Port LightGBM provider

[shengwangsw]

Port LightGBM provider from author Alberto Ferreira ([email protected]).

[codecov]

Codecov Report

Merging #26 into master will increase coverage by 1.49%.
The diff coverage is 80.90%.

@@             Coverage Diff              @@
##             master      #26      +/-   ##
============================================
+ Coverage     76.19%   77.69%   +1.49%     
- Complexity      268      357      +89     
============================================
  Files            28       39      +11     
  Lines           899     1318     +419     
  Branches         79      120      +41     
============================================
+ Hits            685     1024     +339     
- Misses          171      228      +57     
- Partials         43       66      +23     

Impacted Files	Coverage Δ	Complexity Δ
...i/openml/provider/lightgbm/LightGBMMLProvider.java	0.00% <0.00%> (ø)	0.00 <0.00> (?)
...er/lightgbm/LightGBMBinaryClassificationModel.java	57.14% <57.14%> (ø)	8.00 <8.00> (?)
...feedzai/openml/provider/lightgbm/LightGBMSWIG.java	72.13% <72.13%> (ø)	11.00 <11.00> (?)
...eedzai/openml/provider/lightgbm/SWIGResources.java	73.33% <73.33%> (ø)	7.00 <7.00> (?)
...openml/provider/lightgbm/LightGBMModelCreator.java	82.35% <82.35%> (ø)	20.00 <20.00> (?)
...i/openml/provider/lightgbm/SWIGTrainResources.java	83.78% <83.78%> (ø)	9.00 <9.00> (?)
...eedzai/openml/provider/lightgbm/LightGBMUtils.java	86.95% <86.95%> (ø)	4.00 <4.00> (?)
...tgbm/LightGBMBinaryClassificationModelTrainer.java	87.28% <87.28%> (ø)	23.00 <23.00> (?)
...enml/provider/lightgbm/LightGBMDescriptorUtil.java	96.96% <96.96%> (ø)	3.00 <3.00> (?)
...i/openml/provider/lightgbm/LightGBMAlgorithms.java	100.00% <100.00%> (ø)	3.00 <3.00> (?)
... and 12 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 de052b7...60902db. Read the comment docs.

[paulojrp]

Rename the folder of the new provider from openml-lightgbm-provider to openml-lightgbm

[ptraca]

@shenggwang I've finished the review, please see comments.

[AlbertoEAF]

It looks good @shenggwang.

In fact great work with the additions of "final" and fixes in the docstrings.

The visibility modifier changes should be reverted though.
We were extremely careful and purposefully expose only what's needed. For instance, many internal-use-only classes and properties are now public and user-visible, which should not happen.

I have one question for @JPDSousa , by dropping those TODO's, it means we'll have to maintain 2 versions of the provider? The open source and the internal?

[AlbertoEAF]

I have one final question, if we strip away the versions as above, then from now on how do we track the LightGBM provider version internally?

[shengwangsw]

I have one final question, if we strip away the versions as above, then from now on how do we track the LightGBM provider version internally?

I suppose that from now on, we are going to track internally the version of LightGBM on openml-java.

[ptraca]

I've talked this over with @JPDSousa and we need to also move this repo [email protected]:data-science/make-lightgbm.git to github. Maybe we could add it in this repo as a git sub-module replacing the lightgbmlib_builddir.

[AlbertoEAF]

I've talked this over with @JPDSousa and we need to also move this repo [email protected]:data-science/make-lightgbm.git to github. Maybe we could add it in this repo as a git sub-module replacing the lightgbmlib_builddir.

I'd rather keep it apart if possible @ptraca , the make-lightgbm generates a lot of trash and temporary files and stuff that is not needed here. Also, both repos are agnostic to each other, and make-lightgbm serves to build artifacts for any version of LightGBM and Java LightGBM implementation, not only ours.

[shengwangsw]

@ptraca @JPDSousa @AlbertoEAF @paulojrp
I think this is done. Take a look.

[paulojrp]

lgtm

good work