use multiple columns as variables with sapply
I have a dataframe
and I would like to apply a function that takes the values of three columns and computes the minimum difference between the three values.
#dataset
df <- data.frame(a= sample(1:100, 10),b = sample(1:100, 10),c= sample(1:100, 10))
#function
minimum_distance <- function(a,b,c)
{
dist1 <- abs(a-b)
dist2 <- abs(a-c)
dist3 <- abs(b-c)
return(min(dist1,dist2,dist3))
}
I am looking for something like:
df$distance <- sapply(df, function(x) minimum_distance(x$a,x$b,x$c) )
## errormessage
Error in x$a : $ operator is invalid for atomic vectors
While I can use ddply:
df2 <- ddply(df,.(a),function(r) {data.frame(min_distance=minimum_distance(r$a,r$b, r$c))}, .drop=FALSE)
This doesn't keep all of the columns. Any suggestions?
Edit: I ended up using:
df$distance <- mapply(minimum_distance, df$a, df$b, df$c)
Try mapply():
qq <- mapply(minimum_distance, df$a, df$b, df$c)
try this:
do.call("mapply", c(list(minimum_distance), df))
but you can write vectorized version:
pminimum_distance <- function(a,b,c)
{
dist1 <- abs(a-b)
dist2 <- abs(a-c)
dist3 <- abs(b-c)
return(pmin(dist1,dist2,dist3))
}
pminimum_distance(df$a, df$b, df$c)
# or
do.call("pminimum_distance", df)
I know this has been answered but I'd actually take a different approach that takes any number of columns and is more generalizable using an outer approach:
vdiff <- function(x){
y <- outer(x, x, "-")
min(abs(y[lower.tri(y)]))
}
apply(df, 1, vdiff)
I think this is a little cleaner and flexible.
EDIT: Per zach's comments I propose this more formalized function that works on data frames with non numeric columns as well by removing them and acting only on the numeric columns.
cdif <- function(dataframe){
df <- dataframe[, sapply(dataframe, is.numeric)]
vdiff <- function(x){
y <- outer(x, x, "-")
min(abs(y[lower.tri(y)]))
}
return(apply(df, 1, vdiff))
}
#TEST it out
set.seed(10)
(df <- data.frame(a = sample(1:100, 10), b = sample(1:100, 10),
c = sample(1:100, 10), d = LETTERS[1:10]))
cdif(df)