public BootstrapResult Compute()
{
if (Fitting == null)
throw new InvalidOperationException("Fitting function must have been previously defined.");
var models = new BootstrapValues[Subsamples.Length];
if (RunInParallel)
{
Parallel.For(0, Subsamples.Length, i =>
{
int[] trainingSet, validationSet;
// Create training and validation sets
CreatePartitions(i, out trainingSet, out validationSet);
// Fit and evaluate the model
models[i] = fitting(trainingSet, validationSet);
});
}
else
{
for (int i = 0; i < Subsamples.Length; i++)
{
int[] trainingSet, validationSet;
// Create training and validation sets
CreatePartitions(i, out trainingSet, out validationSet);
// Fit and evaluate the model
models[i] = fitting(trainingSet, validationSet);
}
}
// Return cross-validation statistics
return new BootstrapResult(this, models);
}
public void BootstrapConstructorTest2()
{
// Example from Masters, 1995
// Assume a small dataset
double[] data = { 3, 5, 2, 1, 7 };
// indices { 0, 1, 2, 3, 4 }
int[][] resamplings =
{
new [] { 4, 0, 2, 0, 3 }, // indices of { 7, 3, 2, 3, 1 }
new [] { 1, 3, 3, 0, 4 }, // indices of { 5, 1, 1, 3, 7 }
new [] { 2, 2, 4, 3, 0 }, // indices of { 2, 2, 7, 1, 3 }
};
Bootstrap target = new Bootstrap(data.Length, resamplings);
target.Fitting = (int[] trainingSamples, int[] validationSamples) =>
{
double[] subsample = data.Submatrix(trainingSamples);
double mean = subsample.Mean();
return new BootstrapValues(mean, 0);
};
var result = target.Compute();
double actualMean = result.Training.Mean;
double actualVar = result.Training.Variance;
Assert.AreEqual(3.2, actualMean, 1e-10);
Assert.AreEqual(0.04, actualVar, 1e-10);
}
