inst/utils/allFit.R

##
library("lme4")
require("optimx")   ## for optim optimizers
   ## (optimx-specific optimizers require explicit gradients --
   ##  we could use numDeriv::grad, but this seems to defeat
   ##  the intention)
require("nloptr")

meth.tab.0 <- cbind(optimizer=
                      rep(c("bobyqa",
                            "Nelder_Mead",
                            "nlminbw",
                            "nmkbw",
                            "optimx",
                            "nloptwrap" ),
                          c(rep(1,5),2)),
                  method= c(rep("",4), "L-BFGS-B",
                  "NLOPT_LN_NELDERMEAD", "NLOPT_LN_BOBYQA"))

nlminbw   <- lme4:::nlminbwrap
namedList <- lme4:::namedList

if (require("dfoptim")) {
    nmkbw <- function(fn,par,lower,upper,control) {
	if (length(par)==1) {
	    res <- optim(fn=fn,par=par,lower=lower,upper=100*par,
			 method="Brent")
	} else {
	    if (!is.null(control$maxfun)) {
		control$maxfeval <- control$maxfun
		control$maxfun <- NULL
	    }
	    res <- nmkb(fn=fn,par=par,lower=lower,upper=upper,control=control)
	}
	res$fval <- res$value
	res
    }
} else { ## pkg{dfoptim} not available
    ## for nmkb
    meth.tab.0 <- meth.tab.0[meth.tab.0[,"optimizer"] != "nmkbw",]
}



##' Attempt to re-fit a [g]lmer model with a range of optimizers.
##' The default is to use all known optimizers for R that satisfy the
##' requirements (do not require explicit gradients, allow
##' box constraints), in three categories; (i) built-in
##' (minqa::bobyqa, lme4::Nelder_Mead, nlminbwrap), (ii) wrapped via optimx
##' (most of optimx's optimizers that allow box constraints require
##' an explicit gradient function to be specified; the two provided
##' here are really base R functions that can be accessed via optimx,
##' (iii) wrapped via nloptr; (iv)
##'
##' @param m a fitted model
##' @param meth.tab a matrix (or data.frame) with columns
##' - method  the name of a specific optimization method to pass to the optimizer
##'           (leave blank for built-in optimizers)
##' - optimizer  the \code{optimizer} function to use
##' @param verbose print progress messages?
##' @return a list of fitted \code{merMod} objects
##' @seealso slice, slice2D in the bbmle package
##' @examples
##' library(lme4)
##' gm1 <- glmer(cbind(incidence, size - incidence) ~ period + (1 | herd),
##'                  data = cbpp, family = binomial)
##' gm_all <- allFit(gm1)
##' ss <- summary(gm_all)
##' ss$fixef               ## extract fixed effects
##' ss$llik                ## log-likelihoods
##' ss$sdcor               ## SDs and correlations
##' ss$theta               ## Cholesky factors
##' ss$which.OK            ## which fits worked


allFit <- function(m, meth.tab = meth.tab.0,
                   data=NULL,
                   verbose=TRUE,
                   maxfun=1e5)
{
    stopifnot(length(dm <- dim(meth.tab)) == 2, dm[1] >= 1, dm[2] >= 2,
	      is.character(optimizer <- meth.tab[,"optimizer"]),
	      is.character(method    <- meth.tab[,"method"]))
    fit.names <- gsub("\\.$","",paste(optimizer, method, sep="."))
    res <- setNames(as.list(fit.names), fit.names)
    for (i in seq_along(fit.names)) {
        if (verbose) cat(fit.names[i],": ")
        ctrl <- list(optimizer=optimizer[i])
        ctrl$optCtrl <- switch(optimizer[i],
                               optimx    = list(method   = method[i]),
                               nloptWrap = list(algorithm= method[i]),
                               list(maxfun=maxfun))
        ctrl <- do.call(if(isGLMM(m)) glmerControl else lmerControl, ctrl)
        tt <- system.time(rr <- tryCatch(update(m, control = ctrl),
                                         error = function(e) e))
        attr(rr, "optCtrl") <- ctrl$optCtrl # contains crucial info here
        attr(rr, "time") <- tt  # store timing info
        res[[i]] <- rr
        if (verbose) cat("[OK]\n")
    }

    structure(res, class = "allfit", fit = m, sessionInfo =  sessionInfo(),
              data = data # is dropped if NULL
              )
}

print.allfit <- function(object, width=80, ...) {
    cat("original model:\n")
    f <- attr(object,"fit")
    ss <- function(x) {
        if (nchar(x)>width) {
            strwrap(paste0(substr(x,1,width-3),"..."))
        } else x
    }
    ff <- ss(deparse(formula(f)))
    cat(ff,"\n")
    cat("data: ",deparse(getCall(f)$data),"\n")
    cat("optimizers (",length(object),"): ",
        ss(paste(names(object),collapse=", ")),"\n",
        sep="")
    which.bad <- sapply(object,is,"error")
    if ((nbad <- sum(which.bad))>0) {
        cat(nbad,"optimizers failed\n")
    }
    cat("differences in negative log-likelihoods:\n")
    nllvec <- -sapply(object[!which.bad],logLik)
    print(summary(nllvec-min(nllvec)))
}

summary.allfit <- function(object, ...) {
    which.OK <- !sapply(object, is, "error")
    objOK <- object[which.OK]
    msgs <- lapply(objOK, function(x) x@optinfo$conv$lme4$messages)
    fixef <- do.call(rbind, lapply(objOK, fixef))
    llik <- sapply(objOK, logLik)
    times <- t(sapply(objOK, attr, "time"))
    feval <- sapply(objOK, function(x) x@optinfo$feval)
    sdcor <- do.call(rbind, lapply(objOK, function(x)
        as.data.frame(VarCorr(x))[, "sdcor"]))
    theta <- do.call(rbind, lapply(objOK,
                                   function(x) getME(x, "theta")))
    cnm <- lme4:::tnames(objOK[[1]])
    if (sigma(object[[1]])!=1) cnm <- c(cnm,"sigma")
    colnames(sdcor) <- unname(cnm)
    sdcor <- as.data.frame(sdcor)
    namedList(which.OK, msgs, fixef, llik, sdcor, theta, times, feval)
}

## should add a print method for summary:
##  * fixed effects, random effects: summary of differences?

Try the lme4 package in your browser

Any scripts or data that you put into this service are public.

lme4 documentation built on April 4, 2018, 1:04 a.m.