parSpModel: Generation of Spatiotemporal Models by Bootstrap Aggregating

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

Description

Generation of multiple models using bootstrap aggregating, supporting multi-cores based parallel computing.

Usage

parSpModel(tSet,bnd,fS,tidF="tid",tids, c=1,
     nM=30,mPath,idF="siteid",dateF="date",obsF="pm25")

Arguments

tSet	Dataframe of the training dataset, including the measurements of the target variable and covariates.
bnd	BND object used in saptial effect modeling (BayesX)
fS	Formula string, like that used in BayesX
tidF	time id (ensemble models for each time point)
tids	all the time ids for which multiple models will be trained.
c	CPU core number
nM	Number of ensemble models for each time point.
mPath	Path where the models will be saved.
idF	Unique location name
dateF	Time id
obsF	Target variable name

Details

Batch training of the models using the multi-cores based parallel computing

Value

The model will be saved into 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

# Example the PM2.5 data for Shandong

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

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,mPath,idF="siteid",dateF="date",obsF="pm25")

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