How to add multiple custom columns to IDataView in ML.NET - ml.net

Question

I'd like to add two custom columns after loading my IDataView from file. In each row, the new column values should be natural logarithm of the values.
My environment is:
Windows 10 Pro, Version 10.0.19043 Build 19043
ML.NET version 1.6.0
Microsoft Visual Studio Professional 2022 Preview (64-bit), Version 17.0.0 Preview 3.1
Here is what I have done so far. This is a standalone, reproducible program that demonstrates the problem.
using Microsoft.ML;
using Microsoft.ML.Transforms;
namespace TestLog {
public static class Program {
private class InputData {
public double Velocity { get; set; }
public double Thrust { get; set; }
}
private class CustomMappingOutput {
public double LogVelocity { get; set; }
public double LogThrust { get; set; }
}
private class TransformedData : InputData {
public double LogVelocity { get; set; }
public double LogThrust { get; set; }
}
[CustomMappingFactoryAttribute("LogVelocity")]
private class LogVelocityCustomAction : CustomMappingFactory<InputData, CustomMappingOutput> {
public static void CustomAction(InputData input, CustomMappingOutput output)
=> output.LogVelocity = (float) Math.Log(input.Velocity);
public override Action<InputData, CustomMappingOutput> GetMapping() => CustomAction;
}
[CustomMappingFactoryAttribute("LogThrust")]
private class LogThrustCustomAction : CustomMappingFactory<InputData, CustomMappingOutput> {
public static void CustomAction(InputData input, CustomMappingOutput output)
=> output.LogThrust = (float)Math.Log(input.Thrust);
public override Action<InputData, CustomMappingOutput> GetMapping() => CustomAction;
}
public static void Run() {
var mlContext = new MLContext();
var samples = new List<InputData> {
new InputData { Velocity= 0.006467, Thrust = 1.614237 },
new InputData { Velocity= 0.53451, Thrust = 1.068356 },
new InputData { Velocity= 0.278578, Thrust = 0.216861 },
new InputData { Velocity= 0.014179, Thrust = 0.119712 },
new InputData { Velocity= 0.392814, Thrust = 3.915486 }
};
var data = mlContext.Data.LoadFromEnumerable(samples);
var pipeline = mlContext.Transforms.CustomMapping(new LogVelocityCustomAction().GetMapping(),
contractName: "LogVelocity")
.Append(mlContext.Transforms.CustomMapping(new LogThrustCustomAction().GetMapping(),
contractName: "LogThrust"));
var transformer = pipeline.Fit(data);
// Now save the transform pipeline so that it can be reloaded by another process.
mlContext.Model.Save(transformer, data.Schema, "customTransform.zip");
// We load the saved transform and use it, as if it was in another program.
var loadedTransform = mlContext.Model.Load("customTransform.zip", out _);
// Now we can transform the data.
var transformedIDataView = loadedTransform.Transform(data);
var newDataEnumerable = mlContext.Data.CreateEnumerable<TransformedData>(transformedIDataView,
reuseRowObject: true);
var newIDataView = mlContext.Data.LoadFromEnumerable(newDataEnumerable);
// Save newIDataView as a CSV file so we can verify the transformations.
var path = #"../../../transformed.csv";
var mlContextSave = new MLContext();
using var stream = File.Create(path);
mlContextSave.Data.SaveAsText(newIDataView,
stream,
separatorChar: ',',
headerRow: true,
schema: false);
}
static void Main() {
Program.Run();
Console.WriteLine("Hit return to exit");
Console.ReadLine();
}
}
}
At end of the program, the transformed IDataView is saved as a “.csv” file.
It looks like:
As you can see, the “Velocity” column was not transformed.
Any help or suggestions will be greatly appreciated.
Charles

Answer 1

You're casting the log to float, then assigning it to a variable of type double. Have you tried casting to double?
I'm not sure why it's working for thrust but not velocity. Perhaps the very small number in the first line of velocity is causing an exception.