org.biojava.nbio.survival.cox

Class CoxR

java.lang.Object

org.biojava.nbio.survival.cox.CoxR

public class CoxR
extends Object

This is a port of the R survival code used for doing Cox Regression. The algorithm was a fairly easy port from C code to Java where the challenge was making the code a little more object friendly. In the R code everything is passed around as an array and a large portion of the code is spent extracting data from the array for use in different calculations. By organizing the data in a class for each data point was able to simplify much of the code. Not all variants of different methods that you can select for doing various statistical calculations are implemented. Wouldn't be difficult to go back in add them in if they are important.

In R you can pass in different paramaters to override defaults which requires parsing of the paramaters. In the Java code tried to be a little more exact in the code related to paramaters where using strata, weighting, robust and cluster are advance options. Additionaly code is implemented from Bob Gray to do variance correction when using weighted paramaters in a data set. /Users/Scooter/NetBeansProjects/biojava3-survival/docs/wtexamples.docx

The CoxHelper class is meant to hide some of the implementation details.

Issues

• sign in CoxMart?
• double toler_chol = 1.818989e-12; Different value for some reason
• In robust linear_predictor set to 0 which implies score = 1 but previous score value doesn't get reset

Cox regression fit, replacement for coxfit2 in order to be more frugal about memory: specificly that we don't make copies of the input data.

the input parameters are

      maxiter      :number of iterations
      time(n)      :time of status or censoring for person i
      status(n)    :status for the ith person    1=dead , 0=censored
      covar(nv,n)  :covariates for person i.
                       Note that S sends this in column major order.
      strata(n)    :marks the strata.  Will be 1 if this person is the
                      last one in a strata.  If there are no strata, the
                      vector can be identically zero, since the nth person's
                      value is always assumed to be = to 1.
      offset(n)    :offset for the linear predictor
      weights(n)   :case weights
      init         :initial estimate for the coefficients
      eps          :tolerance for convergence.  Iteration continues until
                      the percent change in loglikelihood is <= eps.
      chol_tol     : tolerance for the Cholesky decompostion
      method       : 0=Breslow, 1=Efron
      doscale      : 0=don't scale the X matrix, 1=scale the X matrix
 

returned parameters

      means(nv)    : vector of column means of X
      beta(nv)     :the vector of answers (at start contains initial est)
      u(nv)        :score vector
      imat(nv,nv)  :the variance matrix at beta=final
                     (returned as a vector)
      loglik(2)    :loglik at beta=initial values, at beta=final
      sctest       :the score test at beta=initial
      flag         :success flag  1000  did not converge
                                  1 to nvar: rank of the solution
      iterations         :actual number of iterations used
 

work arrays

      mark(n)
      wtave(n)
      a(nvar), a2(nvar)
      cmat(nvar,nvar)       ragged array
      cmat2(nvar,nvar)
      newbeta(nvar)         always contains the "next iteration"
 

calls functions: cholesky2, chsolve2, chinv2

the data must be sorted by ascending time within strata

Author:

Scooter Willis

Constructor Summary

Constructors

Constructor and Description
CoxR()

Method Summary

Modifier and Type	Method and Description
static void	calculateWaldTestInfo(CoxInfo ci)
void	coxphfitSCleanup(CoxInfo ci, boolean useWeighted, boolean robust, ArrayList<String> cluster)
double	coxsafe(double x)
static void	main(String[] args)
CoxInfo	process(ArrayList<String> variables, ArrayList<SurvivalInfo> DataT, boolean useStrata, boolean useWeighted, boolean robust, boolean cluster)
CoxInfo	process(ArrayList<String> variables, ArrayList<SurvivalInfo> data, int maxiter, CoxMethod method, double eps, double toler, double[] beta, int doscale, boolean useStrata, boolean useWeighted, boolean robust, boolean cluster)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

CoxR

public CoxR()

Method Detail

process

public CoxInfo process(ArrayList<String> variables,
                       ArrayList<SurvivalInfo> DataT,
                       boolean useStrata,
                       boolean useWeighted,
                       boolean robust,
                       boolean cluster)
                throws Exception

Parameters:

variables -

DataT -

useStrata -

useWeighted -

robust -

cluster -

Returns:

Throws:

Exception

process

public CoxInfo process(ArrayList<String> variables,
                       ArrayList<SurvivalInfo> data,
                       int maxiter,
                       CoxMethod method,
                       double eps,
                       double toler,
                       double[] beta,
                       int doscale,
                       boolean useStrata,
                       boolean useWeighted,
                       boolean robust,
                       boolean cluster)
                throws Exception

Parameters:

variables -

data -

maxiter -

method -

eps -

toler -

beta -

doscale -

useStrata -

useWeighted -

robust -

cluster -

Returns:

Throws:

Exception

coxphfitSCleanup

public void coxphfitSCleanup(CoxInfo ci,
                             boolean useWeighted,
                             boolean robust,
                             ArrayList<String> cluster)
                      throws Exception

Parameters:

ci -

useWeighted -

robust -

cluster -

Throws:

Exception

calculateWaldTestInfo

public static void calculateWaldTestInfo(CoxInfo ci)

main

public static void main(String[] args)

Parameters:

args - the command line arguments

coxsafe

public double coxsafe(double x)

Parameters:

x -

Returns: