weiA2Ens: Ensemble Weighted Prediction of Mixed Models

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

Description

Weighted average and standard deviation of multiple models. The weights can be the model's performance metrics such as R2 or RMSE.

Usage

weiA2Ens(pPath,mFile,metrF="rmse",preF="pre",idF="gid",dateF=NA)

Arguments

pPath	Path for the prediction files from multiple models with the unified format and field names. File format: CSV with head.
mFile	File path for the corresponding multiple models's performance. CSV format:mid, r2, rmse.
metrF	target metric such as rmse or r2 to weigh the model's output
preF	prediction field name
idF	unique identifier string
dateF	date string if any

Value

id and corresponding mean and standard deviation. Format: dataframe

Author(s)

Lianfa Li: [e-mail]

Examples

#First generate the prediction dataset and metrics.
# In application, you can use parSpModel to train models and
# get the models's performance metrics, and use the parATimePredict function to make the prediction

# Simulared data

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

nr=2000;nmod=80
for(i in c(1:nmod)){ # i =1
  dset=data.frame(gid=c(1:nr),rid=sample(c(1:30),size=nr,replace=TRUE),stringsAsFactors = FALSE)
  dset$pre=dset$gid%%nmod+rnorm(nr,mean=5,sd=9)+runif(nr,0,1)
  dset$gid=paste("c",dset$gid,sep="")
  afile=paste(pPath,"/m",i,".csv",sep="")
  write.csv(dset,file=afile,row.names = FALSE)
}

modelsMetrics=data.frame(mid=c(1:nmod),r2=runif(nmod,0.6,0.9),rmse=runif(nmod,20,60))
mfile=paste(dPath,"/model_metrics.csv",sep="")
write.csv(modelsMetrics,file=mfile,row.names = FALSE)
result=weiA2Ens(pPath,mfile,metrF="rmse","pre","gid","gid")

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