Aded catch in R^2 calculation for case with few samples

src/Microsoft.ML.Data/Evaluators/RegressionEvaluator.cs

@@ -102,7 +102,15 @@ public override double RSquared
                 {
                     get
                     {
-                        return SumWeights > 0 ? 1 - TotalL2Loss / (TotalLabelSquaredW - TotalLabelW * TotalLabelW / SumWeights) : 0;
+                        // RSquared value cannot be well-defined with less than two samples.
+                        // Return NaN instead of -Infinity.
+                        if (SumWeights > 0)
+                        {
+                            if ((TotalLabelSquaredW - TotalLabelW * TotalLabelW / SumWeights) == 0)
+                                return double.NaN;
+                            return 1 - TotalL2Loss / (TotalLabelSquaredW - TotalLabelW * TotalLabelW / SumWeights);
+                        }
+                        return 0;
                     }
                 }
 

test/Microsoft.ML.Functional.Tests/Validation.cs

@@ -12,6 +12,7 @@
 using Microsoft.ML.Trainers.LightGbm;
 using Xunit;
 using Xunit.Abstractions;
+using static Microsoft.ML.TrainCatalogBase;
 
 namespace Microsoft.ML.Functional.Tests
 {
@@ -138,5 +139,27 @@ public void TrainWithValidationSet()
             Common.AssertMetrics(trainMetrics);
             Common.AssertMetrics(validMetrics);
         }
+
+        /// <summary>
+        /// Test cross validation R^2 metric to return NaN when given fewer data
+        /// than needed to infer metric calculation. R^2 is NaN when given folds
+        /// with less than 2 rows of training data.
+        /// </summary>
+        [Fact]
+        public void TestCrossValidationResultsWithNotEnoughData()
+        {
+            var mlContext = new MLContext(1);
+            // Get data and set up sample regression pipeline.
+            var data = mlContext.Data.LoadFromTextFile<Iris>(TestCommon.GetDataPath(DataDir, TestDatasets.iris.trainFilename), hasHeader: true);
+            var dataFirstTenRows = mlContext.Data.TakeRows(data, 10);
+            var pipeline = mlContext.Transforms.Concatenate("Features", Iris.Features)
+                .Append(mlContext.Regression.Trainers.OnlineGradientDescent());
+
+            // Check that NaN is returned with fold given less than 2 rows of training data.
+            // With dataset of 10 rows, number of folds will be 6.
+            var cvResults = mlContext.Regression.CrossValidate(dataFirstTenRows, pipeline, numberOfFolds: 6);
+            foreach (CrossValidationResult<RegressionMetrics> result in cvResults)
+                Assert.Equal(double.NaN, result.Metrics.RSquared);
+        }
     }
 }