public MultipleLinearRegressor[] Regress(IAssetFxModel model)
{
if (_regressors != null)
{
return(_regressors);
}
var nPaths = _pathwiseValues.First().Length;
var finalSchedules = _portfolio.Select(x => x.ExpectedFlowsByPath(model)).ToArray();
var finalValues = new double[_dateIndexes.Length][];
ParallelUtils.Instance.For(0, _dateIndexes.Length, 1, d =>
{
var exposureDate = _regressionDates[d];
finalValues[d] = new double[nPaths];
foreach (var schedule in finalSchedules)
{
for (var p = 0; p < finalValues[d].Length; p++)
{
foreach (var flow in schedule[p].Flows)
{
if (flow.SettleDate > exposureDate)
{
finalValues[d][p] += flow.Pv;
}
}
}
}
}).Wait();
var o = new MultipleLinearRegressor[_dateIndexes.Length];
ParallelUtils.Instance.For(0, _dateIndexes.Length, 1, d =>
{
if (_regressionDates[d] <= model.BuildDate)
{
var detVec = new double[_pathwiseValues[d][0].Length + 1];
detVec[0] = finalValues[d].Average();
o[d] = new MultipleLinearRegressor(detVec);
}
else
{
var mlr = MultipleLinearRegression.Regress(_pathwiseValues[d], finalValues[d]);
o[d] = new MultipleLinearRegressor(mlr);
}
}).Wait();
_regressors = o;
return(o);
}
public void Dim2Facts()
{
double intercept = 76;
double w1 = 5;
double w2 = -2;
var testFunc = new Func <double[], double>(xs =>
{
return(intercept + xs[0] * w1 + xs[1] * w2);
});
var nExamples = 7;
var predictors = new double[nExamples][];
var predictions = new double[nExamples];
var R = new System.Random();
for (var e = 0; e < nExamples; e++)
{
predictors[e] = new double[2] {
R.NextDouble(), R.NextDouble()
};
predictions[e] = testFunc(predictors[e]);
}
var weights = MultipleLinearRegression.Regress(predictors, predictions);
Assert.Equal(intercept, weights[0], 8);
Assert.Equal(w1, weights[1], 8);
Assert.Equal(w2, weights[2], 8);
var weights2 = MultipleLinearRegression.RegressHistorical(predictors, predictions);
for (var i = 0; i < weights.Length; i++)
{
Assert.Equal(weights[i], weights2[i], 10);
}
var badPredictions = new double[nExamples - 1];
Assert.Throws <InvalidOperationException>(() => MultipleLinearRegression.RegressHistorical(predictors, badPredictions));
Assert.Throws <InvalidOperationException>(() => MultipleLinearRegression.Regress(predictors, badPredictions));
var z = new MultipleLinearRegressor(weights);
var q = z.Regress(new[] { 0.0, 0.0 });
}
