一个用于节省内存的技巧：(构建Dataset后可将原始数据删除！训练和验证完成后可将训练集和验证集的DataLoader删除！)
import gc
from torch.utils.data import DataLoader, Dataset, random_split
# preprocess data
train_X, train_y = pass
val_X, val_y = pass
# get dataset
train_set = MyDataset(train_X, train_y)
val_set = MyDataset(val_X, val_y)
# remove raw feature to save memory
del train_X, train_y, val_X, val_y
gc.collect()
# get dataloader
train_loader = DataLoader(train_set, batch_size=batch_size, shuffle=True)
val_loader = DataLoader(val_set, batch_size=batch_size, shuffle=False)
"""此处为训练和验证过程..."""
# after training, before testing
del train_loader, val_loader
gc.collect()
