\uc548\ub155\ud558\uc138\uc694! \uc624\ub298\uc740 \uc790\uc5f0\uc5b4 \ucc98\ub9ac \ubd84\uc57c\uc5d0\uc11c \ub110\ub9ac \uc0ac\uc6a9\ub418\ub294 \ud5c8\uae45\ud398\uc774\uc2a4(Hugging Face)\uc758 \ud2b8\ub79c\uc2a4\ud3ec\uba38(Transformers) \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \ud65c\uc6a9\ud558\uc5ec IMDB \ub370\uc774\ud130\uc14b\uc744 \uac00\uc9c0\uace0 \uac10\uc815 \ubd84\uc11d \ubaa8\ub378\uc744 \ud559\uc2b5\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc54c\uc544\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \ub370\uc774\ud130 \uc900\ube44\ubd80\ud130 \ubaa8\ub378 \ud559\uc2b5, \ud3c9\uac00 \ubc0f \uc608\uce21\uae4c\uc9c0 \uc804 \uacfc\uc815\uc744 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
IMDB \ub370\uc774\ud130\uc14b\uc740 \uc601\ud654 \ub9ac\ubdf0\uac00 \ud3ec\ud568\ub41c \ub370\uc774\ud130\uc14b\uc73c\ub85c, \uc8fc\uc5b4\uc9c4 \ub9ac\ubdf0\uac00 \uae0d\uc815\uc801(1) \ub610\ub294 \ubd80\uc815\uc801(0)\uc778\uc9c0\ub97c \ubd84\ub958\ud558\ub294 \uc791\uc5c5\uc5d0 \ub110\ub9ac \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \uc774 \ub370\uc774\ud130\uc14b\uc740 25000\uac1c\uc758 \ub9ac\ubdf0\ub85c \uad6c\uc131\ub418\uc5b4 \uc788\uc73c\uba70, \uac01 \ub9ac\ubdf0\ub294 \uc790\uc5f0\uc5b4\ub85c \uc791\uc131\ub41c \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\uc785\ub2c8\ub2e4. \ub525\ub7ec\ub2dd \ubaa8\ub378\uc740 \uc774 \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\ub97c \uc774\ud574\ud558\uace0, \uac10\uc815\uc744 \ubd84\ub958\ud558\ub294 \ub370\uc5d0 \ub3c4\uc6c0\uc744 \uc90d\ub2c8\ub2e4.<\/p>\n
\uc6b0\uc120, \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud558\uace0 \ud658\uacbd\uc744 \uc124\uc815\ud558\uaca0\uc2b5\ub2c8\ub2e4. \ud5c8\uae45\ud398\uc774\uc2a4 \ud2b8\ub79c\uc2a4\ud3ec\uba38\uc640 \ud568\uaed8 \uc0ac\uc6a9\ud558\ub294 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub294 torch<\/strong>\uc640 datasets<\/strong>\uc785\ub2c8\ub2e4. \uc544\ub798 \ucf54\ub4dc\ub294 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud558\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4.<\/p>\n IMDB \ub370\uc774\ud130\uc14b\uc744 \ub85c\ub4dc\ud558\uae30 \uc704\ud574 datasets<\/strong> \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc0ac\uc6a9\ud558\uaca0\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c \ucf54\ub4dc\ub97c \uc2e4\ud589\ud558\uc5ec \ub370\uc774\ud130\ub97c \ubd88\ub7ec\uc635\ub2c8\ub2e4.<\/p>\n \uc704 \ucf54\ub4dc\ub294 IMDB \ub370\uc774\ud130\uc14b\uc744 \ub85c\ub4dc\ud558\uace0, \ub370\uc774\ud130\uc14b \uad6c\uc870\ub97c \ucd9c\ub825\ud569\ub2c8\ub2e4. \ucd9c\ub825 \uacb0\uacfc\ub85c \ud6c8\ub828 \ub370\uc774\ud130\uc640 \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uc758 \ud06c\uae30\ub97c \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n \ubaa8\ub378\uc774 \uc774\ud574\ud560 \uc218 \uc788\ub3c4\ub85d \ud14d\uc2a4\ud2b8 \ub370\uc774\ud130\ub97c \uc804\ucc98\ub9ac\ud574\uc57c \ud569\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc73c\ub85c \ud544\uc694\ud55c \uc804\ucc98\ub9ac \uacfc\uc815\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n \ud5c8\uae45\ud398\uc774\uc2a4 \ud2b8\ub79c\uc2a4\ud3ec\uba38 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc774\uc6a9\ud574 BERT \ubaa8\ub378\uc744 \uae30\ubc18\uc73c\ub85c \ud55c \ud1a0\ud06c\ub098\uc774\uc800\ub97c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c \ucf54\ub4dc\ub97c \ud1b5\ud574 \ud1a0\ud06c\ub098\uc774\uc800\ub97c \uc124\uc815\ud558\uace0 \ub370\uc774\ud130\ub97c \uc804\ucc98\ub9ac\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n \ud6c8\ub828 \ub370\uc774\ud130\uc14b\uc744 \ud6c8\ub828 \uc138\ud2b8\uc640 \uac80\uc99d \uc138\ud2b8\ub85c \ub098\ub204\uae30 \uc704\ud574 \ub370\uc774\ud130\uc14b\uc744 \ubd88\ub7ec\uc628 \ud6c4, PyTorch\uc758 \ub370\uc774\ud130\ub85c\ub354\ub97c \uc0ac\uc6a9\ud558\uc5ec \ub370\uc774\ud130\ub97c \ubd84\ud560\ud569\ub2c8\ub2e4. \uc544\ub798 \ucf54\ub4dc\ub97c \ucc38\uace0\ud558\uc138\uc694.<\/p>\n \uc774\uc81c \ubaa8\ub378\uc744 \uc124\uc815\ud574\uc57c \ud569\ub2c8\ub2e4. BERT \ubaa8\ub378\uc744 \uac10\uc815 \ubd84\uc11d\uc744 \uc704\ud55c \uc804\uc774\ud559\uc2b5(transfer learning)\uc73c\ub85c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798 \ucf54\ub4dc\ub294 BERT \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4.<\/p>\n \ubaa8\ub378\uc744 \ud6c8\ub828\ud558\uae30 \uc704\ud574 \uc635\ud2f0\ub9c8\uc774\uc800\uc640 \uc190\uc2e4 \ud568\uc218\ub97c \uc124\uc815\ud574\uc57c \ud569\ub2c8\ub2e4. \uc544\ub798 \ucf54\ub4dc\ub294 Adam \uc635\ud2f0\ub9c8\uc774\uc800\ub97c \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ub378\uc744 \ud6c8\ub828\ud558\ub294 \uacfc\uc815\uc744 \ubcf4\uc5ec\uc90d\ub2c8\ub2e4.<\/p>\n \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ud3c9\uac00\ud558\uae30 \uc704\ud574 \uac80\uc99d \uc138\ud2b8\ub97c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ud3c9\uac00 \uba54\ud2b8\ub9ad\uc740 \uc815\ud655\ub3c4(accuracy)\ub85c \uc124\uc815\ud569\ub2c8\ub2e4.<\/p>\n \ud6c8\ub828\uc774 \uc644\ub8cc\ub41c \ud6c4, \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc0c8\ub85c\uc6b4 \ub9ac\ubdf0\uc5d0 \ub300\ud55c \uac10\uc815 \uc608\uce21\uc744 \uc218\ud589\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n \uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 \ud5c8\uae45\ud398\uc774\uc2a4 \ud2b8\ub79c\uc2a4\ud3ec\uba38\ub97c \ud65c\uc6a9\ud558\uc5ec IMDB \ub370\uc774\ud130\uc14b\uc73c\ub85c \uc601\ud654 \ub9ac\ubdf0 \uac10\uc815 \ubd84\uc11d \ubaa8\ub378\uc744 \uad6c\ucd95\ud558\ub294 \uc804 \uacfc\uc815\uc744 \uc0b4\ud3b4\ubcf4\uc558\uc2b5\ub2c8\ub2e4. \ub370\uc774\ud130\uc14b \ub85c\ub4dc, \uc804\ucc98\ub9ac, \ubaa8\ub378 \ud559\uc2b5 \ubc0f \ud3c9\uac00 \ub2e8\uacc4\uae4c\uc9c0, \ubaa8\ub4e0 \uacfc\uc815\uc744 \uac70\uce58\uba70 \ub525\ub7ec\ub2dd\uc744 \ud65c\uc6a9\ud55c \ud14d\uc2a4\ud2b8 \ubd84\ub958\uc758 \ud750\ub984\uc744 \uc774\ud574\ud560 \uc218 \uc788\uc5c8\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4. \ud5c8\uae45\ud398\uc774\uc2a4 \ub77c\uc774\ube0c\ub7ec\ub9ac\ub294 \uac15\ub825\ud55c \uae30\ub2a5\uc744 \uc81c\uacf5\ud558\ubbc0\ub85c \uc5ec\ub7ec NLP \uc791\uc5c5\uc5d0 \ud65c\uc6a9\ud574 \ubcf4\uc2dc\uae30 \ubc14\ub78d\ub2c8\ub2e4.<\/p>\n \uac10\uc0ac\ud569\ub2c8\ub2e4!<\/p>\n <\/body><\/p>\n","protected":false},"excerpt":{"rendered":"!pip install transformers torch datasets<\/code><\/pre>\n3. \ub370\uc774\ud130\uc14b \ub85c\ub4dc<\/h2>\n
from datasets import load_dataset\n\ndataset = load_dataset(\"imdb\")\nprint(dataset)<\/code><\/pre>\n4. \ub370\uc774\ud130 \uc804\ucc98\ub9ac<\/h2>\n
\n
from transformers import BertTokenizer\n\ntokenizer = BertTokenizer.from_pretrained(\"bert-base-uncased\")\n\ndef encode_review(review):\n return tokenizer(review, padding=\"max_length\", truncation=True, max_length=512, return_tensors='pt')['input_ids'][0]\n\n# \ud6c8\ub828 \ub370\uc774\ud130\uc5d0\uc11c \uc77c\ubd80 \ub9ac\ubdf0\ub97c \uc804\ucc98\ub9ac\ntrain_encodings = { 'input_ids': [], 'label': [] }\nfor review, label in zip(dataset['train']['text'], dataset['train']['label']):\n train_encodings['input_ids'].append(encode_review(review))\n train_encodings['label'].append(label)<\/code><\/pre>\n5. \ub370\uc774\ud130\uc14b \ubd84\ud560<\/h2>\n
import torch\n\nclass IMDBDataset(torch.utils.data.Dataset):\n def __init__(self, encodings, labels):\n self.encodings = encodings\n self.labels = labels\n\n def __getitem__(self, idx):\n item = { 'input_ids': self.encodings['input_ids'][idx],\n 'labels': torch.tensor(self.labels[idx]) }\n return item\n\n def __len__(self):\n return len(self.labels)\n\ntrain_dataset = IMDBDataset(train_encodings, train_encodings['label'])<\/code><\/pre>\n6. \ubaa8\ub378 \uc124\uc815<\/h2>\n
from transformers import BertForSequenceClassification\n\nmodel = BertForSequenceClassification.from_pretrained(\"bert-base-uncased\", num_labels=2)<\/code><\/pre>\n7. \ud6c8\ub828<\/h2>\n
from transformers import AdamW\n from transformers import Trainer, TrainingArguments\n\n training_args = TrainingArguments(\n output_dir='.\/results',\n num_train_epochs=3,\n per_device_train_batch_size=8,\n logging_dir='.\/logs',\n )\n\n trainer = Trainer(\n model=model,\n args=training_args,\n train_dataset=train_dataset,\n )\n\n trainer.train()<\/code><\/pre>\n8. \ud3c9\uac00<\/h2>\n
eval_result = trainer.evaluate()\n print(eval_result)<\/code><\/pre>\n9. \uc608\uce21<\/h2>\n
def predict_review(review):\n encoding = encode_review(review)\n with torch.no_grad():\n logits = model(torch.tensor(encoding).unsqueeze(0))[0]\n predicted_label = torch.argmax(logits, dim=-1).item()\n return predicted_label\n\nsample_review = \"This movie was fantastic! I loved it.\"\npredicted_label = predict_review(sample_review)\nprint(f\"Predicted label for the review: {predicted_label}\") # 1: \uae0d\uc815, 0: \ubd80\uc815<\/code><\/pre>\n10. \uacb0\ub860<\/h2>\n