CODEWITHSUNDEEP
pythonnlppytorchlookup-tables
k_p
k_p

Reputation: 313

how to replace torch.Tensor to a value in python

my predictions in a pytorch are coming as torch([0]) , torch([1])....,torch([25]) for respective 26 alphabets i.e. A,B,C....Z. my prediction are coming as torch([0]) which i want as A and so on . Any idea how to do this conversion .

Upvotes: 1

Views: 936

Answers (2)

Iguananaut
Iguananaut

Reputation: 23306

You want Tensor.item()

>>> import torch
>>> t = torch.tensor([0])
>>> t.item()
0

If you want to convert it to a letter from A to Z you can use:

>>> import string
>>> string.ascii_uppercase[t.item()]
'A'

Be careful to check the shape before doing this, or wrap in a try/except for a possible ValueError:

>>> t = torch.tensor([0, 1])
>>> t.item()
Traceback (most recent call last):
  File "<ipython-input-6-dc80242434c0>", line 1, in <module>
    t.item()
ValueError: only one element tensors can be converted to Python scalars

Upvotes: 1

Shai
Shai

Reputation: 114816

To convert indices of the alphabet to the actual letters, you can:

alphabet = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'  # the Alphabet
pred = torch.randint(0, 26, (30,))  # your prediction, int tensor with values in range[0, 25]
# convert to characters
pred_string = ''.join(alphabet[c_] for c_ in pred)

the output would be something like:

'KEFOTIJBNTAPWHSBXUIQKTTJNSCNDF'

This will also work for pred with a single element, in which case the conversion can done more compactly:

alphabet[pred.item()]

Upvotes: 1

Related Questions