Added exception for NaN eigenvectors on PCA (#5349)

Added exception for NaN eigenvectors on PCA
dotnet · Aug 17, 2020 · 1b1872f · 1b1872f
1 parent 2933216
commit 1b1872f
Show file tree

Hide file tree

Showing 2 changed files with 46 additions and 4 deletions.
diff --git a/src/Microsoft.ML.PCA/PcaTrainer.cs b/src/Microsoft.ML.PCA/PcaTrainer.cs
@@ -461,6 +461,9 @@ internal PcaModelParameters(IHostEnvironment env, int rank, float[][] eigenVecto
             {
                 _eigenVectors[i] = new VBuffer<float>(eigenVectors[i].Length, eigenVectors[i]);
                 _meanProjected[i] = VectorUtils.DotProduct(in _eigenVectors[i], in mean);
+                Host.CheckParam(_eigenVectors[i].GetValues().All(FloatUtils.IsFinite),
+                    nameof(eigenVectors),
+                    "The learnt eigenvectors contained NaN values, consider modifying the dataset or lower the rank or oversampling parameters");
             }
 
             _mean = mean;

diff --git a/test/Microsoft.ML.Tests/AnomalyDetectionTests.cs b/test/Microsoft.ML.Tests/AnomalyDetectionTests.cs
@@ -174,10 +174,10 @@ private static void ExecutePipelineWithGivenRandomizedPcaTrainer(MLContext mlCon
         }
 
 
-        /// <summary>
-        /// Help function used to execute trainers defined in <see cref="RandomizedPcaInMemory"/>.
-        /// </summary>
-        private static void ExecuteRandomizedPcaTrainerChangeThreshold(MLContext mlContext, Trainers.RandomizedPcaTrainer trainer)
+            /// <summary>
+            /// Help function used to execute trainers defined in <see cref="RandomizedPcaInMemory"/>.
+            /// </summary>
+            private static void ExecuteRandomizedPcaTrainerChangeThreshold(MLContext mlContext, Trainers.RandomizedPcaTrainer trainer)
         {
             var samples = new List<DataPoint>()
             {
@@ -251,5 +251,44 @@ private IDataView DetectAnomalyInMnistOneClass(string trainPath, string testPath
             var model = trainer.Fit(trainData);
             return model.Transform(testData);
         }
+
+        /// <summary>
+        /// Check that when PCA created invalid eigenvectors with NaNs a readable exception message is thrown.
+        /// </summary>
+        [Fact]
+
+        public void PcaTrainerInvalidEigenvectorsException()
+        {
+            var mlContext = new MLContext(seed: 0);
+
+            var trainer = mlContext.AnomalyDetection.Trainers.RandomizedPca(
+                featureColumnName: nameof(DataPoint.Features), rank: 3);
+
+            var samples = new List<DataPoint>()
+            {
+                new DataPoint(){ Features = new float[3] {1, 0, 2} },
+                new DataPoint(){ Features = new float[3] {2, 0, 4} },
+                new DataPoint(){ Features = new float[3] {4, 0, 8} },
+                new DataPoint(){ Features = new float[3] {8, 0, 16} }
+            };
+
+            var data = mlContext.Data.LoadFromEnumerable(samples);
+
+            bool exceptionThrown = false;
+            try
+            {
+                // Since we provided a dataset where all rows are linearly dependent,
+                // the PCA algorithm will likely fail when extracting 3 eigenvectors
+                // and produce eigenvectors with NaN.
+                var model = trainer.Fit(data);
+            }
+            catch(ArgumentOutOfRangeException ex)
+            {
+                exceptionThrown = true;
+                Assert.Contains("The learnt eigenvectors contained NaN values", ex.Message);
+            }
+
+            Assert.True(exceptionThrown);
+        }
     }
 }