torch. unique()

torch.unique() The function of is similar to the set in mathematics , Is to pick out tensor Independent non repeating elements in .

The parameters of this method are listed in the official interpretation document :torch.unique(input, sorted=True, return_inverse=False, return_counts=False, dim=None)

input: To be dealt with tensor

sorted： Whether to arrange the returned unrepeated tensors in numerical values , By default, it is arranged in sequence

return_inverse: Whether to return to the original tensor The index of each element in this non repeating tensor

return_counts: Count the number of each independent element in the original tensor

dim: Which dimension does the value go along unique To deal with , I didn't understand the mechanism after the experiment . If all the tensors are one-dimensional , Then this need not be ignored .

The following is an experimental explanation and analysis of these different parameters .

import torch
 
x = torch.tensor([4,0,1,2,1,2,3])# Generate a tensor, As experimental input 
print(x)
 
out = torch.unique(x) # All parameters are set to default 
print(out)# Print out the processing results 
# give the result as follows ：
#tensor([0, 1, 2, 3, 4]) # take x The non repeating elements in are singled out , And the default is sequence arrangement 
 
out = torch.unique(x,sorted=False)# Change the default sort order to False
print(out)
# The output is as follows ：
#tensor([3, 2, 1, 0, 4]) # take x Found the independent elements in , Just output in the original order 
 
out = torch.unique(x,return_inverse=True)# Output the index of each element in the original data in the newly generated independent element tensor 
print(out)
# The output is as follows ：
#(tensor([0, 1, 2, 3, 4]), tensor([4, 0, 1, 2, 1, 2, 3])) # The first tensor is the independent tensor output after sorting , The second result corresponds to the index of each element in the original data in the new independent non repeating tensor , such as x[0]=4, The index in the new tensor is 4, x[1]=0, The index in the new tensor is 0,x[6]=3, The index in the new tensor is 3
 
out = torch.unique(x,return_counts=True) # Returns the number of each independent element 
print(out)
# The output is as follows 
#(tensor([0, 1, 2, 3, 4]), tensor([1, 2, 2, 1, 1])) #0 The number of this element in the original data is 1,1 The number of this element in the original data is 2

Reprinted from ：https://blog.csdn.net/t20134297/article/details/108235355