Remove numpy dependency from the core library

[maximsch2]

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

What does this PR do?

Remove numpy dependency that seemed to have been added mostly for one test to make it cleaner. Instead moved the function that needs to support both Tensor and ndarray to test utils and removed the dependency from the core library.

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[pep8speaks]

Hello @maximsch2! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2021-04-30 12:16:53 UTC

[codecov]

Codecov Report

Merging #212 (781300e) into master (263a4c7) will decrease coverage by 0.00%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #212      +/-   ##
==========================================
- Coverage   96.66%   96.65%   -0.01%     
==========================================
  Files         180      180              
  Lines        5750     5744       -6     
==========================================
- Hits         5558     5552       -6     
  Misses        192      192              

Flag	Coverage Δ
Linux	79.73% <66.66%> (-0.03%)	⬇️
Windows	79.73% <66.66%> (-0.03%)	⬇️
cpu	96.65% <100.00%> (-0.01%)	⬇️
gpu	96.65% <ø> (-0.01%)	⬇️
macOS	96.65% <100.00%> (-0.01%)	⬇️
pytest	96.65% <100.00%> (-0.01%)	⬇️
python3.6	95.56% <100.00%> (-0.01%)	⬇️
python3.8	96.62% <100.00%> (-0.01%)	⬇️
python3.9	96.51% <100.00%> (-0.01%)	⬇️
torch1.3.1	95.56% <100.00%> (-0.01%)	⬇️
torch1.4.0	95.68% <100.00%> (-0.01%)	⬇️
torch1.8.1	96.51% <100.00%> (-0.01%)	⬇️

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

Impacted Files	Coverage Δ
torchmetrics/setup_tools.py	93.54% <ø> (ø)
torchmetrics/classification/stat_scores.py	100.00% <100.00%> (ø)
torchmetrics/utilities/data.py	96.29% <100.00%> (-0.14%)	⬇️
__w/2/s/torchmetrics/utilities/data.py	96.29% <0.00%> (-0.14%)	⬇️
__w/2/s/torchmetrics/classification/stat_scores.py	100.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 263a4c7...781300e. Read the comment docs.

[Borda]

numpy is installed with PyTorch anyway and numpy is everywhere so what is the motivation?

[SkafteNicki]

agree with @Borda on this one, no need to remove numpy as requirement as it is already installed with torch

[maximsch2]

I don't think pytorch requires numpy to run, it's likely for tests as well. In fact, we are running pytorch in builds without numpy so it's for sure not required.

In this particular case, the core library doesn't really require numpy, so I'm thinking we should keep it lean especially given it's a no-cost thing for us. The two current uses for numpy in torchmetrics were:

• np.nan (replaced by float('nan') which is same
• helper function that was made to take np.ndarray so that the test will be used.

Both are not needed for metrics to function.

[ananthsub]

@SkafteNicki @Borda one of the concerns is around backwards compatibility for metrics. numpy introduces a large surface area which can make this harder to manage. given how minimal the usage is in torchmetrics now, we'd prefer to keep the library lean and use torch primitives instead to better control for this

[Borda]

Ok, I am fine with dropping it 🐰

[SkafteNicki]

@maximsch2 and @ananthsub great points, also fine with dropping it, did not really have a strong opinion on this :]