"Local regression/LOESS using RapidMiner?"

keith

做es RM have a learner that can do local regression (also known as loess)? I'm thinking of something along the lines of the loess() or locfit() functions in R, or PROC LOESS in SAS.

http://en.wikipedia.org/wiki/Local_regression
http://www.sas.com/rnd/app/papers/loesssugi.pdf

Weka-LWL (locally weighted learning) seem to be similar in concept, but its not clear to me whether they are the same technique or not.

Also, does W-LWL support attribute weights?

Thanks,
Keith

land

Hi Keith,
unfortunatly RapidMiner does not provide something like a distancebased local regression. As far as I know we only provide kNN doing neighbourhood based local regression. But I don't know all the WEKA algorithms and their names are more confusing than helping.

We should add polynomial local regression by the time.

Greetings,
Sebastian