public Func <IList <Tin>, double[][]> CreatePreprocessorFunction(PreprocessorCoefficients coefficients)
{
var means = coefficients.Means;
var sigmas = coefficients.Sigmas;
return((sequence) =>
{
if (sequence.Count == 0)
{
return new double[0][];
}
double[][] result = new double[sequence.Count][];
for (int i = 0; i < sequence.Count; i++)
{
result[i] = Datapoint.AsDblArrayFrom <Tin>(sequence[i]);
}
double[][] zscores = Accord.Statistics.Tools.ZScores(result, means, sigmas);
return Accord.Math.Elementwise.Add(zscores, 10);
});
}
public double[][] Preprocess(double[][] extracted, PreprocessorCoefficients?coefficients = null)
{
// If not using the (more advanced) precalculated full-dataset averages & sigmas, calculate them locally here.
if (coefficients == null)
{
//var dims = Datapoint.From<Tsource>(default(Tsource)).Dimensions;
var dims = Dimensions;
double[] currentAxisValues = new double[extracted.Length]; // Just to save on allocations a tiny bit by reusing it within a single axis.
double[] means = new double[dims];
double[] sigmas = new double[dims];
for (int j = 0; j < dims; j++)
{
for (int i = 0; i < extracted.Length; i++)
{
currentAxisValues[i] = extracted[i][j];
}
means[j] = currentAxisValues.Average();
sigmas[j] = Accord.Statistics.Measures.StandardDeviation(currentAxisValues);
}
// Use our standard crosslinking function to lock values which should share a normalization constant into the same scale as each other.
if (doCrosslink)
{
sigmas = CrosslinkAxes.Apply <Tin>(sigmas); // Locks values in the same space - like X and Y axes - to the same window size.
}
coefficients = new PreprocessorCoefficients()
{
Means = means, Sigmas = sigmas
};
}
// Calculate the z-scores, then make them positive by the simple expedient of adding ten to them. (Yes, yes, but ten-sigma? Gotta mean an error anyway.)
double[][] zscores = Accord.Statistics.Tools.ZScores(extracted, coefficients.Value.Means, coefficients.Value.Sigmas);
return(Accord.Math.Elementwise.Add(zscores, 10));
}
