R-Hmisc impute by cluster result - r

Question

I want to impute a variable x3 by the its mean corresponding to each cluster calculated considering other 2 variables X1 and X2.
I know that you can pass a function to impute from Hmisc package, like "mean" and it does the work. So I would like to pass a function that does all the following.
I use to write the code to do so:
df1 <- data.frame(x1=runif(1000,0,100),
x2=runif(1000,0,100),
x3=c(runif(900,0,100),rep(NA,100)))
I want to pass a function that does all of this:
clust<-kmeans(df1[,-grep('x3', colnames(df1))], 3)
df1$clust<-clust$cluster
library(plyr)
cc<-ddply(df1, 'clust',summarise, mean=mean(x3, na.rm=TRUE))
df2<-merge(df1,cc, by='clust')
df2$x3imputed2<-ifelse(is.na(df2$x3),df2$mean, df2$x3)
Is there a way to pass all this code as a function and use it in Hmisc? (I had a problem with ddply introducing x3 as a variable).
Something like the following:
ff<-function(i) {
clust<-kmeans(df1[,-grep(i, colnames(df1))], 3)
df1$clust<-clust$cluster
cc<-aggregate(df1[,i], by=list(clust=df1$clust), "mean", na.rm=TRUE)
df2<-merge(df1,cc, by='clust')
df2$x3imputed2<-ifelse(is.na(df2[, i]),df2$x, df2[,i])
}
f1$imputedx3<-with(df1, impute(x3,ff))
But I get an error:
empty cluster: try a better set of initial centers
And when I replace it by x3 I don't get the same error.

Answer 1

Try
library(lazyeval)
library(dplyr)
f1 <- function(dat, cname){
#get the third argument i.e, 'cname'
nm1 <- match.call()[[3]]
#paste 'imputed' for renaming the new column later
nm2 <- paste0(nm1, 'imputed')
#create an numeric column index that will be removed in kmeans calc
indx <- grep(cname, colnames(dat))
#get the 'kmeans' of the columns other than the 'cname'
clust <- kmeans(dat[,-indx],3)$cluster
#group by 'clust' and create new column with 'mutate'
dat %>%
group_by(clust=clust) %>%
mutate_(interp(~ifelse(is.na(v), mean(v, na.rm=TRUE), v),
v=as.name(cname))) %>%
#rename the column
setNames(., c(head(names(.),-1), nm2))
}
f1(df1, 'x3')
Or you could pass it without quotes by using v= lazy(cname)
f2 <- function(dat, cname){
nm1 <- match.call()[[3]]
nm2 <- paste0(nm1, 'imputed')
indx <- grep(nm1, colnames(dat))
clust <- kmeans(dat[,-indx],3)$cluster
dat %>%
group_by(clust=clust) %>%
mutate_(interp(~ifelse(is.na(v), mean(v, na.rm=TRUE), v),
v= lazy(cname))) %>%
setNames(., c(head(names(.),-1), nm2))
}
f2(df1, x3)