Share via


StandardTrainersCatalog.PairwiseCoupling<TModel> Method

Definition

Create a PairwiseCouplingTrainer, which predicts a multiclass target using pairwise coupling strategy with the binary classification estimator specified by binaryEstimator.

public static Microsoft.ML.Trainers.PairwiseCouplingTrainer PairwiseCoupling<TModel> (this Microsoft.ML.MulticlassClassificationCatalog.MulticlassClassificationTrainers catalog, Microsoft.ML.Trainers.ITrainerEstimator<Microsoft.ML.ISingleFeaturePredictionTransformer<TModel>,TModel> binaryEstimator, string labelColumnName = "Label", bool imputeMissingLabelsAsNegative = false, Microsoft.ML.IEstimator<Microsoft.ML.ISingleFeaturePredictionTransformer<Microsoft.ML.Calibrators.ICalibrator>> calibrator = default, int maximumCalibrationExampleCount = 1000000000) where TModel : class;
static member PairwiseCoupling : Microsoft.ML.MulticlassClassificationCatalog.MulticlassClassificationTrainers * Microsoft.ML.Trainers.ITrainerEstimator<Microsoft.ML.ISingleFeaturePredictionTransformer<'Model>, 'Model (requires 'Model : null)> * string * bool * Microsoft.ML.IEstimator<Microsoft.ML.ISingleFeaturePredictionTransformer<Microsoft.ML.Calibrators.ICalibrator>> * int -> Microsoft.ML.Trainers.PairwiseCouplingTrainer (requires 'Model : null)
<Extension()>
Public Function PairwiseCoupling(Of TModel As Class) (catalog As MulticlassClassificationCatalog.MulticlassClassificationTrainers, binaryEstimator As ITrainerEstimator(Of ISingleFeaturePredictionTransformer(Of TModel), TModel), Optional labelColumnName As String = "Label", Optional imputeMissingLabelsAsNegative As Boolean = false, Optional calibrator As IEstimator(Of ISingleFeaturePredictionTransformer(Of ICalibrator)) = Nothing, Optional maximumCalibrationExampleCount As Integer = 1000000000) As PairwiseCouplingTrainer

Type Parameters

TModel

The type of the model. This type parameter will usually be inferred automatically from binaryEstimator.

Parameters

catalog
MulticlassClassificationCatalog.MulticlassClassificationTrainers

The multiclass classification catalog trainer object.

binaryEstimator
ITrainerEstimator<ISingleFeaturePredictionTransformer<TModel>,TModel>

An instance of a binary ITrainerEstimator<TTransformer,TModel> used as the base trainer.

labelColumnName
String

The name of the label column.

imputeMissingLabelsAsNegative
Boolean

Whether to treat missing labels as having negative labels, instead of keeping them missing.

calibrator
IEstimator<ISingleFeaturePredictionTransformer<ICalibrator>>

The calibrator. If a calibrator is not explicitly provided, it will default to Microsoft.ML.Calibrators.PlattCalibratorTrainer

maximumCalibrationExampleCount
Int32

Number of instances to train the calibrator.

Returns

Examples

using System;
using System.Collections.Generic;
using System.Linq;
using Microsoft.ML;
using Microsoft.ML.Data;

namespace Samples.Dynamic.Trainers.MulticlassClassification
{
    public static class PairwiseCoupling
    {
        public static void Example()
        {
            // Create a new context for ML.NET operations. It can be used for
            // exception tracking and logging, as a catalog of available operations
            // and as the source of randomness. Setting the seed to a fixed number
            // in this example to make outputs deterministic.
            var mlContext = new MLContext(seed: 0);

            // Create a list of training data points.
            var dataPoints = GenerateRandomDataPoints(1000);

            // Convert the list of data points to an IDataView object, which is
            // consumable by ML.NET API.
            var trainingData = mlContext.Data.LoadFromEnumerable(dataPoints);

            // Define the trainer.
            var pipeline =
                // Convert the string labels into key types.
                mlContext.Transforms.Conversion.MapValueToKey("Label")
                // Apply PairwiseCoupling multiclass meta trainer on top of
                // binary trainer.
                .Append(mlContext.MulticlassClassification.Trainers
                .PairwiseCoupling(
                mlContext.BinaryClassification.Trainers.SdcaLogisticRegression()));

            // Train the model.
            var model = pipeline.Fit(trainingData);

            // Create testing data. Use different random seed to make it different
            // from training data.
            var testData = mlContext.Data
                .LoadFromEnumerable(GenerateRandomDataPoints(500, seed: 123));

            // Run the model on test data set.
            var transformedTestData = model.Transform(testData);

            // Convert IDataView object to a list.
            var predictions = mlContext.Data
                .CreateEnumerable<Prediction>(transformedTestData,
                reuseRowObject: false).ToList();

            // Look at 5 predictions
            foreach (var p in predictions.Take(5))
                Console.WriteLine($"Label: {p.Label}, " +
                    $"Prediction: {p.PredictedLabel}");

            // Expected output:
            //   Label: 1, Prediction: 1
            //   Label: 2, Prediction: 2
            //   Label: 3, Prediction: 2
            //   Label: 2, Prediction: 2
            //   Label: 3, Prediction: 2

            // Evaluate the overall metrics
            var metrics = mlContext.MulticlassClassification
                .Evaluate(transformedTestData);

            PrintMetrics(metrics);

            // Expected output:
            //   Micro Accuracy: 0.90
            //   Macro Accuracy: 0.90
            //   Log Loss: 0.36
            //   Log Loss Reduction: 0.67

            //   Confusion table
            //             ||========================
            //   PREDICTED ||     0 |     1 |     2 | Recall
            //   TRUTH     ||========================
            //           0 ||   150 |     0 |    10 | 0.9375
            //           1 ||     0 |   166 |    11 | 0.9379
            //           2 ||    15 |    15 |   133 | 0.8160
            //             ||========================
            //   Precision ||0.9091 |0.9171 |0.8636 |
        }

        // Generates random uniform doubles in [-0.5, 0.5)
        // range with labels 1, 2 or 3.
        private static IEnumerable<DataPoint> GenerateRandomDataPoints(int count,
            int seed = 0)

        {
            var random = new Random(seed);
            float randomFloat() => (float)(random.NextDouble() - 0.5);
            for (int i = 0; i < count; i++)
            {
                // Generate Labels that are integers 1, 2 or 3
                var label = random.Next(1, 4);
                yield return new DataPoint
                {
                    Label = (uint)label,
                    // Create random features that are correlated with the label.
                    // The feature values are slightly increased by adding a
                    // constant multiple of label.
                    Features = Enumerable.Repeat(label, 20)
                        .Select(x => randomFloat() + label * 0.2f).ToArray()

                };
            }
        }

        // Example with label and 20 feature values. A data set is a collection of
        // such examples.
        private class DataPoint
        {
            public uint Label { get; set; }
            [VectorType(20)]
            public float[] Features { get; set; }
        }

        // Class used to capture predictions.
        private class Prediction
        {
            // Original label.
            public uint Label { get; set; }
            // Predicted label from the trainer.
            public uint PredictedLabel { get; set; }
        }

        // Pretty-print MulticlassClassificationMetrics objects.
        public static void PrintMetrics(MulticlassClassificationMetrics metrics)
        {
            Console.WriteLine($"Micro Accuracy: {metrics.MicroAccuracy:F2}");
            Console.WriteLine($"Macro Accuracy: {metrics.MacroAccuracy:F2}");
            Console.WriteLine($"Log Loss: {metrics.LogLoss:F2}");
            Console.WriteLine(
                $"Log Loss Reduction: {metrics.LogLossReduction:F2}\n");

            Console.WriteLine(metrics.ConfusionMatrix.GetFormattedConfusionTable());
        }
    }
}

Remarks

In the Pairwise coupling (PKPD) strategy, a binary classification algorithm is used to train one classifier for each pair of classes. Prediction is then performed by running these binary classifiers, and computing a score for each class by counting how many of the binary classifiers predicted it. The prediction is the class with the highest score.

Applies to