Updated handling of missing values with LightGBM, and added ability to use (0) as missing value

[mstfbl]

Fixes #4681 , and also adds the ability to use the numerical value (0) as missing value with LightGBM. In addition, I've also updated the baseline results for LightGbm as it is normal for them to change with CursOpt.AllFeatures, and all added a unit test checking that a previously trained LightGBM model with the CursOpt.Features flag produces the baselines it did before.

[mstfbl]

The current CI tests are failing due to the current generated netcoreapp core_manifest.json not exactly matching with the generated baseline core_manifest.json. The difference here is caused by the newly added UseZeroAsMissingValue variable. I am currently investigating how this baseline is generated, so that I can update the generated file to match the correct core_manifest.json.

Output matches baseline: '..\Common\EntryPoints\core_ep-list.tsv'
*** Failure: Output and baseline mismatch at line 11652, expected ' "Name": "MinimumExampleCountPerGroup",' but got ' "Name": "UseZeroAsMissingValue",' : '..\Common\EntryPoints\core_manifest.json'
Test EntryPointCatalog: completed normally: failed

[justinormont]

@mstfbl: We may want to document how to update the core_manifest.json and core_ep-list.tsv in the developer-guide.md.

I think the method is to unskip the RegenerateEntryPointCatalog test, run it, and push changes to the two files.

machinelearning/test/Microsoft.ML.Core.Tests/UnitTests/TestEntryPoints.cs

Lines 230 to 256 in 97c4533

	[Fact(Skip = "Execute this test if you want to regenerate the core_manifest and core_ep_list files")]
	public void RegenerateEntryPointCatalog()
	{
	var (epListContents, jObj) = BuildManifests();
	var buildPrefix = GetBuildPrefix();
	var epListFile = buildPrefix + "_ep-list.tsv";
	var entryPointsSubDir = Path.Combine("..", "Common", "EntryPoints");
	var catalog = Env.ComponentCatalog;
	var epListPath = GetBaselinePath(entryPointsSubDir, epListFile);
	DeleteOutputPath(epListPath);
	File.WriteAllLines(epListPath, epListContents);
	var manifestFile = buildPrefix + "_manifest.json";
	var manifestPath = GetBaselinePath(entryPointsSubDir, manifestFile);
	DeleteOutputPath(manifestPath);
	using (var file = File.OpenWrite(manifestPath))
	using (var writer = new StreamWriter(file))
	using (var jw = new JsonTextWriter(writer))
	{
	jw.Formatting = Formatting.Indented;
	jObj.WriteTo(jw);
	}
	}

[yaeldekel]

Another option is to run EntryPointCatalog locally, and then copy the files created in the bin directory to test\BaselineOutput\Common\EntryPoints.

---

In reply to: 579335552 [](ancestors = 579335552)

[justinormont]

LGTM

[ganik]

Pls try to run old LightGBM model (trained without your changes) with your new code. Baseline numbers should be the same.

[ganik]