Closed carlesmila closed 1 year ago
@carlesmila Thanks! I added the W statistic for sample-to-sample as well. I see you limited it to the use of geographic distances. Can't it be used for feature space distances in the same way?
It certainly can, but I thought it'd be better to explore and validate it first prior to providing it, but feel free to disagree :)
I tested it and it works as expected but I think @JanLinnenbrink is exploring the feature space version of knndm. @JanLinnenbrink let's keep in mind to integrate that once you feel confident with it
@HannaMeyer Added W statistic for test data and/or CV folds in
geodist
as attributes fortype="geo"
.@JanLinnenbrink unfortunately we cannot use this implementation for the simulations since
geodist
doesn't acceptppoints
and we need our W to be exactly comparable between CV methods. Will PR tokNNDM_paper
ASAP.