Exceptions are not super helpful

[Ivanidzo4ka]

var loader = mlContext.Data.CreateTextLoader(new TextLoader.Options()
            {
                HasHeader = true,
                Columns = new[] {
                    new TextLoader.Column("NumericFeatures", DataKind.Single, 0, 189),
                    new TextLoader.Column("CatFeatures", DataKind.String, 189, 229),
                    new TextLoader.Column("Label", DataKind.Int32, 230)
                }
            });
            var data = loader.Load("train.tsv");
            var dic = new Dictionary<int, bool>();
            dic.Add(1, true);
            dic.Add(-1, false);
            var pipeline = mlContext.Transforms.IndicateMissingValues("NumericFeatures").
                Append(mlContext.Transforms.Categorical.OneHotEncoding("CatFeatures")).
                Append(mlContext.Transforms.Concatenate("Features","NumericFeatures", "CatFeatures")).
                Append(mlContext.Transforms.Conversion.MapValue("Label", dic));

System.InvalidOperationException: 'Column 'CatFeatures' has values of Singlewhich is not the same as earlier observed type of Boolean.'

A) would be nice to have space between type and sentence.
B) this exception happened in Concat transform and only indication of that is stack trace.

[TomFinley]

Good point on the message. Related to one of the sub-items raised by #2514. I disagree however that reproducing the information already in the stack trace is the job of an exception message. The facility for telling where an exception was thrown is the stacktrace. So I don't view it as problematic that the only thing that shows where an exception was thrown is the stack trace, since that is more or less the entire purpose why the stack trace exists and is part of every exception.

[zeahmed]

System.InvalidOperationException: 'Column 'CatFeatures' has values of Singlewhich is not the same as earlier observed type of Boolean.'

Do you know why it says 'Boolean' at the end?

[Ivanidzo4ka]

Yep.
So I have NumericFeatures which is array of floats.
I apply IndicateMissingValues to it, and in my head I'm doing NAHandle, but in reality it just creates array of boolean indicators. Which I'm combining with cat features.

Properly it should be IndicateMissingValues("MissingIndicator", "NumericFeatures"). TypeConversion("MissingIndicator", DataKind.Single). Concat("Features","MissingIndicator", "NumericFeatures","CatFeatures")