windows环境下pytorch使用num_worker

pip install torchnet

class DataLoader:
    def __init__(self, ...):
        pass
    
    def get_task_batch(self, ...):
        pass
    
    def get_iterator(self):
        tnt_dataset = torchnet.dataset.ListDataset(
            elem_list=range(self.task_num), load=self.get_task_batch)
        data_loader = tnt_dataset.parallel(
            batch_size=self.batch_size,
            num_workers=self.num_workers,
            drop_last=True,
            # 此函数可以使每个worker使用不同的随机种子
            worker_init_fn=self.worker_init_fn_seed,
            shuffle=(False if self.test else True))
        return data_loader
    
    def worker_init_fn_seed(self, worker_id):
        seed = 10 + 5 * worker_id
        np.random.seed(seed)
    
    def __call__(self):
        return self.get_iterator()

# MiniImagenet为预处理数据的类
dataset = MiniImagenet(data_path, ...)
loader = DataLoader(dataset, num_workers=2, ...)
for step, batch in enumerate(loader()):
    x, y = batch