Add deterministic option for LightGBM

[michaelgsharp]

Fixes #7326.

Adds the LightGBM deterministic option to the LightGBM Options.

[Copilot]

Pull Request Overview

This PR adds a deterministic option along with related options for the LightGBM trainer to ensure reproducible training outcomes. Key changes include:

• Adding new options (Deterministic, ForceRowWise, and ForceColumnWise) in the LightGBM trainer options.
• Updating the options mapping dictionary in the trainer base.
• Updating tests to set the new options for LightGBM estimators.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/Microsoft.ML.LightGbm/LightGbmTrainerBase.cs	Added dictionary entries and properties for deterministic options.
test/Microsoft.ML.Tests/TrainerEstimators/TreeEstimators.cs	Updated tests to initialize Deterministic and ForceRowWise options.

Comments suppressed due to low confidence (2)

src/Microsoft.ML.LightGbm/LightGbmTrainerBase.cs:243

• [nitpick] Consider clarifying the help text for 'Deterministic' to state that it ensures reproducible training outcomes, rather than mentioning 'stable results'.

/// Setting this to true should ensure the stable results when using the same data and the same parameters and different num_threads.

test/Microsoft.ML.Tests/TrainerEstimators/TreeEstimators.cs:69

• Consider adding a test case that explicitly sets and verifies the behavior of the 'ForceColumnWise' option, as it is a new addition not covered in the existing tests.

Deterministic = true,

[tarekgh]

public bool Deterministic = false;

I know this class is exposing fields directly, but I am wondering can the new added stuff be properties?

[michaelgsharp]

I believe we use reflection to instantiate these (like when you run this from ML.NET command line), and if its expecting fields it wouldn't work. I will double check on that before I merge this in.

If it is doing that, its possible that we could change things to either do both or only do properites and update all the options as well.

[ericstj]

Microsoft.ML.RunTests.TestEntryPoints.EntryPointCatalog is failing on all legs. https://helixr1107v0xdeko0k025g8.blob.core.windows.net/dotnet-machinelearning-refs-pull-7415-merge-504196c4ce1c401a92/Microsoft.ML.Core.Tests/1/console.55b4bee6.log?helixlogtype=result

�[m�[30;1m      Output:
�[m�[37m        Comparing /private/tmp/helix/working/A4860978/w/ADD50A36/e/TestOutput/../Common/EntryPoints/core_ep-list.tsv and /tmp/helix/working/A4860978/p/test/BaselineOutput/Common/EntryPoints/core_ep-list.tsv
�[m�[37m        Output matches baseline: '../Common/EntryPoints/core_ep-list.tsv'
�[m�[37m        Comparing /private/tmp/helix/working/A4860978/w/ADD50A36/e/TestOutput/../Common/EntryPoints/core_manifest.json and /tmp/helix/working/A4860978/p/test/BaselineOutput/Common/EntryPoints/netcoreapp/core_manifest.json
�[m�[37m        *** Failure #1: Output and baseline mismatch at line 11895, expected '          "Name": "ParallelTrainer",' but got '          "Name": "Deterministic",' : '../Common/EntryPoints/core_manifest.json'

Looks like you need to update core_ep-list.tsv @michaelgsharp

[michaelgsharp]

@ericstj good catch. Its been updated.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 68.96%. Comparing base (c36975c) to head (cd49ab2).
Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7415      +/-   ##
==========================================
- Coverage   68.97%   68.96%   -0.01%     
==========================================
  Files        1481     1481              
  Lines      273696   273708      +12     
  Branches    28285    28285              
==========================================
- Hits       188782   188769      -13     
- Misses      77526    77546      +20     
- Partials     7388     7393       +5     

Flag	Coverage Δ
Debug	68.96% <100.00%> (-0.01%)	⬇️
production	63.26% <100.00%> (-0.01%)	⬇️
test	89.46% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/Microsoft.ML.LightGbm/LightGbmTrainerBase.cs	80.30% <100.00%> (-0.03%)	⬇️
...osoft.ML.Tests/TrainerEstimators/TreeEstimators.cs	97.85% <100.00%> (+<0.01%)	⬆️

... and 10 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[michaelgsharp]

/ba-g failed tests are known failures and build analysis still isn't configured correctly to go green.