public static string CreateNNetworkAndLearn(List <NnRow> rows)
{
// Prepare data
double trainingSplitRatio = 0.7;
int trainCount = (int)(rows.Count * trainingSplitRatio);
var trainData = new MLNetData[trainCount];
var testData = new MLNetData[rows.Count - trainCount];
MLNetData[] allData = Convert(rows);
// Split into Training and Testing sets
Array.Copy(allData, 0, trainData, 0, trainCount);
Array.Copy(allData, trainCount, testData, 0, rows.Count - trainCount);
var allCollection = CollectionDataSource.Create(allData);
var trainCollection = CollectionDataSource.Create(trainData);
var testCollection = CollectionDataSource.Create(testData);
double acc, auc, f1;
PredictionModel <MLNetData, MLNetPredict> modelAll, modelTrain, modelBest;
//(acc, auc, f1, modelAll) = TrainAndGetMetrics(allCollection, allCollection, new AveragedPerceptronBinaryClassifier ()); // acc 0.83, auc 0.86, f1 0.45
//(acc, auc, f1, modelAll) = TrainAndGetMetrics(allCollection, allCollection, new FastForestBinaryClassifier ()); // acc 0.85, auc 0.89, f1 0.46
(acc, auc, f1, modelBest) = TrainAndGetMetrics(trainCollection, testCollection, new FastTreeBinaryClassifier()); // acc 0.95, auc 0.97, f1 0.85
//(acc, auc, f1, modelAll) = TrainAndGetMetrics(allCollection, allCollection, new FieldAwareFactorizationMachineBinaryClassifier()); // acc 0.85, auc 0.88, f1 0.56
//(acc, auc, f1, modelAll) = TrainAndGetMetrics(allCollection, allCollection, new GeneralizedAdditiveModelBinaryClassifier ()); // acc 0.81, auc 0.80, f1 NaN
//(acc, auc, f1, modelAll) = TrainAndGetMetrics(allCollection, allCollection, new LinearSvmBinaryClassifier ()); // acc 0.82, auc 0.86, f1 0.16
//(acc, auc, f1, modelAll) = TrainAndGetMetrics(allCollection, allCollection, new LogisticRegressionBinaryClassifier ()); // acc 0.84, auc 0.86, f1 0.40
//(acc, auc, f1, modelAll) = TrainAndGetMetrics(allCollection, allCollection, new StochasticDualCoordinateAscentBinaryClassifier()); // acc 0.84, auc 0.86, f1 0.40
//(acc, auc, f1, modelAll) = TrainAndGetMetrics(allCollection, allCollection, new StochasticGradientDescentBinaryClassifier ()); // acc 0.83, auc 0.86, f1 0.29
///(acc, auc, f1, modelAll) = TrainAndGetMetrics(allCollection, allCollection, new EnsembleBinaryClassifier ());
//(acc, auc, f1, modelTrain) = TrainAndGetMetrics(trainCollection, testCollection, new AveragedPerceptronBinaryClassifier ()); // acc 0.82, auc 0.84, f1 0.45
//(acc, auc, f1, modelTrain) = TrainAndGetMetrics(trainCollection, testCollection, new FastForestBinaryClassifier ()); // acc 0.82, auc 0.83, f1 0.23
//(acc, auc, f1, modelTrain) = TrainAndGetMetrics(trainCollection, testCollection, new FastTreeBinaryClassifier ()); // acc 0.82, auc 0.84, f1 0.46
//(acc, auc, f1, modelTrain) = TrainAndGetMetrics(trainCollection, testCollection, new FieldAwareFactorizationMachineBinaryClassifier()); // acc 0.83, auc 0.85, f1 0.37
//(acc, auc, f1, modelTrain) = TrainAndGetMetrics(trainCollection, testCollection, new GeneralizedAdditiveModelBinaryClassifier ()); // acc 0.81, auc 0.75, f1 NaN
//(acc, auc, f1, modelTrain) = TrainAndGetMetrics(trainCollection, testCollection, new LinearSvmBinaryClassifier ()); // acc 0.81, auc 0.83, f1 0.14
//(acc, auc, f1, modelTrain) = TrainAndGetMetrics(trainCollection, testCollection, new LogisticRegressionBinaryClassifier ()); // acc 0.83, auc 0.84, f1 0.39
//(acc, auc, f1, modelTrain) = TrainAndGetMetrics(trainCollection, testCollection, new StochasticDualCoordinateAscentBinaryClassifier()); // acc 0.82, auc 0.84, f1 0.43
//(acc, auc, f1, modelTrain) = TrainAndGetMetrics(trainCollection, testCollection, new StochasticGradientDescentBinaryClassifier ()); // acc 0.83, auc 0.83, f1 0.34
// Evaluate a training model
//Console.WriteLine($"Accuracy: {metrics.Accuracy:P2}");
//Console.WriteLine($"Auc: {metrics.Auc:P2}");
//Console.WriteLine($"F1Score: {metrics.F1Score:P2}");
//var cv = new CrossValidator();
//CrossValidationOutput<MLNetData, MLNetPredict> cvRes = cv.CrossValidate<MLNetData, MLNetPredict>(pipelineAll);
//Console.WriteLine($"Rms = {metrics.Rms}");
//Console.WriteLine($"RSquared = {metrics.RSquared}");
// Train the overall model
string NnModelPath = @"NnInputs\mlDotNet_Datacup.model";
modelBest.WriteAsync(NnModelPath);
return(NnModelPath);
}
public static MLNetData[] Convert(List <NnRow> rows)
{
var allData = new MLNetData[rows.Count];
for (int i = 0; i < rows.Count; i++)
{
allData[i] = new MLNetData();
allData[i].Features = new float[Constants.len];
var arr = Array.ConvertAll(rows[i].nbs, x => (float)x);
Array.Copy(arr, 0, allData[i].Features, 0, arr.Length);
allData[i].Label = rows[i].verdict /* == 0? false: true*/;
}
return(allData);
}
