tensorflow改写为pytorch的方法总结

偏执的太偏执、 2021-12-07 13:55 875阅读 0赞

## **替换方法：** ##

1.  
matrix = tf.get\_variable("Matrix", \[output\_size, input\_size\], dtype=input\_.dtype)  
=  
 matrix = Variable(torch.randn(output\_size, input\_size))

2.  
self.D\_l2\_loss = tf.constant(0.0)  
=  
  self.D\_l2\_loss = torch.Tensor(0.0)

3.  
 losses = tf.nn.softmax\_cross\_entropy\_with\_logits(logits=D\_scores, labels=self.D\_input\_y)  
=  
losses = torch.nn.CrossEntropyLoss(D\_scores, D\_input\_y)

4.  
 tf.reduce\_mean(losses)  
=  
torch.mean(losses)

5.  
embedded\_chars = tf.nn.embedding\_lookup(W\_fe, Feature\_input + 1)  
=  
   embedded\_chars = torch.index\_select(W\_fe, 0, Feature\_input + 1)

6.  
embedded\_chars\_expanded = tf.expand\_dims(embedded\_chars, -1)  
=  
embedded\_chars\_expanded = torch.unsqueeze(embedded\_chars, -1)

7.  
 h\_pool = tf.concat(pooled\_outputs, 3)  
=  
 h\_pool = torch.cat(pooled\_outputs, 3)

8.  
tf.transpose(matrix)  
=  
matrix.transpose(1, 0)

9.  
TensorArray可以看做是具有动态size功能的Tensor数组。通常都是跟while\_loop或map\_fn结合使用。

10.tf.TensorArray简单教程    https://blog.csdn.net/z2539329562/article/details/80639199

ta.stack(name=None) 将TensorArray中元素叠起来当做一个Tensor输出  
ta.unstack(value, name=None) 可以看做是stack的反操作，输入Tensor，输出一个新的TensorArray对象  
ta.write(index, value, name=None) 指定index位置写入Tensor  
ta.read(index, name=None) 读取指定index位置的Tensor

11.  
torch.squeeze  
=？？？？？不确定  
ta.stack

12.  
torch.matmul  
=  
tf.matmul

13.  
tf.expand\_dims  
=  
torch.unsqueeze

14.  
torch.split( torch.cat(\[torch.split(input\_x, i, 1)\[0\], self.padding\_array\], 1),  
                     self.sequence\_length, 1)\[0\]  
=有一点不一样  
tf.split(tf.concat(\[tf.split(input\_x, \[i, self.sequence\_length - i\], 1)\[0\], self.padding\_array\], 1),\[self.sequence\_length, i\], 1)\[0\]

14.  
tf.squeeze()  
=  
torch.squeeze

16，  
sub\_goal = tf.nn.l2\_normalize(sub\_goal, 1)  
=  
sub\_goal = sklearn.preprocessing.normalize(sub\_goal , norm='l2')

17.  
tf.log(tf.nn.softmax(）  
=  
F.log\_softmax（）

18.  
tf.multinomial(log\_prob, 1)  
=  
 torch.multinomial(log\_prob, 1,  replacement=True)

19.  
tf.multiply  
=  
\*或者torch.mul

20.  
tf.one\_hot  
=？？？  
onehot\_encoder = OneHotEncoder(sparse=False)  
onehot\_encoder.fit\_transform(next\_token)

21.  
tf.reduce\_sum（a, 0）  
=  
a.sum(0)