topic Re: HPSplit Guidance On Specific Parameters To Use in Statistical Procedures

HPSplit Guidance On Specific Parameters To Use

NKormanik — Fri, 06 Sep 2019 07:02:29 GMT

Anyone out there with Proc HPSplit experience that can shed light on ANY of the below parameters, specific numbers used where the (?) is -- what worked for you, or what's to be avoided, and why?

How might one decide which parameters to use? Here are the main ones:

cvmethod=random(?)
intervalbins=?
maxbranch=?
maxdepth=?
mincatsize=?
minleafsize=?
nsurrogates=?

grow ?
prune ?

I'm interested in what experiences you've had using Proc HPSplit. How useful did you find it?

If not Proc HPSplit, what other did you find more to your liking?

Thanks!

Nicholas Kormanik

Re: HPSplit Guidance On Specific Parameters To Use

JosvanderVelden — Fri, 06 Sep 2019 09:28:48 GMT

Maybe the paper "An introduction to classification and regression trees with PROC HPSPLIT" has some useful information: http://www.mwsug.org/proceedings/2018/AA/MWSUG-2018-AA-42.pdf

You can also read this paper: https://www.lexjansen.com/wuss/2019/185_Final_Paper_PDF.pdf "PROC DTREE VS PROC HPSPLIT"

Re: HPSplit Guidance On Specific Parameters To Use

NKormanik — Sun, 08 Sep 2019 00:34:28 GMT

Thanks for the leads, JosvanderVelden.

No specific advice on the parameters. Curiously.

HPSplit comes with defaults. Maybe those are 'good enough'?

Why change the defaults, unless one has good reason to. Eh?