diff --git a/test/Microsoft.ML.Functional.Tests/Common.cs b/test/Microsoft.ML.Functional.Tests/Common.cs
index 5756893966..cd7f6757fc 100644
--- a/test/Microsoft.ML.Functional.Tests/Common.cs
+++ b/test/Microsoft.ML.Functional.Tests/Common.cs
@@ -9,6 +9,7 @@
using Microsoft.ML.Data;
using Microsoft.ML.Functional.Tests.Datasets;
using Xunit;
+using Xunit.Sdk;
namespace Microsoft.ML.Functional.Tests
{
@@ -268,6 +269,33 @@ public static void AssertMetricsStatistics(RegressionMetricsStatistics metrics)
AssertMetricStatistics(metrics.LossFunction);
}
+ /// <summary>
+ /// Assert that two float arrays are not equal.
+ /// </summary>
+ /// <param name="array1">An array of floats.</param>
+ /// <param name="array2">An array of floats.</param>
+ public static void AssertNotEqual(float[] array1, float[] array2)
+ {
+ Assert.NotNull(array1);
+ Assert.NotNull(array2);
+ Assert.Equal(array1.Length, array2.Length);
+
+ bool mismatch = false;
+ for (int i = 0; i < array1.Length; i++)
+ try
+ {
+ // Use Assert to test for equality rather than
+ // to roll our own float equality checker.
+ Assert.Equal(array1[i], array2[i]);
+ }
+ catch(EqualException)
+ {
+ mismatch = true;
+ break;
+ }
+ Assert.True(mismatch);
+ }
+
/// <summary>
/// Verify that a float array has no NaNs or infinities.
/// </summary>
diff --git a/test/Microsoft.ML.Functional.Tests/Training.cs b/test/Microsoft.ML.Functional.Tests/Training.cs
new file mode 100644
index 0000000000..c1162e7978
--- /dev/null
+++ b/test/Microsoft.ML.Functional.Tests/Training.cs
@@ -0,0 +1,529 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+using System;
+using System.Linq;
+using Microsoft.ML.Data;
+using Microsoft.ML.Functional.Tests.Datasets;
+using Microsoft.ML.RunTests;
+using Microsoft.ML.TestFramework;
+using Microsoft.ML.Trainers;
+using Microsoft.ML.Trainers.FastTree;
+using Xunit;
+using Xunit.Abstractions;
+
+namespace Microsoft.ML.Functional.Tests
+{
+ public class Training : BaseTestClass
+ {
+ public Training(ITestOutputHelper output) : base(output)
+ {
+ }
+
+ /// <summary>
+ /// Training: It is easy to compare trainer evaluations on the same dataset.
+ /// </summary>
+ [Fact]
+ public void CompareTrainerEvaluations()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ // Get the dataset.
+ var data = mlContext.Data.LoadFromTextFile<TweetSentiment>(GetDataPath(TestDatasets.Sentiment.trainFilename),
+ separatorChar: TestDatasets.Sentiment.fileSeparator,
+ hasHeader: TestDatasets.Sentiment.fileHasHeader,
+ allowQuoting: TestDatasets.Sentiment.allowQuoting);
+ var trainTestSplit = mlContext.Data.TrainTestSplit(data);
+ var trainData = trainTestSplit.TrainSet;
+ var testData = trainTestSplit.TestSet;
+
+ // Create a transformation pipeline.
+ var featurizationPipeline = mlContext.Transforms.Text.FeaturizeText("Features", "SentimentText")
+ .AppendCacheCheckpoint(mlContext);
+
+ // Create a selection of learners.
+ var sdcaTrainer = mlContext.BinaryClassification.Trainers.SdcaCalibrated(
+ new SdcaCalibratedBinaryClassificationTrainer.Options { NumberOfThreads = 1 });
+
+ var fastTreeTrainer = mlContext.BinaryClassification.Trainers.FastTree(
+ new FastTreeBinaryClassificationTrainer.Options { NumberOfThreads = 1 });
+
+ var ffmTrainer = mlContext.BinaryClassification.Trainers.FieldAwareFactorizationMachine();
+
+ // Fit the data transformation pipeline.
+ var featurization = featurizationPipeline.Fit(trainData);
+ var featurizedTrain = featurization.Transform(trainData);
+ var featurizedTest = featurization.Transform(testData);
+
+ // Fit the trainers.
+ var sdca = sdcaTrainer.Fit(featurizedTrain);
+ var fastTree = fastTreeTrainer.Fit(featurizedTrain);
+ var ffm = ffmTrainer.Fit(featurizedTrain);
+
+ // Evaluate the trainers.
+ var sdcaPredictions = sdca.Transform(featurizedTest);
+ var sdcaMetrics = mlContext.BinaryClassification.EvaluateNonCalibrated(sdcaPredictions);
+ var fastTreePredictions = fastTree.Transform(featurizedTest);
+ var fastTreeMetrics = mlContext.BinaryClassification.EvaluateNonCalibrated(fastTreePredictions);
+ var ffmPredictions = sdca.Transform(featurizedTest);
+ var ffmMetrics = mlContext.BinaryClassification.EvaluateNonCalibrated(ffmPredictions);
+
+ // Validate the results.
+ Common.AssertMetrics(sdcaMetrics);
+ Common.AssertMetrics(fastTreeMetrics);
+ Common.AssertMetrics(ffmMetrics);
+ }
+
+ /// <summary>
+ /// Training: Models can be trained starting from an existing model.
+ /// </summary>
+ [Fact]
+ public void ContinueTrainingAveragePerceptron()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ // Get the dataset.
+ var data = mlContext.Data.LoadFromTextFile<TweetSentiment>(GetDataPath(TestDatasets.Sentiment.trainFilename),
+ separatorChar: TestDatasets.Sentiment.fileSeparator,
+ hasHeader: TestDatasets.Sentiment.fileHasHeader,
+ allowQuoting: TestDatasets.Sentiment.allowQuoting);
+
+ // Create a transformation pipeline.
+ var featurizationPipeline = mlContext.Transforms.Text.FeaturizeText("Features", "SentimentText")
+ .AppendCacheCheckpoint(mlContext);
+
+ var trainer = mlContext.BinaryClassification.Trainers.AveragedPerceptron(
+ new AveragedPerceptronTrainer.Options { NumberOfIterations = 1 });
+
+ // Fit the data transformation pipeline.
+ var featurization = featurizationPipeline.Fit(data);
+ var featurizedData = featurization.Transform(data);
+
+ // Fit the first trainer.
+ var firstModel = trainer.Fit(featurizedData);
+ var firstModelWeights = firstModel.Model.Weights;
+
+ // Fist the first trainer again.
+ var firstModelPrime = trainer.Fit(featurizedData);
+ var firstModelWeightsPrime = firstModel.Model.Weights;
+
+ // Fit the second trainer.
+ var secondModel = trainer.Fit(featurizedData, firstModel.Model);
+ var secondModelWeights = secondModel.Model.Weights;
+
+ // Validate that continued training occurred.
+ // Training from the same initial condition, same seed should create the same model.
+ Common.AssertEqual(firstModelWeights.ToArray(), firstModelWeightsPrime.ToArray());
+ // Continued training should create a different model.
+ Common.AssertNotEqual(firstModelWeights.ToArray(), secondModelWeights.ToArray());
+ }
+
+ /// <summary>
+ /// Training: Models can be trained starting from an existing model.
+ /// </summary>
+ [Fact]
+ public void ContinueTrainingFieldAwareFactorizationMachine()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ // Get the dataset.
+ var data = mlContext.Data.LoadFromTextFile<TweetSentiment>(GetDataPath(TestDatasets.Sentiment.trainFilename),
+ separatorChar: TestDatasets.Sentiment.fileSeparator,
+ hasHeader: TestDatasets.Sentiment.fileHasHeader,
+ allowQuoting: TestDatasets.Sentiment.allowQuoting);
+
+ // Create a transformation pipeline.
+ var featurizationPipeline = mlContext.Transforms.Text.FeaturizeText("Features", "SentimentText")
+ .AppendCacheCheckpoint(mlContext);
+
+ var trainer = mlContext.BinaryClassification.Trainers.FieldAwareFactorizationMachine(
+ new FieldAwareFactorizationMachineTrainer.Options { NumberOfIterations = 100 });
+
+ // Fit the data transformation pipeline.
+ var featurization = featurizationPipeline.Fit(data);
+ var featurizedData = featurization.Transform(data);
+
+ // Fit the first trainer.
+ var firstModel = trainer.Fit(featurizedData);
+ var firstModelWeights = firstModel.Model.GetLinearWeights();
+
+ // Fist the first trainer again.
+ var firstModelPrime = trainer.Fit(featurizedData);
+ var firstModelWeightsPrime = firstModel.Model.GetLinearWeights();
+
+ // Fit the second trainer.
+ var secondModel = trainer.Fit(featurizedData, modelParameters: firstModel.Model);
+ var secondModelWeights = secondModel.Model.GetLinearWeights();
+
+ // Validate that continued training occurred.
+ // Training from the same initial condition, same seed should create the same model.
+ Assert.Equal(firstModelWeights, firstModelWeightsPrime);
+ // Continued training should create a different model.
+ Assert.NotEqual(firstModelWeights, secondModelWeights);
+ }
+
+ /// <summary>
+ /// Training: Models can be trained starting from an existing model.
+ /// </summary>
+ [Fact]
+ public void ContinueTrainingLinearSupportVectorMachine()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ // Get the dataset.
+ var data = mlContext.Data.LoadFromTextFile<TweetSentiment>(GetDataPath(TestDatasets.Sentiment.trainFilename),
+ separatorChar: TestDatasets.Sentiment.fileSeparator,
+ hasHeader: TestDatasets.Sentiment.fileHasHeader,
+ allowQuoting: TestDatasets.Sentiment.allowQuoting);
+
+ // Create a transformation pipeline.
+ var featurizationPipeline = mlContext.Transforms.Text.FeaturizeText("Features", "SentimentText")
+ .AppendCacheCheckpoint(mlContext);
+
+ var trainer = mlContext.BinaryClassification.Trainers.LinearSvm(
+ new LinearSvmTrainer.Options { NumberOfIterations = 1 });
+
+ // Fit the data transformation pipeline.
+ var featurization = featurizationPipeline.Fit(data);
+ var featurizedData = featurization.Transform(data);
+
+ // Fit the first trainer.
+ var firstModel = trainer.Fit(featurizedData);
+ var firstModelWeights = firstModel.Model.Weights;
+
+ // Fist the first trainer again.
+ var firstModelPrime = trainer.Fit(featurizedData);
+ var firstModelWeightsPrime = firstModel.Model.Weights;
+
+ // Fit the second trainer.
+ var secondModel = trainer.Fit(featurizedData, firstModel.Model);
+ var secondModelWeights = secondModel.Model.Weights;
+
+ // Validate that continued training occurred.
+ // Training from the same initial condition, same seed should create the same model.
+ Common.AssertEqual(firstModelWeights.ToArray(), firstModelWeightsPrime.ToArray());
+ // Continued training should create a different model.
+ Common.AssertNotEqual(firstModelWeights.ToArray(), secondModelWeights.ToArray());
+ }
+
+ /// <summary>
+ /// Training: Models can be trained starting from an existing model.
+ /// </summary>
+ [Fact]
+ public void ContinueTrainingLogisticRegression()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ // Get the dataset.
+ var data = mlContext.Data.LoadFromTextFile<TweetSentiment>(GetDataPath(TestDatasets.Sentiment.trainFilename),
+ separatorChar: TestDatasets.Sentiment.fileSeparator,
+ hasHeader: TestDatasets.Sentiment.fileHasHeader,
+ allowQuoting: TestDatasets.Sentiment.allowQuoting);
+
+ // Create a transformation pipeline.
+ var featurizationPipeline = mlContext.Transforms.Text.FeaturizeText("Features", "SentimentText")
+ .AppendCacheCheckpoint(mlContext);
+
+ var trainer = mlContext.BinaryClassification.Trainers.LogisticRegression(
+ new LogisticRegressionBinaryClassificationTrainer.Options { NumberOfThreads = 1, MaximumNumberOfIterations = 10 });
+
+ // Fit the data transformation pipeline.
+ var featurization = featurizationPipeline.Fit(data);
+ var featurizedData = featurization.Transform(data);
+
+ // Fit the first trainer.
+ var firstModel = trainer.Fit(featurizedData);
+ var firstModelWeights = firstModel.Model.SubModel.Weights;
+
+ // Fist the first trainer again.
+ var firstModelPrime = trainer.Fit(featurizedData);
+ var firstModelWeightsPrime = firstModel.Model.SubModel.Weights;
+
+ // Fit the second trainer.
+ var secondModel = trainer.Fit(featurizedData, firstModel.Model.SubModel);
+ var secondModelWeights = secondModel.Model.SubModel.Weights;
+
+ // Validate that continued training occurred.
+ // Training from the same initial condition, same seed should create the same model.
+ Common.AssertEqual(firstModelWeights.ToArray(), firstModelWeightsPrime.ToArray());
+ // Continued training should create a different model.
+ Common.AssertNotEqual(firstModelWeights.ToArray(), secondModelWeights.ToArray());
+ }
+
+ /// <summary>
+ /// Training: Models can be trained starting from an existing model.
+ /// </summary>
+ [Fact]
+ public void ContinueTrainingLogisticRegressionMulticlass()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ var data = mlContext.Data.LoadFromTextFile<Iris>(GetDataPath(TestDatasets.iris.trainFilename),
+ hasHeader: TestDatasets.iris.fileHasHeader,
+ separatorChar: TestDatasets.iris.fileSeparator);
+
+ // Create a training pipeline.
+ var featurizationPipeline = mlContext.Transforms.Concatenate("Features", Iris.Features)
+ .Append(mlContext.Transforms.Conversion.MapValueToKey("Label"))
+ .AppendCacheCheckpoint(mlContext);
+
+ var trainer = mlContext.MulticlassClassification.Trainers.LogisticRegression(
+ new LogisticRegressionMulticlassClassificationTrainer.Options { NumberOfThreads = 1, MaximumNumberOfIterations = 10 });
+
+ // Fit the data transformation pipeline.
+ var featurization = featurizationPipeline.Fit(data);
+ var featurizedData = featurization.Transform(data);
+
+ // Fit the first trainer.
+ var firstModel = trainer.Fit(featurizedData);
+ VBuffer<float>[] firstModelWeights = null;
+ firstModel.Model.GetWeights(ref firstModelWeights, out int firstModelNumClasses);
+
+ // Fist the first trainer again.
+ var firstModelPrime = trainer.Fit(featurizedData);
+ VBuffer<float>[] firstModelWeightsPrime = null;
+ firstModel.Model.GetWeights(ref firstModelWeightsPrime, out int firstModelNumClassesPrime);
+
+ // Fit the second trainer.
+ var secondModel = trainer.Fit(featurizedData, firstModel.Model);
+ VBuffer<float>[] secondModelWeights = null;
+ secondModel.Model.GetWeights(ref secondModelWeights, out int secondModelNumClasses);
+
+ // Validate that continued training occurred.
+ // Training from the same initial condition, same seed should create the same model.
+ Assert.Equal(firstModelNumClasses, firstModelNumClassesPrime);
+ for (int i = 0; i < firstModelNumClasses; i++)
+ Common.AssertEqual(firstModelWeights[i].DenseValues().ToArray(), firstModelWeightsPrime[i].DenseValues().ToArray());
+ // Continued training should create a different model.
+ Assert.Equal(firstModelNumClasses, secondModelNumClasses);
+ for (int i = 0; i < firstModelNumClasses; i++)
+ Common.AssertNotEqual(firstModelWeights[i].DenseValues().ToArray(), secondModelWeights[i].DenseValues().ToArray());
+ }
+
+ /// <summary>
+ /// Training: Models can be trained starting from an existing model.
+ /// </summary>
+ [Fact]
+ public void ContinueTrainingOnlineGradientDescent()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ // Get the dataset.
+ var data = mlContext.Data.LoadFromTextFile<HousingRegression>(GetDataPath(TestDatasets.housing.trainFilename),
+ separatorChar: TestDatasets.housing.fileSeparator,
+ hasHeader: TestDatasets.housing.fileHasHeader);
+
+ // Create a transformation pipeline.
+ var featurizationPipeline = mlContext.Transforms.Concatenate("Features", HousingRegression.Features)
+ .Append(mlContext.Transforms.Normalize("Features"))
+ .AppendCacheCheckpoint(mlContext);
+
+ var trainer = mlContext.Regression.Trainers.OnlineGradientDescent(
+ new OnlineGradientDescentTrainer.Options { NumberOfIterations = 10 });
+
+ // Fit the data transformation pipeline.
+ var featurization = featurizationPipeline.Fit(data);
+ var featurizedData = featurization.Transform(data);
+
+ // Fit the first trainer.
+ var firstModel = trainer.Fit(featurizedData);
+ var firstModelWeights = firstModel.Model.Weights;
+
+ // Fist the first trainer again.
+ var firstModelPrime = trainer.Fit(featurizedData);
+ var firstModelWeightsPrime = firstModel.Model.Weights;
+
+ // Fit the second trainer.
+ var secondModel = trainer.Fit(featurizedData, firstModel.Model);
+ var secondModelWeights = secondModel.Model.Weights;
+
+ // Validate that continued training occurred.
+ // Training from the same initial condition, same seed should create the same model.
+ Common.AssertEqual(firstModelWeights.ToArray(), firstModelWeightsPrime.ToArray());
+ // Continued training should create a different model.
+ Common.AssertNotEqual(firstModelWeights.ToArray(), secondModelWeights.ToArray());
+ }
+
+ /// <summary>
+ /// Training: Models can be trained starting from an existing model.
+ /// </summary>
+ [Fact]
+ public void ContinueTrainingPoissonRegression()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ // Get the dataset.
+ var data = mlContext.Data.LoadFromTextFile<HousingRegression>(GetDataPath(TestDatasets.housing.trainFilename),
+ separatorChar: TestDatasets.housing.fileSeparator,
+ hasHeader: TestDatasets.housing.fileHasHeader);
+
+ // Create a transformation pipeline.
+ var featurizationPipeline = mlContext.Transforms.Concatenate("Features", HousingRegression.Features)
+ .Append(mlContext.Transforms.Normalize("Features"))
+ .AppendCacheCheckpoint(mlContext);
+
+ var trainer = mlContext.Regression.Trainers.PoissonRegression(
+ new PoissonRegressionTrainer.Options { NumberOfThreads = 1, MaximumNumberOfIterations = 100 });
+
+ // Fit the data transformation pipeline.
+ var featurization = featurizationPipeline.Fit(data);
+ var featurizedData = featurization.Transform(data);
+
+ // Fit the first trainer.
+ var firstModel = trainer.Fit(featurizedData);
+ var firstModelWeights = firstModel.Model.Weights;
+
+ // Fist the first trainer again.
+ var firstModelPrime = trainer.Fit(featurizedData);
+ var firstModelWeightsPrime = firstModel.Model.Weights;
+
+ // Fit the second trainer.
+ var secondModel = trainer.Fit(featurizedData, firstModel.Model);
+ var secondModelWeights = secondModel.Model.Weights;
+
+ // Validate that continued training occurred.
+ // Training from the same initial condition, same seed should create the same model.
+ Common.AssertEqual(firstModelWeights.ToArray(), firstModelWeightsPrime.ToArray());
+ // Continued training should create a different model.
+ Common.AssertNotEqual(firstModelWeights.ToArray(), secondModelWeights.ToArray());
+ }
+
+ /// <summary>
+ /// Training: Models can be trained starting from an existing model.
+ /// </summary>
+ [Fact]
+ public void ContinueTrainingSymbolicStochasticGradientDescent()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ // Get the dataset.
+ var data = mlContext.Data.LoadFromTextFile<TweetSentiment>(GetDataPath(TestDatasets.Sentiment.trainFilename),
+ separatorChar: TestDatasets.Sentiment.fileSeparator,
+ hasHeader: TestDatasets.Sentiment.fileHasHeader,
+ allowQuoting: TestDatasets.Sentiment.allowQuoting);
+
+ // Create a transformation pipeline.
+ var featurizationPipeline = mlContext.Transforms.Text.FeaturizeText("Features", "SentimentText")
+ .AppendCacheCheckpoint(mlContext);
+
+ var trainer = mlContext.BinaryClassification.Trainers.SymbolicSgd(
+ new SymbolicSgdTrainer.Options
+ {
+ NumberOfThreads = 1,
+ NumberOfIterations = 10
+ });
+
+ // Fit the data transformation pipeline.
+ var featurization = featurizationPipeline.Fit(data);
+ var featurizedData = featurization.Transform(data);
+
+ // Fit the first trainer.
+ var firstModel = trainer.Fit(featurizedData);
+ var firstModelWeights = firstModel.Model.SubModel.Weights;
+
+ // Fist the first trainer again.
+ var firstModelPrime = trainer.Fit(featurizedData);
+ var firstModelWeightsPrime = firstModel.Model.SubModel.Weights;
+
+ // Fit the second trainer.
+ var secondModel = trainer.Fit(featurizedData, firstModel.Model.SubModel);
+ var secondModelWeights = secondModel.Model.SubModel.Weights;
+
+ // Validate that continued training occurred.
+ // Training from the same initial condition, same seed should create the same model.
+ Common.AssertEqual(firstModelWeights.ToArray(), firstModelWeightsPrime.ToArray());
+ // Continued training should create a different model.
+ Common.AssertNotEqual(firstModelWeights.ToArray(), secondModelWeights.ToArray());
+ }
+
+ /// <summary>
+ /// Training: Meta-compononts function as expected. For OVA (one-versus-all), a user will be able to specify only
+ /// binary classifier trainers. If they specify a different model class there should be a compile error.
+ /// </summary>
+ [Fact]
+ public void MetacomponentsFunctionAsExpectedOva()
+ {
+ var mlContext = new MLContext(seed: 1);
+
+ var data = mlContext.Data.LoadFromTextFile<Iris>(GetDataPath(TestDatasets.iris.trainFilename),
+ hasHeader: TestDatasets.iris.fileHasHeader,
+ separatorChar: TestDatasets.iris.fileSeparator);
+
+ // Create a model training an OVA trainer with a binary classifier.
+ var anomalyDetectionTrainer = mlContext.AnomalyDetection.Trainers.RandomizedPca();
+ var anomalyDetectionPipeline = mlContext.Transforms.Concatenate("Features", Iris.Features)
+ .AppendCacheCheckpoint(mlContext)
+ .Append(mlContext.Transforms.Conversion.MapValueToKey("Label"))
+ .Append(mlContext.MulticlassClassification.Trainers.OneVersusAll(anomalyDetectionTrainer))
+ .Append(mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel"));
+
+ // Fit the binary classification pipeline.
+ Assert.Throws<InvalidOperationException>(() => anomalyDetectionPipeline.Fit(data));
+
+ // Create a model training an OVA trainer with a binary classifier.
+ var binaryclassificationTrainer = mlContext.BinaryClassification.Trainers.LogisticRegression(
+ new LogisticRegressionBinaryClassificationTrainer.Options { MaximumNumberOfIterations = 10, NumberOfThreads = 1, });
+ var binaryClassificationPipeline = mlContext.Transforms.Concatenate("Features", Iris.Features)
+ .AppendCacheCheckpoint(mlContext)
+ .Append(mlContext.Transforms.Conversion.MapValueToKey("Label"))
+ .Append(mlContext.MulticlassClassification.Trainers.OneVersusAll(binaryclassificationTrainer));
+
+ // Fit the binary classification pipeline.
+ var binaryClassificationModel = binaryClassificationPipeline.Fit(data);
+
+ // Transform the data
+ var binaryClassificationPredictions = binaryClassificationModel.Transform(data);
+
+ // Evaluate the model.
+ var binaryClassificationMetrics = mlContext.MulticlassClassification.Evaluate(binaryClassificationPredictions);
+
+ // Create a model training an OVA trainer with a clustering trainer.
+ var kmeansTrainer = mlContext.Clustering.Trainers.KMeans(
+ new KMeansTrainer.Options { MaximumNumberOfIterations = 10, NumberOfThreads = 1, });
+
+ Assert.Throws<ArgumentOutOfRangeException>(() =>
+ mlContext.Transforms.Concatenate("Features", Iris.Features)
+ .AppendCacheCheckpoint(mlContext)
+ .Append(mlContext.Transforms.Conversion.MapValueToKey("Label"))
+ .Append(mlContext.MulticlassClassification.Trainers.OneVersusAll(kmeansTrainer))
+ .Append(mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel")));
+
+ // Create a model training an OVA trainer with a multiclass classification trainer.
+ var multiclassTrainer = mlContext.MulticlassClassification.Trainers.LogisticRegression(
+ new LogisticRegressionMulticlassClassificationTrainer.Options { MaximumNumberOfIterations = 10, NumberOfThreads = 1, });
+ Assert.Throws<ArgumentOutOfRangeException>(() =>
+ mlContext.Transforms.Concatenate("Features", Iris.Features)
+ .AppendCacheCheckpoint(mlContext)
+ .Append(mlContext.Transforms.Conversion.MapValueToKey("Label"))
+ .Append(mlContext.MulticlassClassification.Trainers.OneVersusAll(multiclassTrainer))
+ .Append(mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel")));
+
+ // Create a model training an OVA trainer with a ranking trainer.
+ var rankingTrainer = mlContext.Ranking.Trainers.FastTree(
+ new FastTreeRankingTrainer.Options { NumberOfTrees = 2, NumberOfThreads = 1, });
+ // Todo #2920: Make this fail somehow.
+ var rankingPipeline = mlContext.Transforms.Concatenate("Features", Iris.Features)
+ .AppendCacheCheckpoint(mlContext)
+ .Append(mlContext.Transforms.Conversion.MapValueToKey("Label"))
+ .Append(mlContext.MulticlassClassification.Trainers.OneVersusAll(rankingTrainer))
+ .Append(mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel"));
+
+ // Fit the invalid pipeline.
+ Assert.Throws<ArgumentOutOfRangeException>(() => rankingPipeline.Fit(data));
+
+ // Create a model training an OVA trainer with a regressor.
+ var regressionTrainer = mlContext.Regression.Trainers.PoissonRegression(
+ new PoissonRegressionTrainer.Options { MaximumNumberOfIterations = 10, NumberOfThreads = 1, });
+ // Todo #2920: Make this fail somehow.
+ var regressionPipeline = mlContext.Transforms.Concatenate("Features", Iris.Features)
+ .AppendCacheCheckpoint(mlContext)
+ .Append(mlContext.Transforms.Conversion.MapValueToKey("Label"))
+ .Append(mlContext.MulticlassClassification.Trainers.OneVersusAll(regressionTrainer))
+ .Append(mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel"));
+
+ // Fit the invalid pipeline.
+ Assert.Throws<ArgumentOutOfRangeException>(() => regressionPipeline.Fit(data));
+ }
+ }
+}
\ No newline at end of file
diff --git a/test/Microsoft.ML.Tests/Scenarios/Api/Estimators/Metacomponents.cs b/test/Microsoft.ML.Tests/Scenarios/Api/Estimators/Metacomponents.cs
deleted file mode 100644
index 9e9411a2f9..0000000000
--- a/test/Microsoft.ML.Tests/Scenarios/Api/Estimators/Metacomponents.cs
+++ /dev/null
@@ -1,37 +0,0 @@
-// Licensed to the .NET Foundation under one or more agreements.
-// The .NET Foundation licenses this file to you under the MIT license.
-// See the LICENSE file in the project root for more information.
-
-using Microsoft.ML.Data;
-using Microsoft.ML.RunTests;
-using Microsoft.ML.Trainers;
-using Microsoft.ML.Transforms;
-using Xunit;
-
-namespace Microsoft.ML.Tests.Scenarios.Api
-{
- public partial class ApiScenariosTests
- {
- /// <summary>
- /// Meta-components: Meta-components (for example, components that themselves instantiate components) should not be booby-trapped.
- /// When specifying what trainer OVA should use, a user will be able to specify any binary classifier.
- /// If they specify a regression or multi-class classifier ideally that should be a compile error.
- /// </summary>
- [Fact]
- public void Metacomponents()
- {
- var ml = new MLContext();
- var data = ml.Data.LoadFromTextFile<IrisData>(GetDataPath(TestDatasets.irisData.trainFilename), separatorChar: ',');
-
- var sdcaTrainer = ml.BinaryClassification.Trainers.SdcaNonCalibrated(
- new SdcaNonCalibratedBinaryClassificationTrainer.Options { MaximumNumberOfIterations = 100, Shuffle = true, NumberOfThreads = 1, });
-
- var pipeline = new ColumnConcatenatingEstimator (ml, "Features", "SepalLength", "SepalWidth", "PetalLength", "PetalWidth")
- .Append(ml.Transforms.Conversion.MapValueToKey("Label"), TransformerScope.TrainTest)
- .Append(ml.MulticlassClassification.Trainers.OneVersusAll(sdcaTrainer))
- .Append(ml.Transforms.Conversion.MapKeyToValue(("PredictedLabel")));
-
- var model = pipeline.Fit(data);
- }
- }
-}
\ No newline at end of file