subset of data frame on based on multiple conditions - r

Question

I'm actually having a trouble with a particular task of my code. I have a data frame as
n <- 6
set.seed(123)
df <- data.frame(x=paste0("x",seq_along(1:n)), A=sample(c(-2:2),n,replace=TRUE), B=sample(c(-1:3),n,replace=TRUE))
#
# x A B
# 1 x1 -1 1
# 2 x2 1 3
# 3 x3 0 1
# 4 x4 2 1
# 5 x5 2 3
# 6 x6 -2 1
and a decision tree as
A>0;Y;Y;N;N
B>1;Y;N;Y;N
C;1;2;2;1
that I load by
dt <- read.csv2("tmp.csv", header=FALSE)
I'd like to create a loop for all the possible combinations of (A>0) and (B>1) and set the C value to the subset x column that satisfy that condition. So, here's what I did
nr <- 3
nc <- 5
cond <- dt[1:(nr-1),1,drop=FALSE]
rule <- dt[nr,1,drop=FALSE]
subdf <- vector(mode="list",2^(nr-1))
for (i in 2:nc) {
check <- paste0("")
for (j in 1:(nr-1)) {
case <- paste0(dt[j,1])
if (dt[j,i]=="N")
case <- paste0("!",case)
check <- paste0(check, "(", case, ")" )
if (j<(nr-1))
check <- paste0(check, "&")
}
subdf[i] <- subset(df,check)
subdf[i]$C <- dt[nr,i]
}
unlist(subdf)
unfortunately, I got an error using subset as by this, it cannot parse the conditions from my string statements. what should I do?

Answer 1

Your issue is your creating of the subset: the subset commands expects a boolean and you gave it a string. ('check'). So the simplest solution here is to add a 'parse'. I feel there is a more elegant way to solve this problem and I hope someone'll come along and do it, but you can fix the final part of your code with the following
mysubset <- subset(df,with(df,eval(parse(text=check))))
if(nrow(mysubset)>0){
mysubset$C <- dt[nr,i]
}
subdf[[i]]<-mysubset
I have added the parse/eval part to generate a vector of booleans to subset only the 'TRUE' cases, and added a check for whether C could be added (will give error if there are no rows).
Based on the previous answer, I came up with a more elegant/practical way of generating a vector of combined rules, and then applying them all to the data, using apply/lapply.
##create list of formatted rules
#format each 'building' block separately,
#based on rows in 'dt'.
part_conditions <- apply(dt[-nrow(dt),],MARGIN=1,FUN=function(x){
res <- sprintf("(%s%s)", ifelse(x[-1]=="Y","","!"), x[1])
})
# > part_conditions
# 1 2
# [1,] "(A>0)" "(B>1)"
# [2,] "(A>0)" "(!B>1)"
# [3,] "(!A>0)" "(B>1)"
# [4,] "(!A>0)" "(!B>1)"
#combine to vector of conditions
conditions <- apply(part_conditions, MARGIN=1,FUN=paste, collapse="&")
# > conditions
# [1] "(A>0)&(B>1)" "(A>0)&(!B>1)" "(!A>0)&(B>1)" "(!A>0)&(!B>1)"
#for each condition, test in data wheter condition is 'T'
temp <- sapply(conditions, function(rule){
return(with(df, eval(parse(text=rule))))
}
)
rules <- as.numeric(t(dt[nrow(dt),-1]))
#then find which of the (in this case) four is 'T', and put the appropriate rule
#in df
df$C <- rules[apply(temp,1,which)]
> df
x A B C
1 x1 -1 1 1
2 x2 1 3 1
3 x3 0 1 1
4 x4 2 1 2
5 x5 2 3 1
6 x6 -2 1 1