 [04:51<00:00,  1.17it/s, loss=0.329417\ttrain_bleu=82.5218\ttest_bleu=42.2826]\n",
      "EPOCH [39 / 50]: 100%|██████████| 340/340 [04:53<00:00,  1.16it/s, loss=0.179926\ttrain_bleu=93.6889\ttest_bleu=47.5314] \n",
      "EPOCH [40 / 50]: 100%|██████████| 340/340 [04:50<00:00,  1.17it/s, loss=0.530898\ttrain_bleu=55.0151\ttest_bleu=50.6293]\n",
      "EPOCH [41 / 50]: 100%|██████████| 340/340 [04:52<00:00,  1.16it/s, loss=0.261802\ttrain_bleu=79.0397\ttest_bleu=49.5596] \n",
      "EPOCH [42 / 50]: 100%|██████████| 340/340 [04:54<00:00,  1.16it/s, loss=0.240029\ttrain_bleu=90.9759\ttest_bleu=44.1880] \n",
      "EPOCH [43 / 50]: 100%|██████████| 340/340 [04:52<00:00,  1.16it/s, loss=0.636346\ttrain_bleu=79.7074\ttest_bleu=53.6854]\n",
      "EPOCH [44 / 50]: 100%|██████████| 340/340 [04:53<00:00,  1.16it/s, loss=0.409833\ttrain_bleu=76.6959\ttest_bleu=44.3589]\n",
      "EPOCH [45 / 50]: 100%|██████████| 340/340 [04:53<00:00,  1.16it/s, loss=0.240206\ttrain_bleu=76.6929\ttest_bleu=46.3270] \n",
      "EPOCH [46 / 50]: 100%|██████████| 340/340 [04:51<00:00,  1.17it/s, loss=0.136489\ttrain_bleu=94.6765\ttest_bleu=50.3785]\n",
      "EPOCH [47 / 50]: 100%|██████████| 340/340 [04:53<00:00,  1.16it/s, loss=0.340712\ttrain_bleu=75.6676\ttest_bleu=49.5405] \n",
      "EPOCH [48 / 50]: 100%|██████████| 340/340 [04:53<00:00,  1.16it/s, loss=0.299092\ttrain_bleu=90.6250\ttest_bleu=53.1208] \n",
      "EPOCH [49 / 50]: 100%|██████████| 340/340 [05:11<00:00,  1.09it/s, loss=0.421330\ttrain_bleu=75.3915\ttest_bleu=46.0646]\n",
      "EPOCH [50 / 50]: 100%|██████████| 340/340 [05:12<00:00,  1.09it/s, loss=0.293843\ttrain_bleu=86.4688\ttest_bleu=51.4899] "
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Saving model...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "if ENABLE_TRAINING:\n",
    "    print(\"Training...\")\n",
    "\n",
    "    metrics = {\n",
    "        \"loss\": [],\n",
    "        \"train_bleu\": [],\n",
    "        \"test_bleu\": [],\n",
    "    }\n",
    "\n",
    "    for epoch in range(NUM_EPOCHS):\n",
    "\n",
    "        model.train()\n",
    "        mean_train_bleu = 0\n",
    "        mean_test_bleu = 0\n",
    "        mean_loss = 0\n",
    "\n",
    "        epoch_num = f\"[{epoch + 1:0{len(str(NUM_EPOCHS))}d} / {NUM_EPOCHS}]\"\n",
    "        iterator = tqdm(train_iterator, desc=f\"EPOCH {epoch_num}\")\n",
    "        for idx, batch in enumerate(iterator):\n",
    "            src = batch.src.to(DEVICE)\n",
    "            tgt = batch.tgt.to(DEVICE)\n",
    "\n",
    "            out = model(src, tgt[:-1])\n",
    "            out = out.reshape(-1, out.shape[2])\n",
    "            _tgt = tgt[1:].reshape(-1)\n",
    "            optimizer.zero_grad()\n",
    "\n",
    "            loss = criterion(out, _tgt)\n",
    "\n",
    "            current_loss = loss.item()\n",
    "            current_train_bleu = get_batch_bleu(model, src, tgt, bulgarian)\n",
    "            current_test_bleu = random_eval(\n",
    "                model=model,\n",
    "                n_examples=32,\n",
    "                dataset=test_dataset,\n",
    "                src_lang=polish,\n",
    "                tgt_lang=bulgarian,\n",
    "                device=DEVICE,\n",
    "                token_limit=TOKEN_LIMIT,\n",
    "            )[\"score\"]\n",
    "\n",
    "            mean_loss += current_loss / len(train_iterator)\n",
    "            mean_train_bleu += current_train_bleu / len(train_iterator)\n",
    "            mean_test_bleu += current_test_bleu / len(train_iterator)\n",
    "\n",
    "            if idx == len(train_iterator):\n",
    "                iterator.set_postfix_str(\n",
    "                    f\"loss={mean_loss:.4f}\\t\"\n",
    "                    f\"train_bleu={mean_train_bleu:.2f}\\t\"\n",
    "                    f\"test_bleu={mean_test_bleu:.2f}\"\n",
    "                )\n",
    "            else:\n",
    "                iterator.set_postfix_str(\n",
    "                    f\"loss={current_loss:.6f}\\t\"\n",
    "                    f\"train_bleu={current_train_bleu:.4f}\\t\"\n",
    "                    f\"test_bleu={current_test_bleu:.4f}\"\n",
    "                )\n",
    "\n",
    "            loss.backward()\n",
    "            torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=1)\n",
    "            optimizer.step()\n",
    "\n",
    "        metrics[\"loss\"].append(mean_loss)\n",
    "        metrics[\"train_bleu\"].append(mean_train_bleu)\n",
    "        metrics[\"test_bleu\"].append(mean_test_bleu)\n",
    "\n",
    "    print(\"Saving model...\")\n",
    "    json.dump(model_metadata, open(MODEL_PATH + \".json\", \"w\"))\n",
    "    torch.save(model.state_dict(), MODEL_PATH)\n",
    "\n",
    "else:\n",
    "    print(\"Loading pretrained model...\")\n",
    "    print(model.load_state_dict(torch.load(MODEL_PATH)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "json.dump(model_metadata, open(MODEL_PATH + \".json\", \"w\"))\n",
    "torch.save(model.state_dict(), MODEL_PATH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1500x500 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "if ENABLE_TRAINING:\n",
    "    fig, ax = plt.subplots(1, 2, figsize=(15, 5))\n",
    "    ax[0].plot(metrics[\"train_bleu\"], c=\"orange\")\n",
    "    ax[0].plot(metrics[\"test_bleu\"], c=\"green\")\n",
    "    ax[0].legend([\"Train BLEU\", \"Test BLEU\"])\n",
    "    ax[0].set_ylabel(\"Mean BLEU score\")\n",
    "    ax[0].set_xlabel(\"Epoch\")\n",
    "    ax[0].set_title(\"BLEU score over epochs\")\n",
    "    ax[1].plot(metrics[\"loss\"], c=\"orange\")\n",
    "    ax[1].set_ylabel(\"Mean loss\")\n",
    "    ax[1].set_xlabel(\"Epoch\")\n",
    "    ax[1].set_title(\"Loss over epochs\")\n",
    "    plt.savefig(f\"{MODEL_PATH}.png\", dpi=300)\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Ewaluacja"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "evaluation = random_eval(\n",
    "    model=model,\n",
    "    n_examples=32,\n",
    "    dataset=test_dataset,\n",
    "    src_lang=polish,\n",
    "    tgt_lang=bulgarian,\n",
    "    device=DEVICE,\n",
    "    token_limit=TOKEN_LIMIT,\n",
    ")\n",
    "for example in evaluation[\"examples\"]:\n",
    "    print(example, file=open(f\"{MODEL_PATH}.txt\", \"a\"))\n",
    "print(f\"BLEU: {evaluation[\"score\"]}\", file=open(f\"{MODEL_PATH}.txt\", \"a\"))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "nlp_proj",
   "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": "3.12.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}