Fix root cause of bug reported in PR #1412

[renatomello]

The root cause of the issue raised in #1412 was a cast to numpy inside NumpyBackend.samples_to_binary. That is what was turning the subsequent cast into a requirement for torch. Without the former, the latter could be removed.

Checklist:

• Reviewers confirm new code works as expected.
• Tests are passing.
• Coverage does not decrease.
• Documentation is updated.

[alecandido]

Aren't all the other operations in here still NumPy? Or is Torch automatically converting NumPy calls?

I expected self.np. usage, but instead there is np. everywhere...

[renatomello]

I tried self.np and it creates other issues because torch.arange and torch.right_shift work slightly differently

[alecandido]

But is np.right_shift() able to work on Torch tensors? (i.e. taking Torch tensors as input and returning them as output)

I see it could be vaguely possible, but it's still good to know, because I wouldn't have been confident to say it's actually able.

[renatomello]

In [6]: import numpy as np
   ...: import torch
   ...: 
   ...: x = torch.Tensor(torch.arange(10))
   ...: qrange = np.arange(10 - 1, -1, -1, dtype=np.int32)
   ...: type(np.right_shift(x[:, None], qrange))
Out[6]: torch.Tensor

[alecandido]

Nice, thanks :)

[alecandido]

It seems that by now this starts being reliable

https://numpy.org/doc/stable/user/basics.interoperability.html#operating-on-foreign-objects-without-converting
https://numpy.org/neps/nep-0018-array-function-protocol.html#nep18

(ok, most likely since a while, but it is for sure framework-dependent, I should take the time of assessing which ones are fully supporting it)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.94%. Comparing base (3d87ba7) to head (ad3eb17).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1417      +/-   ##
==========================================
- Coverage   99.94%   99.94%   -0.01%     
==========================================
  Files          78       78              
  Lines       11225    11224       -1     
==========================================
- Hits        11219    11218       -1     
  Misses          6        6              

Flag	Coverage Δ
unittests	99.94% <100.00%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.