You can save and load treatment plans. Note: treatments plans are intended to be used with the version of vtreat they were constructed with (though we try to make plans forward-compatible). So it is good idea to have procedures to re-build treatment plans.
The easiest way to save vtreat treatment plans is to use R’s built in saveRDS function.
To save in a file:
library("vtreat")
dTrainC <- data.frame(x=c('a','a','a','b','b',NA,NA),
z=c(1,2,3,4,NA,6,NA),
y=c(FALSE,FALSE,TRUE,FALSE,TRUE,TRUE,TRUE))
treatmentsC <- designTreatmentsC(dTrainC, colnames(dTrainC),
'y', TRUE,
verbose= FALSE)
fileName = paste0(tempfile(c('vtreatPlan')), '.RDS')
saveRDS(treatmentsC,fileName)
rm(list=c('treatmentsC'))And then to restore and use.
library("vtreat")
treatmentsC <- readRDS(fileName)
dTestC <- data.frame(x=c('a','b','c',NA),z=c(10,20,30,NA))
dTestCTreated <- prepare(treatmentsC, dTestC, pruneSig= c())
# clean up
unlink(fileName)Treatment plans can also be stored as binary blobs in databases. Using ideas from here gives us the following through the DBI interface.
con <- NULL
if (requireNamespace('RSQLite', quietly = TRUE) &&
requireNamespace('DBI', quietly = TRUE)) {
library("RSQLite")
con <- dbConnect(drv=SQLite(), dbname=":memory:")
# create table
dbExecute(con, 'create table if not exists treatments
(key varchar(200) primary key,
treatment blob)')
# wrap data
df <- data.frame(
key='treatmentsC',
treatment = I(list(serialize(treatmentsC, NULL))))
# Clear any previous version
dbExecute(con,
"delete from treatments where key='treatmentsC'")
# insert treatmentplan
# depreciated
# dbGetPreparedQuery(con,
# 'insert into treatments (key, treatment) values (:key, :treatment)',
# bind.data=df)
dbExecute(con,
'insert into treatments (key, treatment) values (:key, :treatment)',
params=df)
constr <- paste(capture.output(print(con)),collapse='\n')
paste('saved to db: ', constr)
}## Warning: Factors converted to character## [1] "saved to db: <SQLiteConnection>\n Path: :memory:\n Extensions: TRUE"rm(list= c('treatmentsC', 'dTestCTreated'))And we can read the treatment back in as follows.
if(!is.null(con)) {
treatmentsList <- lapply(
dbGetQuery(con,
"select * from treatments where key='treatmentsC'")$treatment,
unserialize)
treatmentsC <- treatmentsList[[1]]
dbDisconnect(con)
dTestCTreated <- prepare(treatmentsC, dTestC, pruneSig= c())
print(dTestCTreated)
}## x_catP x_catB z_clean z_isBAD x_lev_NA x_lev_x_a x_lev_x_b
## 1 0.4285714 -0.9807709 10.0 0 0 1 0
## 2 0.2857143 -0.2876737 20.0 0 0 0 1
## 3 0.0000000 0.0000000 30.0 0 0 0 0
## 4 0.2857143 9.6158638 3.2 1 1 0 0