Scores to Label mapping

src/Microsoft.ML.Core/Data/ITransformModel.cs

@@ -25,6 +25,13 @@ public interface ITransformModel
         /// </summary>
         ISchema InputSchema { get; }
 
+        /// <summary>
+        /// The resulting schema once applied to this model. The <see cref="InputSchema"/> might have
+        /// columns that are not needed by this transform and these columns will be seen in the 
+        /// <see cref="OutputSchema"/> produced by this transform.
+        /// </summary>
+        ISchema OutputSchema { get; }
+
         /// <summary>
         /// Apply the transform(s) in the model to the given input data.
         /// </summary>

src/Microsoft.ML.Data/EntryPoints/TransformModel.cs

@@ -39,10 +39,14 @@ public sealed class TransformModel : ITransformModel
         /// if transform model A needs column X and model B needs Y, that is NOT produced by A,
         /// then trimming A's input schema would cause composition to fail.
         /// </summary>
-        public ISchema InputSchema
-        {
-            get { return _schemaRoot; }
-        }
+        public ISchema InputSchema => _schemaRoot;
+
+        /// <summary>
+        /// The resulting schema once applied to this model. The <see cref="InputSchema"/> might have
+        /// columns that are not needed by this transform and these columns will be seen in the 
+        /// <see cref="OutputSchema"/> produced by this transform.
+        /// </summary>
+        public ISchema OutputSchema => _chain.Schema;
 
         /// <summary>
         /// Create a TransformModel containing the transforms from "result" back to "input".

src/Microsoft.ML/PredictionModel.cs

@@ -6,6 +6,7 @@
 using Microsoft.ML.Runtime.Api;
 using Microsoft.ML.Runtime.Data;
 using Microsoft.ML.Runtime.EntryPoints;
+using Microsoft.ML.Runtime.Internal.Utilities;
 using System;
 using System.Collections.Generic;
 using System.IO;
@@ -29,6 +30,39 @@ internal Runtime.EntryPoints.TransformModel PredictorModel
             get { return _predictorModel; }
         }
 
+        /// <summary>
+        /// Returns labels that correspond to indices of the score array in the case of 
+        /// multi-class classification problem.
+        /// </summary>
+        /// <param name="mapping">Label to score mapping</param>
+        /// <param name="scoreColumnName">Name of the score column</param>
+        /// <returns></returns>
+        public bool TryGetScoreLabelMapping(out string[] mapping, string scoreColumnName = DefaultColumnNames.Score)
+        {
+            mapping = null;
+            ISchema schema = _predictorModel.OutputSchema;
+            int colIndex = -1;
+            if (!schema.TryGetColumnIndex(scoreColumnName, out colIndex))
+                return false;
+
+            int expectedLabelCount = schema.GetColumnType(colIndex).AsVector.ValueCount;
+            if (!schema.HasSlotNames(colIndex, expectedLabelCount))
+                return false;
+
+            VBuffer<DvText> labels = default;
+            schema.GetMetadata(MetadataUtils.Kinds.SlotNames, colIndex, ref labels);
+            VBufferUtils.Densify(ref labels);
+
+            if (labels.Length != expectedLabelCount)
+                return false;
+
+            mapping = new string[labels.Length];
+            for (int index = 0; index < labels.Count; index++)
+                mapping[index] = labels.Values[index].ToString();
+
+            return true;
+        }
+
         /// <summary>
         /// Read model from file asynchronously.
         /// </summary>

test/Microsoft.ML.Tests/Scenarios/IrisPlantClassificationWithStringLabelTests.cs

@@ -30,6 +30,14 @@ public void TrainAndPredictIrisModelWithStringLabelTest()
             pipeline.Add(new StochasticDualCoordinateAscentClassifier());
 
             PredictionModel<IrisDataWithStringLabel, IrisPrediction> model = pipeline.Train<IrisDataWithStringLabel, IrisPrediction>();
+            string[] scoreLabels;
+            model.TryGetScoreLabelMapping(out scoreLabels);
+
+            Assert.NotNull(scoreLabels);
+            Assert.Equal(3, scoreLabels.Length);
+            Assert.Equal("Iris-setosa", scoreLabels[0]);
+            Assert.Equal("Iris-versicolor", scoreLabels[1]);
+            Assert.Equal("Iris-virginica", scoreLabels[2]);
 
             IrisPrediction prediction = model.Predict(new IrisDataWithStringLabel()
             {