predict.kproto() does not work on data with a single row

[galen-ft]

Issue description

Predict does not work on data with a single row.

The cause is in the way distances for categorical variables are computed in predict.kproto() .

Whenever the input data x has only 1 row, the result of sapply simplifies to a logical vector instead of a matrix and the subsequent call to rowSums fails.

d2 <- sapply(which(catvars), function(j) return(x[,j] != rep(protos[i,j], nrows)) )
d2[is.na(d2)] <- FALSE
if(length(lambda) == 1) d2 <- lambda * rowSums(d2) # <---- rowSums fails

Reproducible example

library(clustMixType)

set.seed(123)
model <- kproto(x = iris, k = 4)
predict(model, iris[1, ])
# which yields
# Error in rowSums(d2) : 'x' must be an array of at least two dimensions

Suggested fix

One quick fix would be to convert d2 to matrix whenever x is a data.frame with one row only:

d2 <- sapply(which(catvars), function(j) return(x[,j] != rep(protos[i,j], nrows)) )
if (NROW(x) == 1) d2 <- matrix(data = d2, nrow = 1, byrow = TRUE, dimnames = list(NULL, names(x)[catvars])) # <- FIX
d2[is.na(d2)] <- FALSE
if(length(lambda) == 1) d2 <- lambda * rowSums(d2) # <---- rowSums now succeeds

[gaborcsardi]

Hi, this is a read only mirror of CRAN, please see the package authors in the DESCRIPTION file. Look for Maintainer, BugReports and URL. Thanks!

[galen-ft]

Hi, thanks for your reply. I've already emailed the maintainer, but thought I'd also share the issue here in case any one stumbles upon it.