min max wrapper

[janhenriklambrechts]

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?

Fixes #49

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 🙃

[Borda]

@SkafteNicki seems to be a duplicate to #506

[Borda]

@janhenriklambrechts how is it going here, ready for review?

[janhenriklambrechts]

@Borda not yet finished, but will work on it this afternoon -- so expect a clean PR by tomorrow :)

[codecov]

Codecov Report

Merging #556 (297943b) into master (93cb842) will decrease coverage by 0%.
The diff coverage is 94%.

@@          Coverage Diff          @@
##           master   #556   +/-   ##
=====================================
- Coverage      95%    95%   -0%     
=====================================
  Files         152    153    +1     
  Lines        5362   5396   +34     
=====================================
+ Hits         5102   5134   +32     
- Misses        260    262    +2     

[janhenriklambrechts]

@Borda @SkafteNicki, I have identified the difference between tests/wrappers/test_minmax.py::compare_fn() and tests/helpers/testers.py::_class_test() that makes the tests fail. However, my understanding of ddp is quite limited so I need some help to understand what the correct behaviour is:

For, compare_fn(), the base_metric gets correctly calculated per batch.
For our test over 10 batches, the raw values of the metric look like this:

0.375
0.34375
0.2916666567325592
0.265625
0.2750000059604645
0.2604166567325592
0.25
0.24609375
0.2465277761220932
0.24375000596046448

The corresponding raw; min; max-array looks like this:

[array(0.24375, dtype=float32), array(0.24375, dtype=float32), array(0.375, dtype=float32)]

These results make sense to me.
However, if we run it within the tests/helpers/testers.py::_class_test(), our raw batch_result for WorldSize=2 is:

0.3438
0.2656
0.2604
0.2461
0.2438

These values correspond to the ones above, but omit half and thus lead to wrong min and max values.
Could you point me to the right direction in what causes this disparity? And how I could fix this?

[Borda]

@SkafteNicki mind have look ^^ 🐰