ONNX training

gistfile1.txt

using System.Collections.Generic;
using System.Linq;
using Microsoft.ML;
using Microsoft.ML.Data;
using Microsoft.ML.Transforms;
using Microsoft.ML.Trainers.LightGbm;
using System;
using System.IO;
using Newtonsoft.Json;
using System.Text.RegularExpressions;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;
using Microsoft.ML.OnnxRuntime;

namespace LGBMRanker
{
    public class Trainer
    {
        MLContext mlContext;
        IDataView transformedTestData;
        IDataView onnxTransformedTestData;

        public Trainer()
        {
            // Create a new context for ML.NET operations. It can be used for
            // exception tracking and logging, as a catalog of available operations
            // and as the source of randomness. Setting the seed to a fixed number
            // in this example to make outputs deterministic.
            this.mlContext = new MLContext(seed: 0);

            //Load Data
            IDataView data = mlContext.Data.LoadFromTextFile<FinalMixtureData>("./data/train.tsv", separatorChar: '\t', hasHeader: true);

            // Define trainer options.
            var options = new LightGbmRankingTrainer.Options
            {
                LearningRate = 0.1,
                NumberOfIterations = 100,
                MinimumExampleCountPerGroup = 1,
                NumberOfLeaves = 31,
                NumberOfThreads = -1,
                MinimumExampleCountPerLeaf = 20,
                EarlyStoppingRound = 0,
                CategoricalSmoothing = 10,
                EvaluationMetric = LightGbmRankingTrainer.Options.EvaluateMetricType.NormalizedDiscountedCumulativeGain,
                HandleMissingValue = true,
                Sigmoid = 1,

                Booster = new GradientBooster.Options
                {
                    FeatureFraction = 1,
                },
                FeatureColumnName = "Feature",
                RowGroupColumnName = "ImpressionIdKey",
                LabelColumnName = "BackProClick"
            };

            // Define Data Prep Estimator
            // 1. Concatenate Size and Historical into a single feature vector output to a new column called Features
            // 2. Normalize Features vector
            IEstimator<ITransformer> dataPrepEstimator = mlContext.Transforms.Categorical.OneHotEncoding(
                    new[]
                    {
                        new InputOutputColumnPair("...", "..."),
                        new InputOutputColumnPair("...", "...")
                    })
                .Append(mlContext.Transforms.Conversion.MapValueToKey(
                    new[]
                    {
                        new InputOutputColumnPair("...", "..."),
                    }, keyOrdinality: Microsoft.ML.Transforms.ValueToKeyMappingEstimator
                    .KeyOrdinality.ByValue))
                .Append(mlContext.Transforms.Concatenate("Feature", "...", "..."))
                ;

            // Create data prep transformer
            ITransformer dataPrepTransformer = dataPrepEstimator.Fit(data);

            // Apply transforms to training data
            IDataView transformedTrainingData = dataPrepTransformer.Transform(data);

            // Define the trainer.
            var pipeline = mlContext.Ranking.Trainers.LightGbm(options);

            // Train the model.
            var model = pipeline.Fit(colSelTrainingData2);

            using (FileStream fs = new FileStream("./model.onnx", FileMode.Truncate, FileAccess.Write))
            {
                mlContext.Model.ConvertToOnnx(model, colSelTrainingData2, fs);
            }
        }
    }

    public class Program
    {
        public static void Main(string[] args)
        {
            //var summary = BenchmarkRunner.Run(typeof(Trainer).Assembly);
            new Trainer();
        }
    }
}