Add CatBoostClassifier/CatBoostRegressor Docs & Fix badge links

[tylerjthomas9]

I added documentation for the python CatBoostClassifier/CatBoostRegressor models. Additionally, the beacon-biosignals links were updated with JuliaAI, so hopefully, all the badges will work again.

#19

[tylerjthomas9]

@ablaom The MLJTestInterface tests are failing because the data is being split down the middle, so only one label is being passed to the CatBoostClassifier in the binary case (using the MLJTestInterface.make_binary() data). This seems to be a new addition to MLJTestInterface. Should I shuffle the data on the catboost side to avoid this, or is this a MLJTestInterface change?

JuliaAI/MLJTestInterface.jl@702c7b9

[ablaom]

Ah. Thanks for reporting this. Ideally, binary classifiers should be able to handle a level not manifest in the training target (but present in the pool) but I understand this can be a pain to fix. One could predict the seen class every time, with probabability one, if probabilistic (and skip the call to fit the core model). Although this ordinarily isn't a problem on larger datasets, I would encourage you to post an issue to have this remedied at some point. In the meantime one could either mark the test as broken (probably better practice) or hack a fix as you suggest.

Some other models failed in this way and we fixed them a little while back. CatBoost slipped through on the old version of the tests (not sure why) but the even stricter tests just released are catching the issue. By the way, I tested that the stricter tests are okay for all registered models, so am inclined to keep the new stricter tests, as I still believe they are correct.

[codecov-commenter]

Codecov Report

Merging #20 (4159f20) into main (bde0968) will increase coverage by 3.95%.
The diff coverage is 100.00%.

📣 This organization is not using Codecov’s GitHub App Integration. We recommend you install it so Codecov can continue to function properly for your repositories. Learn more

@@            Coverage Diff             @@
##             main      #20      +/-   ##
==========================================
+ Coverage   73.61%   77.57%   +3.95%     
==========================================
  Files           6        6              
  Lines         163      165       +2     
==========================================
+ Hits          120      128       +8     
+ Misses         43       37       -6     

Impacted Files	Coverage Δ
src/MLJCatBoostInterface.jl	79.66% <ø> (+10.16%)	⬆️
src/CatBoost.jl	100.00% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[tylerjthomas9]

I just added a temporary fix to ensure the tests work across the platforms. Setting the test to broken didn't fully fix it, so I assume the Python Exception is breaking the model fit, and not being caught. Here is the test table when setting the test to broken.

Test Summary:             | Pass  Error  Broken  Total   Time
MLJ Interface             |    1      1       1      3  53.6s
  MLJ Examples            |                       None   9.8s
  CatBoostClassifier      |                       None   0.4s
  CatBoostRegressor       |                       None   0.1s
  generic interface tests |    1      1       1      3  43.3s
    CatBoostRegressor     |    1                     1  31.5s
    CatBoostClassifier    |           1       1      2  11.8s

[ablaom]

I just added a temporary fix to ensure the tests work across the platforms.

Perhaps @ericphanson will want to comment / review. I've looked over the PR and it looks otherwise good, thanks.

In any case, I suggest posting an issue to revert the test when #22 is addressed, or you can add a comment there.

[ablaom]

Good to go.

[ablaom]

I think a real maintainer should merge this. @ericphanson or @tylerjthomas9 ?

[tylerjthomas9]

I don't have merge permissions. However, this still looks good to me. I am struggling to think of a good way to implement the single class case for the classifier (#22), but that can be tackled later.

[ericphanson]

Sounds like this was included in #23. Thanks both!