R/gen.rec.raw.R

In cacIRT: Classification Accuracy and Consistency under Item Response Theory

#theta is only for labels here. It is unneeded

gen.rec.raw <- function(Pij, theta.names = NULL){ 
  #matches results from R package classify wlord
  
  nn <- dim(Pij)[1] 
  
  Pij[round(Pij,5)==1]<- 1-1e-4
  Pij[round(Pij,5)==0]<- 1e-4
  
  if(length(dim(Pij))==2){
    ni <- dim(Pij)[2]
    sc <- ni+1
    
    Qjt <- 1-Pij
    H <- apply(Qjt,1,prod)
    Zjt	<- Pij/Qjt
    
    zm <- array(0, dim = c(nn, sc, ni))	
    zm[,1,] <- 1
    zm[,2,1] <- Zjt[,1]
    
    for(i in 2: ni){
      for(s in 2:sc){
        zm[,s,i] <- zm[,s,i-1] + Zjt[,i]*zm[,s-1,i-1]}}	
    
    out<- zm[,,ni]*H
    rownames(out) <- theta.names
    colnames(out) <- paste("x = " , 0:ni)
    return(out)
  }else{	
    
    nk <- dim(Pij)[2]	
    ni <- dim(Pij)[3]
    
    zm <- array(0, dim = c(nn, nk*ni, ni))	
    zm[,1:nk,1] <- Pij[,,1] 
    
    for(i in 2:ni){
      for(s in i:(i*nk)){
        
        zm[ , s, i] <- rowSums(
          as.matrix(	zm[,s-1:(min(c(s-1, nk))),i-1] * Pij[,1:(min(c(s-1, nk))),i]))
        
      }} 			
    
    out<- zm[,ni:(nk*ni),ni]
    rownames(out) <- theta.names
    colnames(out) <-paste("x = " , 0:(((nk-1)*ni)))
    return(out)}
}