Add Explainability to ModelBuilder

[chunky]

System Information (please complete the following information):
$ mlnet --version
16.2.0+511cc1082bef7a4bbb2f83ad88c1c425c932079c
$ dotnet --version
3.1.402
$ lsb_release -d
Description: Ubuntu 19.10

Describe the bug
I've trained up a regression algorithm using AutoML on Linux, and am now trying to use Permutation Feature Importance, from the documentation. [URLs in "additional context", below]

Things that are problematic to try and follow:

• Given that modelbuilder gave me a ModelInput, it's unclear how to use a ModelInput as the data: parameter. An elegant way to turn a ModelInput, or a collection of them, into an IDataView would be nice. Currently the only path I've found is to re-instantiate a fresh IDataView from my original CSV data using mlContext.Data.LoadFromFile [code that I lifted from the created ModelBuilder.cs]
• Getting the predictionTransformer parameter is currently an awkward cast [which I was only able to figure out by opening a debugger and checking the types] followed by .LastTransformer. Presumably this isn't going to be copy-pastable to the next model, if modelbuilder chose a different best model.
• I still haven't figured out how to map what PFI says, to the original parameters in the IDataView [or ModelInput]

Expected behavior
Personally I think the perfect product here would be one or more of a few things, any of which would improve my ability to access the functionality:

• An additional .cs file in the wizard-created Sample that can do PFI on the created model [this would be ideal, and probably the least invasive change!]
• Additional functions to instantiate a PermutationFeatureImportance that could take a TransformerChain instead of just the singular last transformer [current workflow also assumes that LastTransformer is even the important part of the chain]
• A way to take an autocreated "ModelInput" and end with an IDataView.
• A way to map ModelInput members to feature indices in the final weights/indices from PFI.

Additional context
https://docs.microsoft.com/en-us/dotnet/machine-learning/how-to-guides/explain-machine-learning-model-permutation-feature-importance-ml-net
and
https://docs.microsoft.com/en-us/dotnet/api/microsoft.ml.permutationfeatureimportanceextensions.permutationfeatureimportance?view=ml-dotnet#Microsoft_ML_PermutationFeatureImportanceExtensions_PermutationFeatureImportance__1_Microsoft_ML_RegressionCatalog_Microsoft_ML_ISingleFeaturePredictionTransformer___0__Microsoft_ML_IDataView_System_String_System_Boolean_System_Nullable_System_Int32__System_Int32_

[LittleLittleCloud]

@JakeRadMSFT @briacht for broadcasting

[JakeRadMSFT]

@chunky I totally agree. I actually helped investigate adding PFI to Model Builder output (Still in progress). Let me share some code that might help!

[JakeRadMSFT]

I'll add some more notes to the code soon! Most of the code you need can be found in "CalculatePFI" function and "CreateModel"

using System;
using System.Collections.Generic;
using System.IO;
using System.Linq;
using Microsoft.ML;
using Microsoft.ML.Data;
using PfiSample.Model;
using System.Collections.Immutable;

namespace PfiSample.ConsoleApp
{
    public static class ModelBuilder
    {
        private static string TRAIN_DATA_FILEPATH = @"C:\Users\anvelazq\Desktop\PfiSample\PfiSample.ConsoleApp\taxi-fare-train.csv";
        private static string MODEL_FILE = ConsumeModel.MLNetModelPath;

        // Create MLContext to be shared across the model creation workflow objects 
        // Set a random seed for repeatable/deterministic results across multiple trainings.
        private static MLContext mlContext = new MLContext(seed: 1);

        public static void CreateModel()
        {
            // Load Data
            IDataView trainingDataView = mlContext.Data.LoadFromTextFile<ModelInput>(
                                            path: TRAIN_DATA_FILEPATH,
                                            hasHeader: true,
                                            separatorChar: ',',
                                            allowQuoting: true,
                                            allowSparse: false);

            // Build training pipeline and Train Model
            IEstimator<ITransformer> trainingPipeline = BuildTrainingPipeline(mlContext);
            var mlModel = TrainModel(mlContext, trainingDataView, trainingPipeline);

            // Option 1: to extract predictor, requires to know the type in advance:
            // var predictor = ((TransformerChain<RegressionPredictionTransformer<LightGbmRegressionModelParameters>>)mlModel).LastTransformer;

            // Option 2: Should work always, as long as you _know_ the predictor is the last transformer in the chain.
            var predictor = ((IEnumerable<ITransformer>)mlModel).Last();

            // Option 3, need to load from disk the model first
            //var path = @"C:\Users\anvelazq\Desktop\PfiSample\model.zip";
            //mlContext.Model.Save(mlModel, trainingDataView.Schema, path);
            //var mlModel2 = mlContext.Model.Load(path, out var _);
            //var predictor = ((TransformerChain<ITransformer>) mlModel2).LastTransformer;

            // NOTE: Whatever option is chosen above, the explicit cast of the predictor seems to be needed here:
            CalculatePFI(mlContext, mlModel.Transform(trainingDataView), (ISingleFeaturePredictionTransformer<object>) predictor);
            
            // Evaluate quality of Model
            Evaluate(mlContext, trainingDataView, trainingPipeline);

            // Save model
            SaveModel(mlContext, mlModel, MODEL_FILE, trainingDataView.Schema);
        }

        public static IEstimator<ITransformer> BuildTrainingPipeline(MLContext mlContext)
        {
            // Data process configuration with pipeline data transformations 
            var dataProcessPipeline = mlContext.Transforms.Categorical.OneHotEncoding(new[] { new InputOutputColumnPair("vendor_id", "vendor_id"), new InputOutputColumnPair("payment_type", "payment_type") })
                                      .Append(mlContext.Transforms.Concatenate("Features", new[] { "vendor_id", "payment_type", "rate_code", "passenger_count", "trip_time_in_secs", "trip_distance" }));
            // Set the training algorithm 
            var trainer = mlContext.Regression.Trainers.LightGbm(labelColumnName: "fare_amount", featureColumnName: "Features");

            var trainingPipeline = dataProcessPipeline.Append(trainer);

            return trainingPipeline;
        }

        public static ITransformer TrainModel(MLContext mlContext, IDataView trainingDataView, IEstimator<ITransformer> trainingPipeline)
        {
            Console.WriteLine("=============== Training  model ===============");

            ITransformer model = trainingPipeline.Fit(trainingDataView);

            Console.WriteLine("=============== End of training process ===============");
            return model;
        }

        private static void Evaluate(MLContext mlContext, IDataView trainingDataView, IEstimator<ITransformer> trainingPipeline)
        {
            // Cross-Validate with single dataset (since we don't have two datasets, one for training and for evaluate)
            // in order to evaluate and get the model's accuracy metrics
            Console.WriteLine("=============== Cross-validating to get model's accuracy metrics ===============");
            var crossValidationResults = mlContext.Regression.CrossValidate(trainingDataView, trainingPipeline, numberOfFolds: 5, labelColumnName: "fare_amount");
            PrintRegressionFoldsAverageMetrics(crossValidationResults);
        }

        private static void CalculatePFI(MLContext mlContext, IDataView preprocessedTrainData, ISingleFeaturePredictionTransformer<object> model)
        {
            VBuffer<ReadOnlyMemory<char>> nameBuffer = default;
            preprocessedTrainData.Schema["Features"].Annotations.GetValue("SlotNames", ref nameBuffer); // NOTE: The column name "Features" needs to match the featureColumnName used in the trainer, the name "SlotNames" is always the same regardless of trainer.
            var featureColumnNames = nameBuffer.DenseValues().ToList();

            ImmutableArray<RegressionMetricsStatistics> permutationFeatureImportance =
                mlContext
                .Regression
                .PermutationFeatureImportance(model, preprocessedTrainData, permutationCount: 1, labelColumnName: "fare_amount");

            var featureImportanceMetrics =
                permutationFeatureImportance
                .Select((metric, index) => new { index, metric.RSquared })
                .OrderByDescending(myFeatures => Math.Abs(myFeatures.RSquared.Mean));

            Console.WriteLine("Feature\tPFI");

            foreach (var feature in featureImportanceMetrics)
            {
                Console.WriteLine($"{featureColumnNames[feature.index],-20}|\t{feature.RSquared.Mean:F6}");
            }
        }

        private static void SaveModel(MLContext mlContext, ITransformer mlModel, string modelRelativePath, DataViewSchema modelInputSchema)
        {
            // Save/persist the trained model to a .ZIP file
            Console.WriteLine($"=============== Saving the model  ===============");
            mlContext.Model.Save(mlModel, modelInputSchema, GetAbsolutePath(modelRelativePath));
            Console.WriteLine("The model is saved to {0}", GetAbsolutePath(modelRelativePath));
        }

        public static string GetAbsolutePath(string relativePath)
        {
            FileInfo _dataRoot = new FileInfo(typeof(Program).Assembly.Location);
            string assemblyFolderPath = _dataRoot.Directory.FullName;

            string fullPath = Path.Combine(assemblyFolderPath, relativePath);

            return fullPath;
        }

        public static void PrintRegressionMetrics(RegressionMetrics metrics)
        {
            Console.WriteLine($"*************************************************");
            Console.WriteLine($"*       Metrics for Regression model      ");
            Console.WriteLine($"*------------------------------------------------");
            Console.WriteLine($"*       LossFn:        {metrics.LossFunction:0.##}");
            Console.WriteLine($"*       R2 Score:      {metrics.RSquared:0.##}");
            Console.WriteLine($"*       Absolute loss: {metrics.MeanAbsoluteError:#.##}");
            Console.WriteLine($"*       Squared loss:  {metrics.MeanSquaredError:#.##}");
            Console.WriteLine($"*       RMS loss:      {metrics.RootMeanSquaredError:#.##}");
            Console.WriteLine($"*************************************************");
        }

        public static void PrintRegressionFoldsAverageMetrics(IEnumerable<TrainCatalogBase.CrossValidationResult<RegressionMetrics>> crossValidationResults)
        {
            var L1 = crossValidationResults.Select(r => r.Metrics.MeanAbsoluteError);
            var L2 = crossValidationResults.Select(r => r.Metrics.MeanSquaredError);
            var RMS = crossValidationResults.Select(r => r.Metrics.RootMeanSquaredError);
            var lossFunction = crossValidationResults.Select(r => r.Metrics.LossFunction);
            var R2 = crossValidationResults.Select(r => r.Metrics.RSquared);

            Console.WriteLine($"*************************************************************************************************************");
            Console.WriteLine($"*       Metrics for Regression model      ");
            Console.WriteLine($"*------------------------------------------------------------------------------------------------------------");
            Console.WriteLine($"*       Average L1 Loss:       {L1.Average():0.###} ");
            Console.WriteLine($"*       Average L2 Loss:       {L2.Average():0.###}  ");
            Console.WriteLine($"*       Average RMS:           {RMS.Average():0.###}  ");
            Console.WriteLine($"*       Average Loss Function: {lossFunction.Average():0.###}  ");
            Console.WriteLine($"*       Average R-squared:     {R2.Average():0.###}  ");
            Console.WriteLine($"*************************************************************************************************************");
        }
    }
}

[JakeRadMSFT]

@chunky did the example code help?

We're planning to make these easier in upcoming versions of ML.NET and Tooling

[JakeRadMSFT]

@chunky closing for now ... hopefully the example code helped. We'll work on making the API easier to use in the future.

[chunky]

The example code helped a lot, thanks!

I would recommend that something be created automatically by the wizard that demonstrates it for the specific model chosen by that wizard, but the sample got me sorted for now, thank-you