parATimePredict: Batch Prediction for Time Series Using the Ensemble Models

In lspatial/sptemUS: Constrained Spatiotemporal Mixed-Effect Models for Exposure Estimation of Air Pollutants

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

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.

Value

The prediction result will be saved in the assigned path

Author(s)

Lianfa Li [e-mail]

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

#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")

lspatial/sptemUS documentation built on May 29, 2019, 3:42 a.m.