Add half precision testing [1/n]

[SkafteNicki]

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?

Issue #57
Adds testing interface for half precision and apply to few metrics.
Follow up PRs will add the testing to remaining metrics and add documentation to metrics where it will not work.

Note, that we cannot really do anything about it not working for some of the metrics as some pytorch operations (e.g torch.log) does not support half precision + cpu.

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 🙃

[codecov]

Codecov Report

Merging #77 (bf78331) into master (64f3c12) will increase coverage by 0.21%.
The diff coverage is 100.00%.

❗ Current head bf78331 differs from pull request most recent head 569657e. Consider uploading reports for the commit 569657e to get more accurate results

@@            Coverage Diff             @@
##           master      #77      +/-   ##
==========================================
+ Coverage   96.67%   96.89%   +0.21%     
==========================================
  Files         148       74      -74     
  Lines        4632     2319    -2313     
==========================================
- Hits         4478     2247    -2231     
+ Misses        154       72      -82     

Flag	Coverage Δ
Linux	79.17% <60.00%> (-0.02%)	⬇️
Windows	79.17% <60.00%> (-0.02%)	⬇️
cpu	96.89% <100.00%> (+0.34%)	⬆️
gpu	?
macOS	96.89% <100.00%> (+0.34%)	⬆️
pytest	96.89% <100.00%> (+0.21%)	⬆️
python3.6	95.67% <100.00%> (-0.86%)	⬇️
python3.8	?
python3.9	96.89% <100.00%> (+0.34%)	⬆️
torch1.3.1	95.67% <100.00%> (+0.35%)	⬆️
torch1.4.0	?
torch1.8.1	96.89% <100.00%> (+0.34%)	⬆️

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

Impacted Files	Coverage Δ
...cs/functional/regression/mean_squared_log_error.py	100.00% <ø> (ø)
torchmetrics/regression/mean_squared_log_error.py	100.00% <ø> (ø)
torchmetrics/regression/psnr.py	100.00% <ø> (ø)
...etrics/functional/regression/mean_squared_error.py	100.00% <100.00%> (ø)
torchmetrics/functional/regression/psnr.py	96.96% <100.00%> (+0.09%)	⬆️
torchmetrics/utilities/imports.py	78.57% <100.00%> (+0.79%)	⬆️
...rics/functional/classification/hamming_distance.py
__w/1/s/torchmetrics/__init__.py
.../s/torchmetrics/classification/confusion_matrix.py
__w/1/s/torchmetrics/regression/ssim.py
... and 71 more

---

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 64f3c12...569657e. Read the comment docs.

[Borda]

can we rather use this function version? Lightning-AI/pytorch-lightning#6434

[SkafteNicki]

can we rather use this function version? PyTorchLightning/pytorch-lightning#6434

Of course :]

[tchaton]

LGTM !

[justusschock]

Not sure if we really should skip here or show a more meaningful error message.

[Borda]

@SkafteNicki ^^

[SkafteNicki]

@justusschock you got any proposal?

[ananyahjha93]

@justusschock more meaningful message in test or the class itself? wouldn't be easier to test for half precision tensors in the base metric class and check if torch version >= 1.6

[Borda]

@SkafteNicki mind resolve conflicts... :]