lightweight explained variance

[thomasgaudelet]

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?

Propose a lightweight version of explained variance without storing all predictions and ground truth up to compute.

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 🙃

[SkafteNicki]

@thomasgaudelet could you please explain the intention behind this PR?

[thomasgaudelet]

Sorry thought this was sufficient Propose a lightweight version of explained variance without storing all predictions and ground truth up to compute.

The issue with the current version of explained variance is that it stores all predictions and ground truths until the compute step (as shown by the RankZero warning). This is fine for small datasets but it doesn't scale. The PR replaces storing potentially large vectors with just tracking 5 sums. It's in a similar spirit to the current implementation of MSE and the likes.

[SkafteNicki]

@thomasgaudelet, thanks for the extra explaination, great improvement :]
Could you please check the failing tests?

[codecov]

Codecov Report

Merging #68 (ee45fb3) into master (179117b) will decrease coverage by 18.44%.
The diff coverage is 96.66%.

@@             Coverage Diff             @@
##           master      #68       +/-   ##
===========================================
- Coverage   97.07%   78.62%   -18.45%     
===========================================
  Files         118       59       -59     
  Lines        3898     1951     -1947     
===========================================
- Hits         3784     1534     -2250     
- Misses        114      417      +303     

Flag	Coverage Δ
Linux	78.62% <96.66%> (+0.07%)	⬆️
Windows	78.62% <96.66%> (+0.07%)	⬆️
cpu	78.62% <96.66%> (-18.45%)	⬇️
gpu	?
macOS	78.62% <96.66%> (-18.45%)	⬇️
pytest	78.62% <96.66%> (-18.45%)	⬇️
python3.6	?
python3.8	?
python3.9	?
torch1.3.1	?
torch1.4.0	?
torch1.8.0	?

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

Impacted Files	Coverage Δ
...etrics/functional/regression/explained_variance.py	90.62% <94.11%> (-9.38%)	⬇️
torchmetrics/regression/explained_variance.py	96.00% <100.00%> (+0.34%)	⬆️
torchmetrics/utilities/distributed.py	38.23% <0.00%> (-58.83%)	⬇️
torchmetrics/classification/auc.py	45.00% <0.00%> (-55.00%)	⬇️
torchmetrics/functional/classification/auroc.py	45.90% <0.00%> (-39.35%)	⬇️
torchmetrics/metric.py	55.07% <0.00%> (-39.16%)	⬇️
torchmetrics/functional/regression/psnr.py	61.29% <0.00%> (-35.49%)	⬇️
...chmetrics/functional/classification/stat_scores.py	64.91% <0.00%> (-35.09%)	⬇️
torchmetrics/classification/checks.py	65.44% <0.00%> (-34.56%)	⬇️
torchmetrics/functional/classification/accuracy.py	66.66% <0.00%> (-33.34%)	⬇️
... and 81 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 179117b...ee45fb3. Read the comment docs.

[thomasgaudelet]

@SkafteNicki done, there was a couple things mishandled but seems to pass all now

[SkafteNicki]

Could you also add entry under the Changed section in changelog?
Something like:
"Changed ExplainedVariance from storing all preds/targets to tracking 5 statistics..."

[SkafteNicki]

LGTM