比较与torch.nn.Sequential的差异

torch.nn.Sequential

torch.nn.Sequential(
    *args
)

更多内容详见torch.nn.Sequential。

mindspore.nn.SequentialCell

mindspore.nn.SequentialCell(
    *args
)

更多内容详见mindspore.nn.SequentialCell。

差异对比

PyTorch：构造Cell顺序容器。Sequential按照传入List的顺序依次将Cell添加。此外，也支持OrderedDict作为构造器传入。

MindSpore：构造Cell顺序容器。入参类型和PyTorch一致。和PyTorch相比，MindSpore支持append()，在容器末尾添加Cell。

分类	子类	PyTorch	MindSpore	差异
参数	参数1	args	args	传入容器的参数，支持List和OrderedDict类型。

代码示例

import collections

# In MindSpore
import mindspore as ms

model = ms.nn.SequentialCell(
          ms.nn.Conv2d(1,20,5),
          ms.nn.ReLU(),
          ms.nn.Conv2d(20,64,5),
          ms.nn.ReLU()
        )
print(model)
# Out:
# SequentialCell<
#   (0): Conv2d<input_channels=1, output_channels=20, kernel_size=(5, 5), stride=(1, 1), pad_mode=same, padding=0, dilation=(1, 1), group=1, has_bias=False, weight_init=normal, bias_init=zeros, format=NCHW>
#   (1): ReLU<>
#   (2): Conv2d<input_channels=20, output_channels=64, kernel_size=(5, 5), stride=(1, 1), pad_mode=same, padding=0, dilation=(1, 1), group=1, has_bias=False, weight_init=normal, bias_init=zeros, format=NCHW>
#   (3): ReLU<>
#   >

# Example of using Sequential with OrderedDict
model = ms.nn.SequentialCell(collections.OrderedDict([
          ('conv1', ms.nn.Conv2d(1,20,5)),
          ('relu1', ms.nn.ReLU()),
          ('conv2', ms.nn.Conv2d(20,64,5)),
          ('relu2', ms.nn.ReLU())
        ]))
print(model)
# Out:
# SequentialCell<
#   (conv1): Conv2d<input_channels=1, output_channels=20, kernel_size=(5, 5), stride=(1, 1), pad_mode=same, padding=0, dilation=(1, 1), group=1, has_bias=False, weight_init=normal, bias_init=zeros, format=NCHW>
#   (relu1): ReLU<>
#   (conv2): Conv2d<input_channels=20, output_channels=64, kernel_size=(5, 5), stride=(1, 1), pad_mode=same, padding=0, dilation=(1, 1), group=1, has_bias=False, weight_init=normal, bias_init=zeros, format=NCHW>
#   (relu2): ReLU<>
#   >


# In PyTorch
import torch

model = torch.nn.Sequential(
          torch.nn.Conv2d(1,20,5),
          torch.nn.ReLU(),
          torch.nn.Conv2d(20,64,5),
          torch.nn.ReLU()
        )
print(model)
# Out
# Sequential(
#   (0): Conv2d(1, 20, kernel_size=(5, 5), stride=(1, 1))
#   (1): ReLU()
#   (2): Conv2d(20, 64, kernel_size=(5, 5), stride=(1, 1))
#   (3): ReLU()
# )

# Example of using Sequential with OrderedDict
model = torch.nn.Sequential(collections.OrderedDict([
          ('conv1', torch.nn.Conv2d(1,20,5)),
          ('relu1', torch.nn.ReLU()),
          ('conv2', torch.nn.Conv2d(20,64,5)),
          ('relu2', torch.nn.ReLU())
        ]))
print(model)
# Out：
# Sequential(
#   (conv1): Conv2d(1, 20, kernel_size=(5, 5), stride=(1, 1))
#   (relu1): ReLU()
#   (conv2): Conv2d(20, 64, kernel_size=(5, 5), stride=(1, 1))
#   (relu2): ReLU()
# )