Pytorch sharpening chapter ｜ argmax and argmin functions

One 、 Grammar format

Format 1 （ Only aim at argmax function ）：

​​torch.argmax(input) → LongTensor​​

function ：

Returns the indices of the maximum value of all elements in the ​​input tensor.​​

namely ： Returns the index corresponding to the maximum value of all elements in the input tensor （ Search by line ）; If there are multiple identical values , Then return the index corresponding to the value encountered for the first time .

give an example ：

In [28]: r=torch.tensor([[1,2,3,4,5],[6,7,8,9,10],[11,12,13,14,15]])

In [29]: torch.argmax(r)

Out[29]: tensor(14)

Format two ：

［1］torch.argmax(input, dim=None, keepdim=False)

function ：

Returns the indices of the maximum values of a tensor across a dimension.

• input(​ ​Tensor​​) – the input tensor. namely ： Output tensor .
• dim(​ ​int​​) – the dimension to reduce. If​​None​​, the argmax of the flattened input is returned. namely ： Dimension to reduce .

• keepdim(​ ​bool​​​) – whether the output tensor has ​​​dim​​​ retained or not. Ignored if ​​​dim=None​​​. namely ：

give an example ：

Log in and copy

      
      

       
       In [30]: a = torch.randn(4, 4)
       
       

       
       
In [31]: a
       
       
Out[31]: 
       
       
tensor([[ 1.4360,  0.6342, -0.5233,  0.4902],
       
       
        [ 1.1998, -0.8644,  0.5244,  0.2690],
       
       
        [ 0.0998, -1.5043,  0.1619, -1.4634],
       
       
        [ 0.0992, -1.0843, -1.3829,  0.5790]])
       
       

       
       
In [32]: torch.argmax(a)
       
       
Out[32]: tensor(0)
       
       

       
       
In [33]: torch.argmax(a,dim=0)
       
       
Out[33]: tensor([0, 0, 1, 3])
       
       

       
       
In [34]: torch.argmax(a,dim=1)
       
       
Out[34]: tensor([0, 0, 2, 3])
      
      
      
      

       
       
1.
       
       
2.
       
       
3.
       
       
4.
       
       
5.
       
       
6.
       
       
7.
       
       
8.
       
       
9.
       
       
10.
       
       
11.
       
       
12.
       
       
13.
       
       
14.
       
       
15.
       
       
16.
       
       
17.
      
      

• about tensor(0) Output , The meaning is as follows ：

The first 0 individual ： 1.4360	The first 1 individual ： 0.6342	The first 2 individual ： -0.5233	The first 3 individual ： 0.4902	The first 4 individual ： 1.1998	The first 5 individual ： -0.8644	The first 6 individual ： 0.5244	The first 7 individual ： 0.2690	The first 8 individual ： 0.0998	The first 9 individual ： -1.5043
The first 10 individual ： 0.1619	The first 11 individual ： -1.4634	The first 12 individual ： 0.0992	The first 13 individual ： -1.0843	The first 14 individual ： -1.3829	The first 15 individual ： 0.5790

• about tensor([0, 0, 1, 3]) Output , The meaning is as follows ：

At this time , Each column is regarded as subscript from 0 To 3 An array of . Easy to see , Each column from left to right （ Array ） The maximum values in are ：1.4360、0.6342、0.5244、0.5790, The subscripts in their corresponding one-dimensional array are 0、0、1、3, So we get the tensor tensor([0, 0, 1, 3]).

• ​ about tensor([0, 0, 2, 3]) Output ：

The meaning is easy to understand . Look horizontally from left to right, from top to bottom , Each row corresponds to an array , The subscripts left and right are 0、1、2、3. therefore , this 4 The maximum values in the arrays are 1.4360、1.1998、0.1619、1.3829, The subscripts in their corresponding one-dimensional array are 0、0、2、3, So we get the tensor tensor([0, 0, 2, 3]).

function ：

［2］torch.argmin(input, dim=None, keepdim=False) → LongTensor

argmin function ：Returns the indices of the minimum value(s) of the flattened tensor or along a dimension.

Understand something like the above argmax The second format of the function , Corresponding to dim=0 and dim=1, Return the tensor composed of the column direction array and the row direction array composed of the subscript corresponding to the minimum value in turn .