/**
* @param ci
* @param useWeighted
* @param robust
* @param cluster
* @throws Exception
*/
public void coxphfitSCleanup(
CoxInfo ci, boolean useWeighted, boolean robust, ArrayList<String> cluster) throws Exception {
// Do cleanup found after coxfit6 is called in coxph.fit.S
// infs <- abs(coxfit$u %*% var)
// [ a1 b1] * [a1 b1]
// [a2 b2]
double[][] du = new double[1][ci.u.length];
du[0] = ci.u;
double[] infs = Matrix.abs(Matrix.multiply(ci.u, ci.getVariance()));
// StdArrayIO.print(infs);
ArrayList<CoxCoefficient> coxCoefficients =
new ArrayList<CoxCoefficient>(ci.getCoefficientsList().values());
for (int i = 0; i < infs.length; i++) {
double inf = infs[i];
double coe = coxCoefficients.get(i).getCoeff();
if (inf > ci.eps && inf > (ci.toler * Math.abs(coe))) {
ci.message = "Loglik converged before variable ";
}
}
// sum(coef*coxfit$means)
double sumcoefmeans = 0;
for (CoxCoefficient cc : coxCoefficients) {
sumcoefmeans = sumcoefmeans + cc.getCoeff() * cc.getMean();
}
// coxph.fit.S line 107
// lp <- c(x %*% coef) + offset - sum(coef*coxfit$means)
for (SurvivalInfo si : ci.survivalInfoList) {
double offset = si.getOffset();
double lp = 0;
for (CoxCoefficient cc : coxCoefficients) {
String name = cc.getName();
double coef = cc.getCoeff();
double value = si.getVariable(name);
lp = lp + value * coef;
}
lp = lp + offset - sumcoefmeans;
si.setLinearPredictor(lp);
si.setScore(Math.exp(lp));
// System.out.println("lp score " + si.order + " " + si.time + " " + si.getWeight()
// + " " + si.getClusterValue() + " " + lp + " " + Math.exp(lp));
}
// ci.dump();
// begin code after call to coxfit6 in coxph.fit.S
// Compute the martingale residual for a Cox model
// appears to be C syntax error for = - vs -=
// (if (nullmodel) in coxph.fit
double[] res = CoxMart.process(ci.method, ci.survivalInfoList, false);
for (int i = 0; i < ci.survivalInfoList.size(); i++) {
SurvivalInfo si = ci.survivalInfoList.get(i);
si.setResidual(res[i]);
}
// this represents the end of coxph.fit.S code and we pickup
// after call to fit <- fitter(X, Y, strats ....) in coxph.R
if (robust) {
ci.setNaiveVariance(ci.getVariance());
double[][] temp;
double[][] temp0;
if (cluster != null) {
temp = ResidualsCoxph.process(ci, ResidualsCoxph.Type.dfbeta, useWeighted, cluster);
// # get score for null model
// if (is.null(init))
// fit2$linear.predictors <- 0*fit$linear.predictors
// else
// fit2$linear.predictors <- c(X %*% init)
// Set score to 1
double[] templp = new double[ci.survivalInfoList.size()];
double[] tempscore = new double[ci.survivalInfoList.size()];
int i = 0;
for (SurvivalInfo si : ci.survivalInfoList) {
templp[i] = si.getLinearPredictor();
tempscore[i] = si.getScore();
si.setLinearPredictor(0);
si.setScore(1.0); // this erases stored value which isn't how the R code does it
i++;
}
temp0 = ResidualsCoxph.process(ci, ResidualsCoxph.Type.score, useWeighted, cluster);
i = 0;
for (SurvivalInfo si : ci.survivalInfoList) {
si.setLinearPredictor(templp[i]);
si.setScore(tempscore[i]); // this erases stored value which isn't how the R code does it
i++;
}
} else {
temp = ResidualsCoxph.process(ci, ResidualsCoxph.Type.dfbeta, useWeighted, null);
// fit2$linear.predictors <- 0*fit$linear.predictors
double[] templp = new double[ci.survivalInfoList.size()];
double[] tempscore = new double[ci.survivalInfoList.size()];
int i = 0;
for (SurvivalInfo si : ci.survivalInfoList) {
templp[i] = si.getLinearPredictor();
tempscore[i] = si.getScore();
si.setLinearPredictor(0);
si.setScore(1.0);
}
temp0 = ResidualsCoxph.process(ci, ResidualsCoxph.Type.score, useWeighted, null);
i = 0;
for (SurvivalInfo si : ci.survivalInfoList) {
si.setLinearPredictor(templp[i]);
si.setScore(tempscore[i]); // this erases stored value which isn't how the R code does it
i++;
}
}
// fit$var<- t(temp) % * % temp
double[][] ttemp = Matrix.transpose(temp);
double[][] var = Matrix.multiply(ttemp, temp);
ci.setVariance(var);
// u<- apply(as.matrix(temp0), 2, sum)
double[] u = new double[temp0[0].length];
for (int i = 0; i < temp0[0].length; i++) {
for (int j = 0; j < temp0.length; j++) {
u[i] = u[i] + temp0[j][i];
}
}
// fit$rscore <- coxph.wtest(t(temp0)%*%temp0, u, control$toler.chol)$test
double[][] wtemp = Matrix.multiply(Matrix.transpose(temp0), temp0);
double toler_chol = 1.818989e-12;
// toler_chol = ci.toler;
WaldTestInfo wti = WaldTest.process(wtemp, u, toler_chol);
// not giving the correct value
ci.setRscore(wti.getTest());
}
if (ci.getNumberCoefficients() > 0) {
double toler_chol = 1.818989e-12;
// toler_chol = ci.toler;
double[][] b = new double[1][ci.getNumberCoefficients()];
int i = 0;
for (CoxCoefficient coe : ci.getCoefficientsList().values()) {
b[0][i] = coe.getCoeff();
i++;
}
ci.setWaldTestInfo(WaldTest.process(ci.getVariance(), b, toler_chol));
}
}
