No ONNX model export support for {FixedPlatt, Isotonic, Naive}CalibratorEstimator(s)

[mstfbl]

For tracking purposes. The following CalibratorEstimators do not have ONNX model conversion support:

• FixedPlattCalibratorEstimator
• IsotonicCalibratorEstimator
• NaiveCalibratorEstimator

PlattCalibratorEstimator does have ONNX export support. This issue will be closed when the remaining calibrator estimators have ONNX export support, and are being tested for ONNX conversion in the manner below as PlattCalibratorEstimator is being tested:

machinelearning/test/Microsoft.ML.Tests/OnnxConversionTest.cs

Lines 265 to 300 in 81357ba

	public void PlattCalibratorOnnxConversionTest()
	{
	var mlContext = new MLContext(seed: 1);
	string dataPath = GetDataPath("breast-cancer.txt");
	// Now read the file (remember though, readers are lazy, so the actual reading will happen when the data is accessed).
	var dataView = mlContext.Data.LoadFromTextFile<BreastCancerBinaryClassification>(dataPath, separatorChar: '\t', hasHeader: true);
	List<IEstimator<ITransformer>> estimators = new List<IEstimator<ITransformer>>()
	{
	mlContext.BinaryClassification.Trainers.AveragedPerceptron(),
	mlContext.BinaryClassification.Trainers.FastForest(),
	mlContext.BinaryClassification.Trainers.FastTree(),
	mlContext.BinaryClassification.Trainers.LbfgsLogisticRegression(),
	mlContext.BinaryClassification.Trainers.LinearSvm(),
	mlContext.BinaryClassification.Trainers.Prior(),
	mlContext.BinaryClassification.Trainers.SdcaLogisticRegression(),
	mlContext.BinaryClassification.Trainers.SdcaNonCalibrated(),
	mlContext.BinaryClassification.Trainers.SgdCalibrated(),
	mlContext.BinaryClassification.Trainers.SgdNonCalibrated(),
	mlContext.BinaryClassification.Trainers.SymbolicSgdLogisticRegression(),
	};
	if (Environment.Is64BitProcess)
	{
	estimators.Add(mlContext.BinaryClassification.Trainers.LightGbm());
	}
	var initialPipeline = mlContext.Transforms.ReplaceMissingValues("Features").
	Append(mlContext.Transforms.NormalizeMinMax("Features"));
	foreach (var estimator in estimators)
	{
	var pipeline = initialPipeline.Append(estimator).Append(mlContext.BinaryClassification.Calibrators.Platt());
	var onnxFileName = $"{estimator}-WithPlattCalibrator.onnx";
	TestPipeline(pipeline, dataView, onnxFileName, new ColumnComparison[] { new ColumnComparison("Score", 3), new ColumnComparison("PredictedLabel"), new ColumnComparison("Probability", 3) });
	}
	Done();
	}

[antoniovs1029]

For the record, support for onnx export of PlattCalibrator was added on the PR:
#4699

EDIT: After a quick look, it seems that doing this for PlattCalibrator was straightforward because it already had its own implementation of how to export it to Onnx. But for the other calibrators, I think it would also be necessary to make that implementation. (Note: I'm not sure about Naive or Isotonic, but at least FixedPlatt might be able to share the same logic to save as Onnx as the one used for Platt).

[mstfbl]

Update: Merged PR #5289 has added and verified ONNX export support for the Naive and FixedPlatt calibrator estimators. PR #5298 is set to add ONNX export support for the Isotonic calibrator estimator.