Exception using CollectionDataSource

[mjmckp]

System information

• Windows 10 Pro 64 Bit
• .NET framework 4.7
• libraries built from master branch

Issue

The code below (a self-contained repro) throws a null reference exception on line

machinelearning/src/Microsoft.ML.Api/DataViewConstructionUtils.cs

Line 233 in 83f9bac

var peek = DataView._peeks[index] as Peek<TRow, TDst>;

since peek is null when attempting to get the value of the Label column.

        public class Data
        {
            [ColumnName("Features")]
            [VectorType(2)]
            public float[] Features;

            [ColumnName("Label")]
            public bool Label;
        }

        public class Prediction
        {
            [ColumnName("PredictedLabel")]
            public bool PredictedLabel;
        }

        static void Train(IEnumerable<Data> data)
        {
            var pipeline = new LearningPipeline();
            pipeline.Add(CollectionDataSource.Create(data));
            pipeline.Add(new FastForestBinaryClassifier());
            var model = pipeline.Train<Data, Prediction>();
        }

        static void Main(string[] args)
        {
            var data = new Data[1];
            data[0] = new Data();
            data[0].Features = new float[] { 0.0f, 1.0f };
            data[0].Label = false;

            Train(data);
        }

[Ivanidzo4ka]

It sounds weird but we expect label column to be float value instead of bool. @TomFinley can probably add more details. Meanwhile, should we add converter inside pipeline? @GalOshri @shauheen

[mjmckp]

That's highly unintuitive given that the 1) label type, 2) predicted label type, and 3) classifier output type are all boolean.

[TomFinley]

Hi @Ivanidzo4ka are we confident in this diagnosis? I think the trainer can accept bool values?

[Ivanidzo4ka]

Should work now, addressed in #183.