AdaptiveAvgPool1d(N)函数解读

• 函数作用
  AdaptiveAvgPool1d(N) 对于一个输入BCL的tensor进行一维的pool，变为BCN
• 代码示例

>>> import torch
>>>  as nn
>>> a&#s(2,3,4)
>>> a[0,1,2]=0
>>> a
tensor([[[1., 1., 1., 1.],[1., 1., 0., 1.],[1., 1., 1., 1.]],[[1., 1., 1., 1.],[1., 1., 1., 1.],[1., 1., 1., 1.]]])
>>> nn.AdaptiveAvgPool1d(5)(a)
tensor([[[1.0000, 1.0000, 1.0000, 1.0000, 1.0000],[1.0000, 1.0000, 0.5000, 0.5000, 1.0000],[1.0000, 1.0000, 1.0000, 1.0000, 1.0000]],[[1.0000, 1.0000, 1.0000, 1.0000, 1.0000],[1.0000, 1.0000, 1.0000, 1.0000, 1.0000],[1.0000, 1.0000, 1.0000, 1.0000, 1.0000]]])
>>> nn.AdaptiveAvgPool1d(5)(a).shape
torch.Size([2, 3, 5])
>>> nn.AdaptiveAvgPool1d(2)(a)
tensor([[[1.0000, 1.0000],[1.0000, 0.5000],[1.0000, 1.0000]],[[1.0000, 1.0000],[1.0000, 1.0000],[1.0000, 1.0000]]])
>>> nn.AdaptiveAvgPool1d(2)(a).shape
torch.Size([2, 3, 2])