sptemExp (version 0.1.4)

parATimePredict: Batch Prediction for Time Series Using the Ensemble Models

Description

Batch predictions for the time series using ensemble models generated by the function, parSpModel

Usage

parATimePredict(mdPath,newPnts,cols=NA,bnd,c=1,outPath="/tmp",idF="siteid",ridF="rid")

Arguments

mdPath

The path where multiple ensemble models are saved by parSpModel

newPnts

New data locations corresponding to the predcitions.

cols

Columns where there are NAs. NAs must be removed before prediction. Default: NA

bnd

The same BND object as that used in parSpModel, for spatial effect models.

c

CPU cores to support parallel computing.

outPath

The output file path, file named after the model id.

idF

Unique identifier

ridF

Region id used in spatial effect modeling

Value

The prediction result will be saved in the assigned path

Details

This function aims to use the muiltiple models with their performance metrics to make the predictions for the new dataset with their spatial location.

References

Breiman, L., 1996. Bagging Predictors. Machine Learning 24, 123-140. Lianfa Li et al, 2017, Constrained Mixed-Effect Models with Ensemble Learning for Prediction of Nitrogen Oxides Concentrations at High Spatiotemporal Resolution, ES & T, DOI: 10.1021/acs.est.7b01864

Examples

# NOT RUN {
# }
# NOT RUN {
#Use the PM2.5 examples

dPath=tempdir()
modelPath=paste(dPath,"/models",sep="")
unlink(modelPath,recursive = TRUE)
dir.create(modelPath)

prePath=paste(dPath,"/preds",sep="")
unlink(prePath,recursive = TRUE)
dir.create(prePath)

data("trainsample","bnd")
aform=paste0('logpm25 ~sx(rid,bs ="mrf",map =bnd)+sx(monthAv,bs="rw2")')
aform=paste0(aform,'+sx(ndvi,bs="rw2")+sx(aod,bs="rw2")+sx(wnd_avg,bs="rw2")')

formulaStrs=c(aform)

trainsample$tid=as.numeric(strftime(trainsample$date, format = "%j"))

trainsample$logpm25=log(trainsample$pm25)
tids=c(91)

parSpModel(trainsample,bnd,formulaStrs,tidF="tid",tids,c=2,
       nM=3,modelPath,idF="siteid",dateF="date",obsF="pm25")

amodelPath=paste(dPath,"/models/t_",tids[1],"_models",sep="")
data("shd140401pcovs","bnd")
shd140401pcovs_part=shd140401pcovs[c(1:1000),]

cols=c("aod","ndvi","wnd_avg","monthAv")
parATimePredict(amodelPath,newPnts=shd140401pcovs_part,
       cols,bnd=bnd,c=2,prePath,idF="gid",ridF="rid")
# }
# NOT RUN {
# }