A. Lets work this out.

Let's define a tensor with 10 integers from 1 to 10 randomly permuted.



In [3]:

    
t=torch.randperm(10)



In [4]:

    
t









    Out[4]:





  8
  3
  6
  1
  5
  7
  4
  9
 10
  2
[torch.DoubleTensor of dimension 10]

Now, find all instances of "7" in the tensor.

Let us do two methods.

Method (1) is a simple for-loop.



In [5]:

    
for i=1,t:size(1) do 
    if t[i] == 7 then
        print('Found 7 at location: ' .. i)
    end
end









    Out[5]:





Found 7 at location: 6

Method (2) which is more efficient and faster:

We use the :eq function, which is part of the logical operators.



In [6]:

    
t:eq(7)









    Out[6]:





 0
 0
 0
 0
 0
 1
 0
 0
 0
 0
[torch.ByteTensor of dimension 10]

As you see, it retured a mask, where there is a value 1 wherever there was the value 7. Now, let us extract a new tensor with just the value.



In [7]:

    
t[t:eq(7)]









    Out[7]:





 7
[torch.DoubleTensor of dimension 1]

Okay, but let's say you actually want the Index number of all locations of 7. You can technically do this with a linspace:



In [8]:

    
idx = torch.linspace(1,t:size(1), t:size(1))



In [9]:

    
idx[t:eq(7)]









    Out[9]:





 6
[torch.DoubleTensor of dimension 1]

Hope this helps make it clear.



In [ ]: