#展平一個(gè)連續(xù)范圍的維度，輸出類型為Tensor
torch.flatten(input, start_dim=0, end_dim=-1) → Tensor
# Parameters：input (Tensor) – 輸入為Tensor
#start_dim (int) – 展平的開始維度
#end_dim (int) – 展平的最后維度
#example
#一個(gè)3x2x2的三維張量
>>> t = torch.tensor([[[1, 2],
                       [3, 4]],
                      [[5, 6],
                       [7, 8]],
                  [[9, 10],
                       [11, 12]]])
#當(dāng)開始維度為0，最后維度為-1，展開為一維
>>> torch.flatten(t)
tensor([ 1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12])
#當(dāng)開始維度為0，最后維度為-1，展開為3x4，也就是說第一維度不變，后面的壓縮
>>> torch.flatten(t, start_dim=1)
tensor([[ 1,  2,  3,  4],
        [ 5,  6,  7,  8],
        [ 9, 10, 11, 12]])
>>> torch.flatten(t, start_dim=1).size()
torch.Size([3, 4])
#下面的和上面進(jìn)行對(duì)比應(yīng)該就能看出是，當(dāng)鎖定最后的維度的時(shí)候
#前面的就會(huì)合并
>>> torch.flatten(t, start_dim=0, end_dim=1)
tensor([[ 1,  2],
        [ 3,  4],
        [ 5,  6],
        [ 7,  8],
        [ 9, 10],
        [11, 12]])
>>> torch.flatten(t, start_dim=0, end_dim=1).size()
torch.Size([6, 2])

Class torch.nn.Flatten(start_dim=1, end_dim=-1)
#Flattens a contiguous range of dims into a tensor. 
#For use with Sequential. :
#param start_dim: first dim to flatten (default = 1). 
#param end_dim: last dim to flatten (default = -1).
#能力有限，個(gè)人認(rèn)為是用于卷積中的
#Shape:
#Input: (N, *dims)(N,?dims)
#Output: (N, \prod *dims)(N,∏?dims) (for the default case).
#官方example
>>> m = nn.Sequential(
>>>     nn.Conv2d(1, 32, 5, 1, 1),
>>>     nn.Flatten()
>>> )
#源代碼為 TORCH.NN.MODULES.FLATTEN
from .module import Module
[docs]class Flatten(Module):
    r"""
    Flattens a contiguous range of dims into a tensor. For use with :class:`~nn.Sequential`.
    Args:
        start_dim: first dim to flatten (default = 1).
        end_dim: last dim to flatten (default = -1).
    Shape:
        - Input: :math:`(N, *dims)`
        - Output: :math:`(N, \prod *dims)` (for the default case).
    Examples::
        >>> m = nn.Sequential(
        >>>     nn.Conv2d(1, 32, 5, 1, 1),
        >>>     nn.Flatten()
        >>> )
    """
    __constants__ = ['start_dim', 'end_dim']
    def __init__(self, start_dim=1, end_dim=-1):
        super(Flatten, self).__init__()
        self.start_dim = start_dim
        self.end_dim = end_dim
    def forward(self, input):
        return input.flatten(self.start_dim, self.end_dim)

import torch
input = torch.tensor([[1, 2, 3],
                      [4, 5, 6]])
output = torch.flatten(input, start_dim=0, end_dim=1)
print(output)
tensor([1, 2, 3, 4, 5, 6])