R/predict.gagam.R

In markcus1/gagam: Genetic Algorithm for Generalized Additive Models

#' @title Prediction from GAGAM
#'
#' @description Predict method for \code{gagam} objects.
#'
#' @param object \code{gagam} object (result of the \code{gagam()} function)
#' @param newdata Matrix or data frame with explanatory variables. The order of columns and their names must be the same as that used to construct the gagam object. If missing, the function extracts in-sample predictions.
#' @param reduc If prediction should be from one of the reduced models, specify the index of reduction. Specify only one index at a time. Default is NULL.
#' @param ... Any other arguments to pass to \code{\link[mgcv]{predict.gam}}.
#'
#' @return A vector of predictions.
#' @rdname predict.gagam
#' @export
#'
#' @examples
#' N <- 1000
#' set.seed(123)
#' xdat2 <- matrix(rnorm(N*10,0,1),nrow=N,ncol=10)
#' ydat2 <- 4*xdat2[,1]+5*xdat2[,2]+6*xdat2[,3]+(xdat2[,4])^2 + 4 + rnorm(N,0,0.25)
#'
#' xdat <- xdat2[1:500,]
#' ydat <- ydat2[1:500]
#'
#' \dontrun{
#' example_gagam <- gagam(ydat,xdat,Kvar = 6,no_gen = 50)
#' }
#'
#' xdat <- xdat2[501:1000,]
#' ydat <- ydat2[501:1000]
#'
#' newdata <- as.data.frame(cbind(as.matrix(ydat),xdat))
#'
#' \dontrun{
#' predict_gagam(example_gagam,newdata)
#' }
#'
#' @import utils
#' @import stats
#' @import mgcv
#' @import parallel
#' @import foreach
#' @import doParallel
#' @import doMC
predict.gagam <- function(object,..., newdata, reduc=NULL){
  gagam <- object
  if (is.null(reduc)){
    gam_obj <- gagam$best_gam

    if (missing(newdata)){
      predict.gam(gam_obj,...)
    } else {
      ndata <- as.data.frame(newdata)
      if (is.null(colnames(newdata))){
        colnames(ndata) <- paste0("x",1:as.numeric(NCOL(ndata)-1))
      } else {
        colnames(ndata) <- colnames(newdata)
      }

      predict.gam(gam_obj,newdata=ndata,...)
    }
  } else {
    if (is.null(gagam[[paste0("best_gam_red",reduc)]])){
      stop("gagam object doesn't contain the specified reduced model")
    }
    if (length())
    gam_obj <- gagam[[paste0("best_gam_red",reduc)]]

    if (missing(newdata)){
      predict.gam(gam_obj,...)
    } else {
      ndata <- as.data.frame(newdata)
      if (is.null(colnames(newdata))){
        colnames(ndata) <- paste0("x",1:as.numeric(NCOL(ndata)-1))
      } else {
        colnames(ndata) <- colnames(newdata)
      }
      colnames(ndata) <- c("y",paste0("x",1:as.numeric(NCOL(ndata)-1)))

      predict.gam(gam_obj,newdata=ndata,...)
    }

  }

}