{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inference"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import(\n",
    "    EncoderDecoderModel,\n",
    "    PreTrainedTokenizerFast,\n",
    "    # XLMRobertaTokenizerFast,\n",
    "    BertJapaneseTokenizer,\n",
    "    BertTokenizerFast,\n",
    ")\n",
    "\n",
    "import torch\n",
    "import csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "The tokenizer class you load from this checkpoint is not the same type as the class this function is called from. It may result in unexpected tokenization. \n",
      "The tokenizer class you load from this checkpoint is 'GPT2Tokenizer'. \n",
      "The class this function is called from is 'PreTrainedTokenizerFast'.\n"
     ]
    }
   ],
   "source": [
    "encoder_model_name = \"cl-tohoku/bert-base-japanese-v2\"\n",
    "decoder_model_name = \"skt/kogpt2-base-v2\"\n",
    "\n",
    "src_tokenizer = BertJapaneseTokenizer.from_pretrained(encoder_model_name)\n",
    "trg_tokenizer = PreTrainedTokenizerFast.from_pretrained(decoder_model_name)\n",
    "model = EncoderDecoderModel.from_pretrained(\"./dump/best_model\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'길가메시 토벌전'"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "text = \"ギルガメッシュ討伐戦\"\n",
    "# text = \"ギルガメッシュ討伐戦に行ってきます。一緒に行きましょうか？\"\n",
    "\n",
    "def translate(text_src):\n",
    "    embeddings = src_tokenizer(text_src, return_attention_mask=False, return_token_type_ids=False, return_tensors='pt')\n",
    "    embeddings = {k: v for k, v in embeddings.items()}\n",
    "    output = model.generate(**embeddings)[0, 1:-1]\n",
    "    text_trg = trg_tokenizer.decode(output.cpu())\n",
    "    return text_trg\n",
    "\n",
    "print(translate(text))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction\n",
    "smoothie = SmoothingFunction().method4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Testing:   0%|          | 0/267 [00:00<?, ?it/s]/home/tikim/.local/lib/python3.8/site-packages/transformers/generation/utils.py:1288: UserWarning: Using `max_length`'s default (20) to control the generation length. This behaviour is deprecated and will be removed from the config in v5 of Transformers -- we recommend using `max_new_tokens` to control the maximum length of the generation.\n",
      "  warnings.warn(\n",
      "Testing: 100%|██████████| 267/267 [01:01<00:00,  4.34it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Bleu score: 0.9619225967540574\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "from tqdm import tqdm\n",
    "from statistics import mean\n",
    "\n",
    "bleu = []\n",
    "f1 = []\n",
    "\n",
    "DATA_ROOT = './output'\n",
    "FILE_JP_KO_TEST = 'ja_ko_test.csv'\n",
    "FILE_FFAC_TEST = 'ffac_test.csv'\n",
    "\n",
    "with torch.no_grad(), open(f'{DATA_ROOT}/{FILE_FFAC_TEST}', 'r') as fd:\n",
    "# with torch.no_grad(), open(f'{DATA_ROOT}/{FILE_JP_KO_TEST}', 'r') as fd:\n",
    "    reader = csv.reader(fd)\n",
    "    next(reader)\n",
    "    datas = [row for row in reader]    \n",
    "\n",
    "    for data in tqdm(datas, \"Testing\"):\n",
    "        input, label = data\n",
    "        embeddings = src_tokenizer(input, return_attention_mask=False, return_token_type_ids=False, return_tensors='pt')\n",
    "        embeddings = {k: v for k, v in embeddings.items()}\n",
    "        with torch.no_grad():\n",
    "            output = model.generate(**embeddings)[0, 1:-1]\n",
    "        preds = trg_tokenizer.decode(output.cpu())\n",
    "\n",
    "        bleu.append(sentence_bleu([label.split()], preds.split(), weights=[1,0,0,0], smoothing_function=smoothie))\n",
    "\n",
    "print(f\"Bleu score: {mean(bleu)}\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.10"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}