I'm not sure how this is relevant - the difficult part is doing the retrieval based off of similarity matching in an efficient way. Calling the cdist function to compare all vectors would be very slow.