R/Module_multiFC.R

In MichaelFolkes/forecastR_package: Forecast stuff

# This module loops through multiple FC and generates overall summaries
# it uses the modules prepData(), fitModel(), and calcFC()


#' @title Process multiple models and produce forecasts with summaries
#'
#' @param data.file The data file read in with \code{read.csv}.
#' @param settings.list A list with list sub-elements. See details.
#' @param do.retro A Boolean (default is FALSE). Run a retrospective
#' @param retro.min.yrs An integer of length one. Default is 15.
#' @param out.type A character vector of length one. The choices are: "short" or
#'   "full". See value for details.
#' @param int.type Either "None","Retrospective","Prediction", or "Bootstrap"
#' @param int.n  number of samples used for the interval calculations
#' @param boot.type = "meboot" or "stlboot", only applies if int.type = "Bootstrap"
#' @param tracing A Boolean (default is FALSE)
#'
#' @details The settings.list argument is a list defining each forecast model in
#'   separate lists that can be uniqely named. Each model list must have the
#'   elements  "model.type" and "settings". The settings element is also a list.
#'   For example: \code{settings.list= list(Naive1 =
#'   list(model.type="Naive",settings=list(avg.yrs=1)), SibRegSimple =
#'   list(model.type="SibRegSimple",settings=NULL))}
#'
#' @return A list is produced. If argument out.type = "short" and do.retro = FALSE,
#' then only generate a summary table of point forecasts.
#'   If out.type = "short" and do.retro= TRUE, then generate 3 versions of
#'   retrospective summary and array of fitted performance measures. If out.type
#'   ="full" and do.retro= TRUE, then also store the model fits for each retro
#'   year. If out.type ="short" and = TRUE, then TBD. If out.type ="full"
#'   and do.boot= TRUE, then TBD
#' @export
#'
#' @examples
multiFC <- function(data.file, settings.list, do.retro = FALSE, retro.min.yrs=15, out.type=c("short", "full"), int.type = c("None","Retrospective","Prediction", "Bootstrap"), int.n = 100, boot.type = "meboot", tracing=FALSE){

## NEED TO FIX
# Settings.list -> fit.settings
# add in fc.settings (just like in doRetro)

# do.boot = TRUE -> !!!)

out.type <- match.arg(out.type)


dat.prepped <-  prepData(data.file,out.labels="v2")  # prep data for the model fit


# start a list for storing
out.list <- list()

model.list <- names(settings.list)
ages.list <- names(dat.prepped$data )

# create the int.array anyway, even of int.type= None
# this way the GUI can still grab the empty array for display

			if(length(ages.list)>1){
			int.array <- array(NA, dim= c(length(model.list),5,length(ages.list)+1),
										dimnames=list(model.list,paste0("p",c(10,25,50,75,90)),c(ages.list,"Total"))
									)}

			if(length(ages.list)==1){  # for no age data
			int.array <- array(NA, dim= c(length(model.list),5,1),
										dimnames=list(model.list,paste0("p",c(10,25,50,75,90)),"Total")
									)}



for(model.name in names(settings.list) ){

print(paste("starting",model.name))

	#extract the settings
	model.use <- settings.list[[model.name]]$model.type
	settings.use <-  settings.list[[model.name]]$settings

	if(tracing){
		print("---------------------------------")
		print(paste("starting",model.name))
		print(paste("model.type =",model.use))
		print("settings =")
		print(settings.use)
		}

	model.fitted <- fitModel(model= model.use, data = dat.prepped$data, settings = settings.use,tracing=FALSE)


	fc.calc <- calcFC(fit.obj= model.fitted,data = dat.prepped$data, fc.yr= dat.prepped$specs$forecastingyear,
					settings = settings.use, tracing=tracing, predictors =  dat.prepped$predictors, covariates = dat.prepped$covariates)

	out.list[[model.name]] <- fc.calc



	if(int.type=="Prediction"){



			int.quants <- doSampleFromInt(fc.obj=fc.calc, interval.n=int.n,interval.quants=TRUE)



			if(tracing){
			print("prediction quants")
			print(int.quants)
			}
		} # end if prediction interval



	if(do.retro | int.type=="Retrospective"){



	 retro.out  <- 	doRetro(model= model.use, data = dat.prepped$data,
 			  predictors =  dat.prepped$predictors, covariates = dat.prepped$covariates,
				retro.settings = list(min.yrs=retro.min.yrs),
				fit.settings = settings.use,
				fc.settings = settings.use,
				tracing=tracing,out.type=out.type,
				interval.n = int.n,
				interval.quants = TRUE,
				pt.fc.in = fc.calc)


		if(int.type=="Retrospective"){
			int.quants <- retro.out$retro.interval

			if(tracing){
				print("retrospective quants")
				print(int.quants)
				}

			}

			out.list[[model.name]] <- c(out.list[[model.name]],list(retro=retro.out ))




			} # end if retro interval


		if(int.type=="Bootstrap"){




			boot.int <- doBoot(data= dat.prepped, args.fitmodel= list(model= model.use, settings = settings.use),
						args.calcfc = list(fc.yr= dat.prepped$specs$forecastingyear,  settings = settings.use),
						args.boot = list(boot.type=boot.type, boot.n= int.n , plot.diagnostics=FALSE),
						full.out = TRUE, plot.out=FALSE)

			int.quants <-  as.data.frame(lapply(boot.int ,function(x){quantile(x,probs=c(0.1,0.25,0.5,0.75,0.9))}))

			if(tracing){
				print("bootstrap quants")
				print(int.quants)
				}


			} # end if bootstrap interval


if(int.type != "None"){ int.array[model.name,,] <- as.matrix(int.quants) }


}  # end looping through models



# create a summary array of all the point forecasts and quantiles
extract.ptfc <- function(x){return(x$pt.fc)}
table.ptfc <- as.data.frame(do.call(rbind,lapply(out.list,FUN=extract.ptfc)))
dimnames(table.ptfc)[[1]] <- names(settings.list)







if(do.retro){


	# create a summary array of all the retrospective pm - 3 versions
	# do this in nested loops for now, given that have to reorg into array as well

	retros <- list()

	for(retro.type in c("fitted.pm.last", "retro.pm.all.constantyrs","retro.pm.all.varyrs" ,"retro.pm.bal")){

	sample.retro <- out.list[[1]]$retro[[retro.type]]

	array.retro <- array(NA,dim=c(length(out.list), dim(sample.retro)),
				dimnames=list(names(out.list),dimnames(sample.retro)[[1]],dimnames(sample.retro)[[2]]))

	for(model.name in names(out.list)){array.retro[model.name , , ] <- out.list[[model.name]]$retro[[retro.type]]}

	retros[[retro.type]] <- array.retro

	} # end looping through retro summary types

	} # end if retro

if(out.type=="short" & !do.retro ) {return.list <-  list(table.ptfc = table.ptfc)}
if(out.type=="short" & do.retro) {return.list <-  list(table.ptfc = table.ptfc, retro.pm = retros)}
if(out.type=="full" & do.retro) {return.list <-  list(table.ptfc = table.ptfc, retro.pm = retros,retro.details = out.list)}


if(int.type %in% c("Retrospective","Prediction", "Bootstrap")){
			return.list <- c(return.list, list(int.array = int.array))
			}

return(return.list)

} # end multiFC