|
| 1 | +using System; |
| 2 | +using System.Collections.Generic; |
| 3 | +using System.Linq; |
| 4 | +using Microsoft.ML; |
| 5 | +using Microsoft.ML.Data; |
| 6 | +using Microsoft.ML.Trainers; |
| 7 | +using Microsoft.ML.Trainers.FastTree; |
| 8 | +using Newtonsoft.Json; |
| 9 | + |
| 10 | +namespace Microsoft.ML.Samples.OneDal |
| 11 | +{ |
| 12 | + class Program |
| 13 | + { |
| 14 | + public static IDataView[] LoadData( |
| 15 | + MLContext mlContext, string trainingFile, string testingFile, |
| 16 | + string task, string label = "target", char separator = ',') |
| 17 | + { |
| 18 | + List<IDataView> dataList = new List<IDataView>(); |
| 19 | + System.IO.StreamReader file = new System.IO.StreamReader(trainingFile); |
| 20 | + string header = file.ReadLine(); |
| 21 | + file.Close(); |
| 22 | + string[] headerArray = header.Split(separator); |
| 23 | + List<TextLoader.Column> columns = new List<TextLoader.Column>(); |
| 24 | + foreach (string column in headerArray) |
| 25 | + { |
| 26 | + if (column == label) |
| 27 | + { |
| 28 | + if (task == "binary") |
| 29 | + columns.Add(new TextLoader.Column(column, DataKind.Boolean, Array.IndexOf(headerArray, column))); |
| 30 | + else |
| 31 | + columns.Add(new TextLoader.Column(column, DataKind.Single, Array.IndexOf(headerArray, column))); |
| 32 | + } |
| 33 | + else |
| 34 | + { |
| 35 | + columns.Add(new TextLoader.Column(column, DataKind.Single, Array.IndexOf(headerArray, column))); |
| 36 | + } |
| 37 | + } |
| 38 | + |
| 39 | + var loader = mlContext.Data.CreateTextLoader( |
| 40 | + separatorChar: separator, |
| 41 | + hasHeader: true, |
| 42 | + columns: columns.ToArray() |
| 43 | + ); |
| 44 | + dataList.Add(loader.Load(trainingFile)); |
| 45 | + dataList.Add(loader.Load(testingFile)); |
| 46 | + return dataList.ToArray(); |
| 47 | + } |
| 48 | + |
| 49 | + public static string[] GetFeaturesArray(IDataView data, string labelName = "target") |
| 50 | + { |
| 51 | + List<string> featuresList = new List<string>(); |
| 52 | + var nColumns = data.Schema.Count; |
| 53 | + var columnsEnumerator = data.Schema.GetEnumerator(); |
| 54 | + for (int i = 0; i < nColumns; i++) |
| 55 | + { |
| 56 | + columnsEnumerator.MoveNext(); |
| 57 | + if (columnsEnumerator.Current.Name != labelName) |
| 58 | + featuresList.Add(columnsEnumerator.Current.Name); |
| 59 | + } |
| 60 | + |
| 61 | + return featuresList.ToArray(); |
| 62 | + } |
| 63 | + |
| 64 | + public static double[] RunRandomForestClassification(MLContext mlContext, IDataView trainingData, IDataView testingData, string labelName, int numberOfTrees, int numberOfLeaves) |
| 65 | + { |
| 66 | + var featuresArray = GetFeaturesArray(trainingData, labelName); |
| 67 | + var preprocessingPipeline = mlContext.Transforms.Concatenate("Features", featuresArray); |
| 68 | + var preprocessedTrainingData = preprocessingPipeline.Fit(trainingData).Transform(trainingData); |
| 69 | + var preprocessedTestingData = preprocessingPipeline.Fit(trainingData).Transform(testingData); |
| 70 | + |
| 71 | + FastForestBinaryTrainer.Options options = new FastForestBinaryTrainer.Options(); |
| 72 | + options.LabelColumnName = labelName; |
| 73 | + options.FeatureColumnName = "Features"; |
| 74 | + options.NumberOfTrees = numberOfTrees; |
| 75 | + options.NumberOfLeaves = numberOfLeaves; |
| 76 | + options.MinimumExampleCountPerLeaf = 5; |
| 77 | + options.FeatureFraction = 1.0; |
| 78 | + |
| 79 | + var trainer = mlContext.BinaryClassification.Trainers.FastForest(options); |
| 80 | + |
| 81 | + ITransformer model = trainer.Fit(preprocessedTrainingData); |
| 82 | + |
| 83 | + IDataView trainingPredictions = model.Transform(preprocessedTrainingData); |
| 84 | + var trainingMetrics = mlContext.BinaryClassification.EvaluateNonCalibrated(trainingPredictions, labelColumnName: labelName); |
| 85 | + IDataView testingPredictions = model.Transform(preprocessedTestingData); |
| 86 | + var testingMetrics = mlContext.BinaryClassification.EvaluateNonCalibrated(testingPredictions, labelColumnName: labelName); |
| 87 | + |
| 88 | + double[] metrics = new double[4]; |
| 89 | + metrics[0] = trainingMetrics.Accuracy; |
| 90 | + metrics[1] = testingMetrics.Accuracy; |
| 91 | + metrics[2] = trainingMetrics.F1Score; |
| 92 | + metrics[3] = testingMetrics.F1Score; |
| 93 | + return metrics; |
| 94 | + } |
| 95 | + |
| 96 | + public static double[] RunRandomForestRegression(MLContext mlContext, IDataView trainingData, IDataView testingData, string labelName, int numberOfTrees, int numberOfLeaves) |
| 97 | + { |
| 98 | + var featuresArray = GetFeaturesArray(trainingData, labelName); |
| 99 | + var preprocessingPipeline = mlContext.Transforms.Concatenate("Features", featuresArray); |
| 100 | + var preprocessedTrainingData = preprocessingPipeline.Fit(trainingData).Transform(trainingData); |
| 101 | + var preprocessedTestingData = preprocessingPipeline.Fit(trainingData).Transform(testingData); |
| 102 | + |
| 103 | + FastForestRegressionTrainer.Options options = new FastForestRegressionTrainer.Options(); |
| 104 | + options.LabelColumnName = labelName; |
| 105 | + options.FeatureColumnName = "Features"; |
| 106 | + options.NumberOfTrees = numberOfTrees; |
| 107 | + options.NumberOfLeaves = numberOfLeaves; |
| 108 | + options.MinimumExampleCountPerLeaf = 5; |
| 109 | + options.FeatureFraction = 1.0; |
| 110 | + |
| 111 | + var trainer = mlContext.Regression.Trainers.FastForest(options); |
| 112 | + |
| 113 | + ITransformer model = trainer.Fit(preprocessedTrainingData); |
| 114 | + |
| 115 | + IDataView trainingPredictions = model.Transform(preprocessedTrainingData); |
| 116 | + var trainingMetrics = mlContext.Regression.Evaluate(trainingPredictions, labelColumnName: labelName); |
| 117 | + IDataView testingPredictions = model.Transform(preprocessedTestingData); |
| 118 | + var testingMetrics = mlContext.Regression.Evaluate(testingPredictions, labelColumnName: labelName); |
| 119 | + |
| 120 | + double[] metrics = new double[4]; |
| 121 | + metrics[0] = trainingMetrics.RootMeanSquaredError; |
| 122 | + metrics[1] = testingMetrics.RootMeanSquaredError; |
| 123 | + metrics[2] = trainingMetrics.RSquared; |
| 124 | + metrics[3] = testingMetrics.RSquared; |
| 125 | + return metrics; |
| 126 | + } |
| 127 | + |
| 128 | + public static double[] RunOLSRegression(MLContext mlContext, IDataView trainingData, IDataView testingData, string labelName) |
| 129 | + { |
| 130 | + var featuresArray = GetFeaturesArray(trainingData, labelName); |
| 131 | + var preprocessingPipeline = mlContext.Transforms.Concatenate("Features", featuresArray); |
| 132 | + var preprocessedTrainingData = preprocessingPipeline.Fit(trainingData).Transform(trainingData); |
| 133 | + var preprocessedTestingData = preprocessingPipeline.Fit(trainingData).Transform(testingData); |
| 134 | + |
| 135 | + OlsTrainer.Options options = new OlsTrainer.Options(); |
| 136 | + options.LabelColumnName = labelName; |
| 137 | + options.FeatureColumnName = "Features"; |
| 138 | + |
| 139 | + var trainer = mlContext.Regression.Trainers.Ols(options); |
| 140 | + |
| 141 | + ITransformer model = trainer.Fit(preprocessedTrainingData); |
| 142 | + |
| 143 | + IDataView trainingPredictions = model.Transform(preprocessedTrainingData); |
| 144 | + var trainingMetrics = mlContext.Regression.Evaluate(trainingPredictions, labelColumnName: labelName); |
| 145 | + IDataView testingPredictions = model.Transform(preprocessedTestingData); |
| 146 | + var testingMetrics = mlContext.Regression.Evaluate(testingPredictions, labelColumnName: labelName); |
| 147 | + |
| 148 | + double[] metrics = new double[4]; |
| 149 | + metrics[0] = trainingMetrics.RootMeanSquaredError; |
| 150 | + metrics[1] = testingMetrics.RootMeanSquaredError; |
| 151 | + metrics[2] = trainingMetrics.RSquared; |
| 152 | + metrics[3] = testingMetrics.RSquared; |
| 153 | + return metrics; |
| 154 | + } |
| 155 | + |
| 156 | + static void Main(string[] args) |
| 157 | + { |
| 158 | + // args[0] - training data filename |
| 159 | + // args[1] - testing data filename |
| 160 | + // args[2] - machine learning task (regression, binary) |
| 161 | + // args[3] - machine learning algorithm (RandomForest, OLS) |
| 162 | + // Random Forest parameters: |
| 163 | + // args[4] - NumberOfTrees |
| 164 | + // args[5] - NumberOfLeaves |
| 165 | + var mlContext = new MLContext(seed: 42); |
| 166 | + // data[0] - training subset |
| 167 | + // data[1] - testing subset |
| 168 | + IDataView[] data = LoadData(mlContext, args[0], args[1], args[2]); |
| 169 | + string labelName = "target"; |
| 170 | + |
| 171 | + var mainWatch = System.Diagnostics.Stopwatch.StartNew(); |
| 172 | + double[] metrics; |
| 173 | + if (args[3] == "RandomForest") |
| 174 | + { |
| 175 | + int numberOfTrees = Int32.Parse(args[4]); |
| 176 | + int numberOfLeaves = Int32.Parse(args[5]); |
| 177 | + if (args[2] == "binary") |
| 178 | + { |
| 179 | + |
| 180 | + metrics = RunRandomForestClassification(mlContext, data[0], data[1], labelName, numberOfTrees, numberOfLeaves); |
| 181 | + mainWatch.Stop(); |
| 182 | + Console.WriteLine("algorithm,all workflow time[ms],training accuracy,testing accuracy,training F1 score,testing F1 score"); |
| 183 | + Console.WriteLine($"Random Forest Binary,{mainWatch.Elapsed.TotalMilliseconds},{metrics[0]},{metrics[1]},{metrics[2]},{metrics[3]}"); |
| 184 | + } |
| 185 | + else |
| 186 | + { |
| 187 | + metrics = RunRandomForestRegression(mlContext, data[0], data[1], labelName, numberOfTrees, numberOfLeaves); |
| 188 | + mainWatch.Stop(); |
| 189 | + Console.WriteLine("algorithm,all workflow time[ms],training RMSE,testing RMSE,training R2 score,testing R2 score"); |
| 190 | + Console.WriteLine($"Random Forest Regression,{mainWatch.Elapsed.TotalMilliseconds},{metrics[0]},{metrics[1]},{metrics[2]},{metrics[3]}"); |
| 191 | + } |
| 192 | + } |
| 193 | + else if (args[3] == "OLS") |
| 194 | + { |
| 195 | + metrics = RunOLSRegression(mlContext, data[0], data[1], labelName); |
| 196 | + mainWatch.Stop(); |
| 197 | + Console.WriteLine("algorithm,all workflow time[ms],training RMSE,testing RMSE,training R2 score,testing R2 score"); |
| 198 | + Console.WriteLine($"OLS Regression,{mainWatch.Elapsed.TotalMilliseconds},{metrics[0]},{metrics[1]},{metrics[2]},{metrics[3]}"); |
| 199 | + } |
| 200 | + } |
| 201 | + } |
| 202 | +} |
| 203 | + |
0 commit comments