void Discriminant_getPartialDiscriminationProbability (Discriminant me, long numberOfDimensions, double *p_prob, double *p_chisq, double *p_df)
{
long g = my numberOfGroups;
long p = my eigen -> dimension, k = numberOfDimensions;
long numberOfFunctions = Discriminant_getNumberOfFunctions (me);
double degreesOfFreedom = Discriminant_getDegreesOfFreedom (me);
double prob = NUMundefined, chisq = NUMundefined, df = NUMundefined;
if (k < numberOfFunctions) {
double lambda = NUMwilksLambda (my eigen -> eigenvalues, k + 1, numberOfFunctions);
if (lambda != 1.0) {
chisq = - (degreesOfFreedom + (g - p) / 2.0 - 1.0) * log (lambda);
df = (p - k) * (g - k - 1);
if (p_prob) {
prob = NUMchiSquareQ (chisq, df);
}
}
}
if (p_prob) {
*p_prob = prob;
}
if (p_chisq) {
*p_chisq = chisq;
}
if (p_df) {
*p_df = df;
}
}
void CCA_getZeroCorrelationProbability (CCA me, long index, double *p_prob, double *p_chisq, double *p_df) {
double lambda = 1.0, *ev = my y -> eigenvalues;
long nev = my y -> numberOfEigenvalues;
long ny = my y -> dimension, nx = my x -> dimension;
double chisq = NUMundefined, prob = NUMundefined, df = NUMundefined;
if (index >= 1 && index <= nev) {
for (long i = index; i <= nev; i ++) {
lambda *= 1.0 - ev [i];
}
df = (ny - index + 1) * (nx - index + 1);
chisq = - (my numberOfObservations - (ny + nx + 3.0) / 2.0) * log (lambda);
prob = NUMchiSquareQ (chisq, df);
}
if (p_chisq) {
*p_chisq = chisq;
}
if (p_df) {
*p_df = df;
}
if (p_prob) {
*p_prob = prob;
}
}
double ContingencyTable_chisqProbability (ContingencyTable me) {
double chisq, df;
ContingencyTable_chisq (me, &chisq, &df);
if (chisq == 0.0 && df == 0.0) {
return 0.0;
}
return NUMchiSquareQ (chisq, df);
}
void CCA_getZeroCorrelationProbability (CCA me, long index, double *chisq, long *ndf, double *probability) {
double lambda = 1.0, *ev = my y -> eigenvalues;
long nev = my y -> numberOfEigenvalues;
long ny = my y -> dimension, nx = my x -> dimension;
*chisq = *probability = NUMundefined;
*ndf = 0;
if (index < 1 || index > nev) {
return;
}
for (long i = index; i <= nev; i++) {
lambda *= (1.0 - ev[i]);
}
*ndf = (ny - index + 1) * (nx - index + 1);
*chisq = - (my numberOfObservations - (ny + nx + 3.0) / 2.0) * log (lambda);
*probability = NUMchiSquareQ (*chisq, *ndf);
}
void PCA_getEqualityOfEigenvalues (PCA me, long from, long to, int conservative, double *p_prob, double *p_chisq, double *p_df) {
double sum = 0, sumln = 0;
double prob = NUMundefined, df = NUMundefined, chisq = NUMundefined;
if (from == 0 && to == 0) {
to = 1; from = my numberOfEigenvalues;
}
if (from < to && from > 0 && to <= my numberOfEigenvalues) {
long i;
for (i = from; i <= to; i++) {
if (my eigenvalues[i] <= 0) {
break;
}
sum += my eigenvalues[i];
sumln += log (my eigenvalues[i]);
}
if (sum == 0) {
return;
}
long r = i - from;
long n = my numberOfObservations - 1;
if (conservative) {
n -= from + (r * (2 * r + 1) + 2) / (6 * r);
}
df = r * (r + 1) / 2 - 1;
chisq = n * (r * log (sum / r) - sumln);
prob = NUMchiSquareQ (chisq, df);
}
if (p_prob) {
*p_prob = prob;
}
if (p_chisq) {
*p_chisq = chisq;
}
if (p_df) {
*p_df = df;
}
}
void Discriminant_getPartialDiscriminationProbability (Discriminant me,
long numberOfDimensions, double *probability, double *chisq, long *ndf) {
long g = my numberOfGroups;
long p = my dimension, k = numberOfDimensions;
long numberOfFunctions = Discriminant_getNumberOfFunctions (me);
double degreesOfFreedom = Discriminant_getDegreesOfFreedom (me);
double lambda;
*probability = 1; *chisq = 0; *ndf = 0;
if (k >= numberOfFunctions) {
return;
}
lambda = NUMwilksLambda (my eigenvalues, k + 1, numberOfFunctions);
if (lambda == 1) {
return;
}
*chisq = - (degreesOfFreedom + (g - p) / 2 - 1) * log (lambda);
*ndf = (p - k) * (g - k - 1);
*probability = NUMchiSquareQ (*chisq, *ndf);
}
void PCA_getEqualityOfEigenvalues (PCA me, long from, long to, int conservative, double *probability, double *chisq, long *ndf) {
double sum = 0, sumln = 0;
*probability = 1;
*ndf = 0;
*chisq = 0;
if ((from > 0 && to == from) || to > my numberOfEigenvalues) {
return;
}
if (to <= from) {
from = 1;
to = my numberOfEigenvalues;
}
long i;
for (i = from; i <= to; i++) {
if (my eigenvalues[i] <= 0) {
break;
}
sum += my eigenvalues[i];
sumln += log (my eigenvalues[i]);
}
if (sum == 0) {
return;
}
long r = i - from;
long n = my numberOfObservations - 1;
if (conservative) {
n -= from + (r * (2 * r + 1) + 2) / (6 * r);
}
*ndf = r * (r + 1) / 2 - 1;
*chisq = n * (r * log (sum / r) - sumln);
*probability = NUMchiSquareQ (*chisq, *ndf);
}
