ml.net-7.cs

private static void ReportOnFeatureImportance(MLContext context, ITransformer model, IDataView data)
{
    // Need to cast from the ITransformer interface to gain access to the LastTransformer property.
    var typedModel = (TransformerChain<RegressionPredictionTransformer<FastForestRegressionModelParameters>>)model;

    // Calculate metrics.
    var permutationMetrics = context.Regression.PermutationFeatureImportance(typedModel.LastTransformer, model.Transform(data), PredictionLabel);

    // Combine metrics with feature names and format for display.
    var columnsToExclude = new[] { PredictionLabel, "Code", "Name", "IdPreservationColumn" };
    var featureNames = data.Schema.AsEnumerable()
        .Select(column => column.Name)
        .Where(name => !columnsToExclude.Contains(name))
        .ToArray();
    var results = featureNames
        .Select((t, i) => new FeatureImportance
            {
                Name = t,
                RSquaredMean = Math.Abs(permutationMetrics[i].RSquared.Mean),
                CorrelationCoefficient = 0 // TBC
        })
        .OrderByDescending(x => x.RSquaredMean);

    OutputFeatureImportanceResults(results);
}

private static void OutputFeatureImportanceResults(IEnumerable<FeatureImportance> results)
{
    Console.WriteLine("Feature importance:");

    var table = new ConsoleTable("Feature", "R Squared Mean", "Correlation Coefficient");
    foreach (var result in results)
    {
        table.AddRow(result.Name, result.RSquaredMean.ToString("G4"), result.CorrelationCoefficient.ToString("N2"));
    }

    table.Write();
    Console.WriteLine();
}