FairGBM openml API implementation

[AndreFCruz]

Implementation of the Java openml API for the C++ FairGBM algorithm (to be open-sourced soon).

As FairGBM was built on top of LightGBM, I just hijacked the current LightGBM openml API implementation and added a new algorithm descriptor for the new FairGBM parameters.

Other considerations:

• this implies an upgrade of the LightGBM version from 3.0.0.99 (at github.com/feedzai/LightGBM) to 3.2.1.99 (at github.com/feedzai/FairGBM)
• all LightGBM tests pass on this new version
• we're missing FairGBM-specific tests
• see this issue for why we changed some LightGBM test files

[AndreFCruz]

Note: this CI build fail is related to the FairGBM repo not being public as of yet; I'll have to change the make-lightgbm submodule to allow building from a local repo or using a token.

[AndreFCruz]

Also, everything seems to work regarding backwards compatibility when loading old LightGBM model files. We needed to test this because the model.txt representation changed with the bump in LightGBM version.

Tested it using the following notebook: test-lightgbm-retrocompatible-model-loading.zip

[AndreFCruz]

I'll review all this feedback today.

One other thing: we were asked to have FairGBM as the first model on the Pulse UI model choices, followed by LightGBM and every other model.

[codecov[bot]]

Codecov Report

Merging #116 (05165d9) into master (de77c62) will increase coverage by 0.30%. The diff coverage is 85.18%.

@@             Coverage Diff              @@
##             master     #116      +/-   ##
============================================
+ Coverage     79.90%   80.21%   +0.30%     
- Complexity      428      465      +37     
============================================
  Files            43       47       +4     
  Lines          1498     1607     +109     
  Branches        138      157      +19     
============================================
+ Hits           1197     1289      +92     
- Misses          224      231       +7     
- Partials         77       87      +10     

Impacted Files	Coverage Δ
...enml/provider/lightgbm/LightGBMDescriptorUtil.java	96.77% <0.00%> (-0.20%)	:arrow_down:
...ai/openml/provider/lightgbm/FairGBMMLProvider.java	40.00% <40.00%> (ø)
...i/openml/provider/lightgbm/LightGBMMLProvider.java	40.00% <50.00%> (+40.00%)	:arrow_up:
...i/openml/provider/lightgbm/AlgoDescriptorUtil.java	66.66% <66.66%> (ø)
...tgbm/LightGBMBinaryClassificationModelTrainer.java	83.72% <82.69%> (-3.15%)	:arrow_down:
...eedzai/openml/provider/lightgbm/SWIGTrainData.java	84.74% <84.00%> (-4.45%)	:arrow_down:
...penml/provider/lightgbm/FairGBMDescriptorUtil.java	93.75% <93.75%> (ø)
...enml/provider/lightgbm/FairGBMParamParserUtil.java	100.00% <100.00%> (ø)
...i/openml/provider/lightgbm/LightGBMAlgorithms.java	100.00% <100.00%> (ø)
...openml/provider/lightgbm/LightGBMModelCreator.java	87.23% <100.00%> (ø)
... and 4 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 de77c62...05165d9. Read the comment docs.

[AndreFCruz]

Are we ready to merge this? :tada:

[gandola]

LGTM! :1st_place_medal:

Cheers