public double logPdf(double[] x) {
Matrix W = new Matrix(x, dim, dim);
double logDensity = 0;
// System.err.println("here");
// double det = 0;
// try {
// det = W.determinant();
// } catch (IllegalDimension illegalDimension) {
// illegalDimension.printStackTrace();
// }
// if( det < 0 ) {
// System.err.println("not positive definite");
// return Double.NEGATIVE_INFINITY;
// }
try {
logDensity = Math.log(W.determinant());
logDensity *= -0.5;
logDensity *= df + dim + 1;
Matrix product = S.product(W.inverse());
for (int i = 0; i < dim; i++) logDensity -= 0.5 * product.component(i, i);
} catch (IllegalDimension illegalDimension) {
illegalDimension.printStackTrace();
}
logDensity += logNormalizationConstant;
return logDensity;
}
private void computeNormalizationConstant() {
logNormalizationConstant = 0;
try {
logNormalizationConstant = df / 2.0 * Math.log(new Matrix(scaleMatrix).determinant());
} catch (IllegalDimension illegalDimension) {
illegalDimension.printStackTrace();
}
logNormalizationConstant -= df * dim / 2.0 * Math.log(2);
logNormalizationConstant -= dim * (dim - 1) / 4.0 * Math.log(Math.PI);
for (int i = 1; i <= dim; i++) {
logNormalizationConstant -= GammaFunction.lnGamma((df + 1 - i) / 2.0);
}
}
