R/gdm_est_skillspace_traits.R

In alexanderrobitzsch/CDM: Cognitive Diagnosis Modeling

## File Name: gdm_est_skillspace_traits.R
## File Version: 0.13

#####################################################
# estimation of skill space
gdm_est_skillspace_traits <- function( n.ik, a, b, theta.k, Qmatrix, I, K, TP,
        TD, numdiff.parm, max.increment, msteps, convM )
{
    n.ik0 <- apply( n.ik, c(1,2,3), sum )
    h <- numdiff.parm
    parchange <- 1000
    iter <- 1
    se.theta.k <- 0 * theta.k
    Q1 <- matrix( 0, nrow=TP, ncol=TD)

    #-- define likelihood function and list of arguments
    prob_fct <- gdm_calc_prob
    prob_args <- list( a=a, b=b, thetaDes=theta.k, Qmatrix=Qmatrix, I=I, K=K, TP=TP, TD=TD )
    parm_args_varname <- "thetaDes"

    #--------- begin M-steps
    while( ( iter <=msteps ) & (parchange > convM ) ){
        theta.k0 <- theta.k
        for ( dd in 1:TD){
            Q0 <- Q1
            Q0[,dd] <- 1
            # calculate log-likelihood
            prob_args[[ parm_args_varname ]] <- theta.k0
            pjk <- do.call( what=prob_fct, args=prob_args)

            prob_args[[ parm_args_varname ]] <- theta.k0 + h*Q0
            pjk1 <- do.call( what=prob_fct, args=prob_args)

            prob_args[[ parm_args_varname ]] <- theta.k0 - h*Q0
            pjk2 <- do.call( what=prob_fct, args=prob_args)

            #-- compute increments
            res <- gdm_numdiff_index( pjk=pjk, pjk1=pjk1, pjk2=pjk2, n.ik=n.ik0, max.increment=max.increment,
                        numdiff.parm=numdiff.parm )
            increment <- res$increment
            d2 <- res$d2
            theta.k[,dd] <- theta.k[,dd] + increment
            se.theta.k[,dd] <- 1 / sqrt( abs(d2) )
        }
        iter <- iter + 1
        parchange <- max( abs( theta.k - theta.k0 ))
    }

    #--- OUTPUT
    res <- list( theta.k=theta.k, se.theta.k=se.theta.k )
    return(res)
}
##########################################################

.gdm.est.skillspace.traits <- gdm_est_skillspace_traits