class Flatten(Layer):
 def __init__(self, **kwargs):
  super(Flatten, self).__init__(**kwargs)
  self.input_spec = InputSpec(min_ndim=3)

 def compute_output_shape(self, input_shape):
  if not all(input_shape[1:]):
   raise ValueError('The shape of the input to "Flatten" '
        'is not fully defined '
        '(got ' + str(input_shape[1:]) + '. '
        'Make sure to pass a complete "input_shape" '
        'or "batch_input_shape" argument to the first '
        'layer in your model.')
  return (input_shape[0], np.prod(input_shape[1:]))

 def call(self, inputs):
  return K.batch_flatten(inputs)

from keras import backend as K
from keras.engine.topology import Layer
import tensorflow as tf
import numpy as np

class MyFlatten(Layer):
 def __init__(self, **kwargs):
  self.supports_masking = True
  super(MyFlatten, self).__init__(**kwargs)

 def compute_mask(self, inputs, mask=None):
  if mask==None:
   return mask
  return K.batch_flatten(mask)

 def call(self, inputs, mask=None):
  return K.batch_flatten(inputs)

 def compute_output_shape(self, input_shape):
  return (input_shape[0], np.prod(input_shape[1:]))

from keras.layers import *
from keras.models import Model
from MyFlatten import MyFlatten
from MySumLayer import MySumLayer
from keras.initializers import ones

data = [[1,0,0,0],
  [1,2,0,0],
  [1,2,3,0],
  [1,2,3,4]]

A = Input(shape=[4]) # None * 4
emb = Embedding(5, 3, mask_zero=True, embeddings_initializer=ones())(A) # None * 4 * 3
fla = MyFlatten()(emb) # None * 12
out = MySumLayer(axis=1)(fla) # None * 1

model = Model(inputs=[A], outputs=[out])
print model.predict(data)

import torch
import numpy as np
a = np.arange(24)
b = a.reshape(4,3,2)
print(np.shape(a))
print(b,np.shape(b))

'''结果
(24,)
[[[ 0 1]
 [ 2 3]
 [ 4 5]]

 [[ 6 7]
 [ 8 9]
 [10 11]]

 [[12 13]
 [14 15]
 [16 17]]

 [[18 19]
 [20 21]
 [22 23]]] (4, 3, 2)
'''

#官方例子
> x = torch.randn(2, 3)
> x
tensor([[ 0.9068, 1.8803, -0.5021],
  [-0.6576, 0.6334, -0.8961]])
> torch.transpose(x, 0, 1)
tensor([[ 0.9068, -0.6576],
  [ 1.8803, 0.6334],
  [-0.5021, -0.8961]])

#官方例子
> t = torch.tensor([[[1, 2],
        [3, 4]],
        [[5, 6],
        [7, 8]]])
> torch.flatten(t)
tensor([1, 2, 3, 4, 5, 6, 7, 8])
> torch.flatten(t, start_dim=1)
tensor([[1, 2, 3, 4],
  [5, 6, 7, 8]])

> m = torch.nn.Sequential(
 torch.nn.Conv2d(1, 32, 5, 1, 1),
 torch.nn.Flatten(),
 torch.nn.Linear(160,10))
> m
Sequential(
 (0): Conv2d(1, 32, kernel_size=(5, 5), stride=(1, 1), padding=(1, 1))
 (1): Flatten()
 (2): Linear(in_features=160, out_features=10, bias=True)
)