5. pytorch-项目流程

1. 数据准备

基本步骤: 生成Dataset(或其子类)对象->传入DataLoader(为可迭代对象，可以用for迭代)

1.1 Dataset类

Dataset为抽象类

• 注意
  • 直接从Dataset中取出的数据是没有经过transform的，只有通过Dataloader加载才可以

    training_data=torchvision.datasets.MNIST(root="./mnist", train=True,
                               transform=torchvision.transforms.ToTensor(), download=True)
    # 像素点的范围仍然是0-255, 不是0-1
    print(training_data.train_data[0])
    

1.1.1 Dataset子类：TensorDataset

• 源码阅读

  class TensorDataset(Dataset):
    """Dataset wrapping tensors.
  
    Each sample will be retrieved by indexing tensors along the first dimension.
  
    Arguments:
        *tensors (Tensor): tensors that have the same size of the first dimension.
    代码示例:
        x = torch.linspace(1, 10, 10)
        y = torch.linspace(10, 1, 10)
        dataset = TensorDataset(x, y)
    """
    def __init__(self, *tensors):
        """
        &1
        tensors[0]为x; tensor[1]为y。因为x,y的batch_size要相同，所以要assert
        TensorDataset(x, y, z...)传入任意多参数都是可以的
        """
        assert all(tensors[0].size(0) == tensor.size(0) for tensor in tensors)
        self.tensors = tensors
        
  
    def __getitem__(self, index):
        """
        &2
        相当于重载[]运算符
        """
        return tuple(tensor[index] for tensor in self.tensors)
  
    def __len__(self):
        return self.tensors[0].size(0)
  

• 示例代码

  import torch
  import torch.utils.data as Data
  
  if __name__ == "__main__":
    x = torch.linspace(1, 10, 10)
    y = torch.linspace(10, 1, 10)
    dataset = Data.TensorDataset(x, y)
    # &1
    # 当最后一个step不足5个(假设仅剩2个)，则仅会返回2个
    # shuffle: 训练时为True则打乱数据集
    # num_workers为子进程数量
    dataloader = Data.DataLoader(dataset=dataset, batch_size=5,
                                 shuffle=True, num_workers=2)
    for epoch in range(3):
        for step,input_data in enumerate(dataloader):
            print(f"{epoch}-{step}:\n{input_data}")
  

2. 网络搭建

2.1 class模式

2.2 Sequential模式

net = torch.nn.Sequential(
        torch.nn.Linear(2, 10),
        torch.nn.ReLU(),
        torch.nn.Linear(10, 2)
    )