It could be a problem if you have many variable that need to handle in this way.
I'll add this feature into woebin function. Thanks.

from scorecard.

Thanks. You can specify the breakpoints via option break_list in the function woebin. And you can get the optimal binning based on the dataset that excludes the two special values.

library(scorecard) library(data.table) dat<-data.table( y=c(0,0,0,1,1,1,1,1,0,0,1,1,1,0,0,0,1,1,1,0), x=c(1,2,3,4,5,888,888,888,9,10,666,666,666,666,15,16,17,18,19,20))

# get optimal breakpoints for rest dataset
bins <- woebin(dat[x != 666 & x != 888], "y")

# specify the breakpoints
bins2 <- woebin(dat, "y", breaks_list = list(x=c(16, 18, 666, 888)))
woebin_plot(bins2)

from scorecard.

Thanks. very detailed answer. In my case, i mean the value '666' and '888' is a categorical variable. so we should convert it as a factor before woebin.
wheile,the following code regard '666' and '888' as numerical variable.
bins2 <- woebin(dat, "y", breaks_list = list(x=c(16, 18, 666, 888)))
in my opinion, i can do following
library(scorecard)
library(data.table)
library(dplyr)
dat<-data.table( y=c(0,0,0,1,1,0,0,1,0,0,1,1,1,0,0,0,1,1,1,0), x=c(1,2,3,4,5,888,888,888,9,10,666,666,666,666,15,16,17,18,19,20))

special value

sp<-c(666,888)
#get the special data
dat_sp<-filter(dat, x %in% sp)
#get normal data
dat_nor<-filter(dat, !x %in% sp)

convert it to factor

dat_sp$x<-as.factor(dat_sp$x)
bins_sp <- woebin(dat_sp, "y")
woebin_plot(bins_sp)

bin for normal data

bins_nor <- woebin(dat_nor, "y")
woebin_plot(bins_nor)

and now the question is 1) how to combine these two plot in one plot. 2) how to combine these two woe for the variable in this case becase we can't do it by rbind function simply. if we have many such variable , how to get woe ? what I really warried is that we can't do bin for many variables automatically. for example,many functions in your package support batch process, obviously,if we bin for special value and normal value respectively,it destroies batch process.

from scorecard.

get optimal breakpoints for rest dataset

bins <- woebin(dat[x != 666 & x != 888], "y")
here, if we could really get the optimal breakponits once we omit some special samples?
and i realized that compute woe separately is wrong.

from scorecard.

Thanks again for your nice solution! Looking forward to your improved version.

from scorecard.

see the following example:

library(scorecard)

dat <- data.frame(y=c(0,0,0,1,1,1,1,1,0,0,1,1,1,0,0,0,1,1,1,0),
x=c(1,2,3,4,5,888,888,888,9,10,666,666,666,666,15,16,17,18,19,20))

#' specify two values as two class
bin = woebin(dat, "y", special_values = c(666,888))
#' specify two values as one class
bin2 = woebin(dat, "y", special_values = c("666%,%888"))

from scorecard.