Issue with predicting cluster labeling, using DBSCAN object and Gower distance matrix for new data in R about dbscan

[peekxc]

I'm unable to reproduce the problem with the following test data:

library("dbscan")
data("moons")
dist_moons <- cluster::daisy(moons, metric = "gower", stand = FALSE)
model <- dbscan(dist_moons, eps = .23, minPts = 50)
predict(model, newdata = moons[1:5,], data = moons)

To answer one of your question, inspection of dbscan's predict method reveals what it does; using the original parameters you passed into the dbscan function, the predict method will recomputes the nearest neighbors at the given radius `eps' for all the data points (train and test), then assigns the new points to their corresponding clusters.

If you can attach the data you're working with or (if it's private data) attach a reproducible example set of data, I can further help you along.

I also may be misunderstanding the issue.

[mhahsler]

predict defaults to Euclidean distance and so does Gower's distance for only numeric attributes. Here is an example that breaks predict because it also uses species for clustering:

data(iris)
d <- cluster::daisy(iris, metric = "gower", stand = TRUE)
model <- dbscan(d, eps = .23, minPts = 50)
predict(model, newdata = iris[1:5,], data = iris)

Error in frNN(rbind(data, newdata), eps = object$eps, sort = TRUE, ...) : 
  x has to be a numeric matrix. 

I am not sure how to fix this, but I guess we need to pass on some sort of distance (or distance function to be used) between newdata and data.