The library is designed with a massive parallelization use in mind,
and this results in multiple independent rData files
(res.filepath) containing formula performances.
Periodically, you have to join all these res.filepath in
a shared base.filepath. This code is an example on how to
do that.
Base filepaths maybe split by formula length. I find this useful to avoid having a single big rData file.
l1.filepath <- paste0('regression/regression',type,'.exploration.l1.rData')
l2.filepath <- paste0('regression/regression',type,'.exploration.l2.rData')
l3.filepath <- paste0('regression/regression',type,'.exploration.l3.rData')
l1.res <- readRDS(l1.filepath)
l2.res <- readRDS(l2.filepath)
l3.res <- readRDS(l3.filepath)You may want to know what fraction of the formula space you have explored. This is a way to do that, the assumption is that you have a transformation list in your environnment as defined in get-started vignette.
max.n.squares <- 2
complete.regressors <- symbolicr::compute.regressors.names(regressors.df, max.n.squares, transformations)
tot.regressors <- length(complete.regressors)
# percentage of L=1
100*nrow(l1.res)/choose(tot.regressors,1)
# percentage of L=2
100*nrow(l2.res)/choose(tot.regressors,2)
# 10.61434
# percentage of L=3
100*nrow(l3.res)/choose(tot.regressors,3)# all possible table names
hyperparams <- expand.grid(formula.len=seq(1,3), n.squares=seq(0,2), seed=seq(0,1010))
# read them ALL!
new.res.l <- apply(hyperparams, MARGIN=1, simplify = F, FUN=function(row){
flen <- row[['formula.len']]
nsq <- row[['n.squares']]
seed <- row[['seed']]
if(seed == 0){
res.filepath <- file.path(out.folder, paste0('regression',type,'.exploration.fl.',flen,'.ord.',nsq,'.rData'))
}else{
res.filepath <- file.path(out.folder, paste0('regression',type,'.exploration.fl.',flen,'.ord.',nsq,'seed.',seed,'.rData'))
}
if(file.exists(res.filepath)){
print(res.filepath)
res <- readRDS(res.filepath)
cres <- res
}else{
cres <- data.frame(base.pe=double(),base.cor=double(),base.r.squared=double(),
base.max.pe=double(), base.iqr.pe=double(), base.max.cooksd=double(), base.max.cooksd.name=character(),
vars=character(), n.squares=integer(), formula.len=integer())
}
return(cres)
})
# drop empty
new.res.l<- new.res.l[which(sapply(new.res.l, nrow) > 0)]
new.res <- do.call(rbind, new.res.l)
dedup.res <- new.res[!duplicated(new.res$vars), ]
# how many NEW formulas of length 1/2/3/4?
l1.dedup.res <- dedup.res[dedup.res$formula.len==1, ]
l2.dedup.res <- dedup.res[dedup.res$formula.len==2, ]
l3.dedup.res <- dedup.res[dedup.res$formula.len==3, ]You may want to take note of the number of new results that will be appended to the rData, so that you can evaluate only the new results.
# take note for analyze results..
nrow(l1.dedup.res)
nrow(l2.dedup.res)
nrow(l3.dedup.res)