parse_args()的使用：

• 建立解析对象：parser = argparse.ArgumentParser()；
• 给parser实例添加属性：parser.add_argument('-epoches', type=int, default=15, help='batch size for dataloader')；
• 增加属性；
• 通过args = parser.parse_args()把刚才的属性从parser给args，后面直接通过args使用。

import argparse

if __name__ == "__main__":
    # 建立解析对象
    parser = argparse.ArgumentParser()
    
    # 给parser实例添加属性
    parser.add_argument('-gpu', action='store_true', default=True, help='use gpu or not')
    parser.add_argument('-bs', type=int, default=128, help='batch size for dataloader')
    parser.add_argument('-epoches', type=int, default=15, help='batch size for dataloader')
    
    # 把刚才的属性给args实例，后面就可以直接使用
    args = parser.parse_args()

    continous_feature_names = ['releaseYear', 'movieRatingCount', 'movieAvgRating', 'movieRatingStddev',
                               'userRatingCount', 'userAvgRating', 'userRatingStddev']
    categorial_feature_names = ['userGenre1', 'userGenre2', 'userGenre3', 'userGenre4', 'userGenre5',
                                'movieGenre1', 'movieGenre2', 'movieGenre3', 'userId', 'movieId']

    categorial_feature_vocabsize = [20] * 8 + [30001] + [1001]
    
    # build dataset for train and test
    batch_size = args.bs
    train_data = build_dataset(args.train_path)
    loader_train = DataLoader(train_data, batch_size=batch_size, num_workers=64, shuffle=True, pin_memory=True)
    test_data = build_dataset(args.test_path)
    loader_test = DataLoader(test_data, batch_size=batch_size, num_workers=64)

    device = torch.device("cuda" if args.gpu else "cpu")
    # train model
    model = WideDeep(categorial_feature_vocabsize, continous_feature_names, categorial_feature_names, embed_dim=64)
    if args.gpu:
        model = model.to(device)
    optimizer = optim.Adam(model.parameters(), lr=1e-3, weight_decay=1e-3)
    best_acc = 0

	# 这里就直接使用args.epoches
    for ep in range(args.epoches):
        train(ep)
        best_acc = test(ep, best_acc)

Reference

[1] Py官网：argparse — 命令行选项、参数和子命令解析器