R4 label works with some but not all binary classification trainers #2750
Description
This code:
using System;
using Microsoft.ML.Data;
namespace Microsoft.ML.Samples
{
internal static class Program
{
static void Main(string[] args)
{
var context = new MLContext();
var options = new TextLoader.Options()
{
Columns = new TextLoader.Column[]
{
new TextLoader.Column("Label", DataKind.Single, 0),
new TextLoader.Column("Sentiment", DataKind.String, 1)
},
HasHeader = true
};
var loader = context.Data.CreateTextLoader(options);
var data = loader.Read(@"C:\AutoMLDotNet\src\Samples\Data\wikipedia-detox-250-line-data.tsv");
var estimator = context.Transforms.Text.FeaturizeText("Features", "Sentiment")
.Append(context.BinaryClassification.Trainers.AveragedPerceptron());
var transformer = estimator.Fit(data);
var scoredData = transformer.Transform(data);
var metrics = context.BinaryClassification.EvaluateNonCalibrated(scoredData);
Console.WriteLine(metrics.Accuracy);
Console.WriteLine("Press any key...");
Console.ReadLine();
}
}
}works great!
If you replace AveragedPerceptron with LogisticRegression, it throws the exception:
'Schema mismatch for label column: expected Bool, got R4'
This may indicate a bug in label schema validation with some binary learners