feat: add metric parameter to lightgbm learners

[imatiach-msft]

resolves #631 and #628

[imatiach-msft]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[codecov]

Codecov Report

Merging #672 into master will decrease coverage by 0.17%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #672      +/-   ##
==========================================
- Coverage    79.9%   79.72%   -0.18%     
==========================================
  Files         225      225              
  Lines        9033     9037       +4     
  Branches      483      484       +1     
==========================================
- Hits         7218     7205      -13     
- Misses       1815     1832      +17

Impacted Files	Coverage Δ
.../com/microsoft/ml/spark/lightgbm/TrainParams.scala	100% <100%> (ø)	⬆️
...m/microsoft/ml/spark/lightgbm/LightGBMParams.scala	85.71% <100%> (+0.62%)	⬆️
...m/microsoft/ml/spark/lightgbm/LightGBMRanker.scala	56.09% <100%> (ø)	⬆️
...icrosoft/ml/spark/lightgbm/LightGBMRegressor.scala	66.66% <100%> (ø)	⬆️
...crosoft/ml/spark/lightgbm/LightGBMClassifier.scala	86.74% <100%> (-1.35%)	⬇️
...a/com/microsoft/ml/spark/io/http/HTTPClients.scala	54.71% <0%> (-16.99%)	⬇️
...a/com/microsoft/ml/spark/lightgbm/TrainUtils.scala	87.94% <0%> (-2.24%)	⬇️
...osoft/ml/spark/io/http/PartitionConsolidator.scala	93.33% <0%> (-2.23%)	⬇️
...om/microsoft/ml/spark/lightgbm/LightGBMUtils.scala	95.29% <0%> (-1.18%)	⬇️

---

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 9805996...792fc34. Read the comment docs.

[mhamilton723]

amazing

[imatiach-msft]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[imatiach-msft]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[imatiach-msft]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[imatiach-msft]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[candalfigomoro]

@imatiach-msft
Does it support multiple evaluation metrics (just like LightGBM)?

[imatiach-msft]

@candalfigomoro yes it should, with comma separated metrics. For early stopping we also use all metrics specified (note early stopping is not provided in native api so R/python and our spark implementation all have their own implementation):
https://github.com/Azure/mmlspark/blob/master/src/main/scala/com/microsoft/ml/spark/lightgbm/TrainUtils.scala#L211

[dr-drunkensteinn]

@imatiach-msft
Does it support just ndcg or ndcg@N as well ?

[imatiach-msft]

@nishant-ai I think it should support all lightgbm metrics as specified here:
https://github.com/microsoft/LightGBM/blob/master/docs/Parameters.rst#metric
but for NDCG@N I believe we need to expose the parameter eval_at based on a quick look through the lightgbm docs:
https://github.com/microsoft/LightGBM/blob/master/docs/Parameters.rst#eval_at
specifically for early stopping, the scala version of lightgbm does use the correct comparison function here:
https://github.com/Azure/mmlspark/blob/master/src/main/scala/com/microsoft/ml/spark/lightgbm/TrainUtils.scala#L215

[dr-drunkensteinn]

Yes, I meant the eval_at parameter only. I had a look at the code and the eval_at parameter used for @n metrics didn't seem to be exposed, just wanted to confirm that.
Can you please make the changes needed so that NDCG@N can be used as well ? Thanks.

[dr-drunkensteinn]

Thanks @imatiach-msft !