Add requirements.txt

inference.py

@@ -0,0 +1,62 @@
+import argparse
+
+import pandas as pd
+
+from tqdm.auto import tqdm
+
+import transformers
+import torch
+import torchmetrics
+import pytorch_lightning as pl
+
+
+### TODO : path, filename 정해지면 수정
+# from PATH import CUSTOMDATAMODULE as DataModule
+# from PATH import CUSTOMMODEL as Model
+# from PATH import CUSTOMTRAINER as Trainer
+# from PATH import CUSTOMPREPROCESS as preprocess
+
+
+# fix random seeds for reproducibility
+SEED = 123
+seed_everything(SEED, workers=True) # pl seed
+
+
+
+def inference(args):
+    ### TODO : preprocess 함수도 인자로 넘겼으면..
+    datamodule = DataModule(args.model_name, 
+                            args.batch_size, args.shuffle, 
+                            args.train_path, args.dev_path,
+                            args.test_path, args.predict_path)
+
+    model = Model(args.model_name, 
+                  args.learning_rate)
+
+    trainer = Trainer(gpus=1, 
+                      max_epochs=arg.max_epoch,
+                      log_every_n_steps=1)
+
+    # Inference
+    model = torch.load(args.best_model_path)
+    predictions = trainer.predict(model=model,
+                                  datamodule=datamodule)
+
+    predictions = list(round(float(i), 1) for i in torch.cat(predictions))
+
+    output = pd.read_csv('./data/sample_submission.csv')
+    output['target'] = predictions
+
+    output_name = args.best_model_path[:-3] + '.csv'
+    output.to_csv('output.csv', index=False)
+
+
+if __name__ == '__main__':
+    ### TODO : best model의 config값 불러와서 argument로 지정
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--best_model_path', default='model.pt')
+
+    args = parser.parse_args(args=[])
+
+    inference(args)
+

requirements.txt

@@ -0,0 +1,8 @@
+pyYAML==5.4.1
+pytorch-lightning==1.7.3
+tqdm==4.64.1
+torchmetrics==0.10.0
+torch==1.12.1
+transformers==4.21.2
+numpy==1.20.3
+pandas==1.5.0

train.py

@@ -10,6 +10,24 @@
 from trainer import Trainer
 from utils import prepare_device
 
+from datetime import datetime
+
+import pytorch_lightning as pl
+from pytorch_lightning import seed_everything
+
+### TODO : path, filename 정해지면 수정
+# from PATH import CUSTOMDATAMODULE as DataModule
+# from PATH import CUSTOMMODEL as Model
+# from PATH import CUSTOMTRAINER as Trainer
+# from PATH import CUSTOMPREPROCESS as preprocess
+
+### TODO : utils에 scatterplot 구현하기
+# "일례로, 위 그림은 모두 0.82의 상관계수를 가지나 모두 생김새가 다릅니다. 그러므로, 결과 분석 시 꼭 산점도를 그려보며 모델의 취약점을 평가해야합니다."
+# from utils import scatterplot
+
+### TODO : utils에 save_config 구현하기
+# from utils import save_config
+
 
 # fix random seeds for reproducibility
 SEED = 123
@@ -18,6 +36,75 @@
 torch.backends.cudnn.benchmark = False
 np.random.seed(SEED)
 
+seed_everything(SEED, workers=True) # pl seed
+
+
+
+def train(args):
+    ### TODO : preprocess 함수도 인자로 넘겼으면..
+    datamodule = DataModule(args.model_name, 
+                            args.batch_size, args.shuffle, 
+                            args.train_path, args.dev_path,
+                            args.test_path, args.predict_path)
+
+    model = Model(args.model_name, 
+                  args.learning_rate)
+
+    trainer = Trainer(gpus=1, 
+                      max_epochs=arg.max_epoch,
+                      log_every_n_steps=1)
+
+    trainer.fit(model=model, datamodule=datamodule)
+    test_dict = trainer.test(model=model, datamodule=datamodule) # val로 테스트함. 
+
+
+    # 모델 이름 자동 생성
+    # 형식: val 성능 + 모델이름 + 저장시각 + 실험자.pt
+    saved_model_path = './save/model/'
+    test_pearson = test_dict[0]['test_pearson']
+
+    now = datetime.now()
+    now_str = now.strftime('%Y-%m-%d_%H:%M:%S')
+    saved_model_name = '+'.join([str(test_pearson), 
+                                args.model_name, 
+                                now_str,
+                                args.expreimenter])
+
+    torch.save(model, saved_model_path + saved_model_name + '.pt')
+
+
+    ### TODO : scatterplot 그리고 저장하는 함수
+    # scatterplot(data, './save/plot/' + saved_model_name + '.png')
+
+
+    ### TODO : config 저장하는 함수
+    # save_config(args, './save/config/' + saved_model_name + '.config')
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--model_name', default='klue/roberta-small', type=str)
+    parser.add_argument('--batch_size', default=16, type=int)
+    parser.add_argument('--max_epoch', default=1, type=int)
+    parser.add_argument('--shuffle', default=True)
+    parser.add_argument('--learning_rate', default=1e-5, type=float)
+    parser.add_argument('--train_path', default='../data/train.csv')
+    parser.add_argument('--dev_path', default='../data/dev.csv')
+    parser.add_argument('--test_path', default='../data/dev.csv')
+    parser.add_argument('--predict_path', default='../data/test.csv')
+
+    ### TODO : custom argument
+    parser.add_argument('--preprocess', default="")
+    parser.add_argument('--experimenter', default='bc-nlp-03')
+    parser.add_argument('--optimizer', default='AdamW')
+    parser.add_argument('--loss_func', default='L1Loss')
+    parser.add_argument('--num_workers', default=4)
+
+    args = parser.parse_args(args=[])
+
+    train(args)
+
+"""
 def main(config):
     logger = config.get_logger('train')
 
@@ -71,3 +158,4 @@ def main(config):
     ]
     config = ConfigParser.from_args(args, options)
     main(config)
+"""