s-news
[Top] [All Lists]

Re: [S] rpart

To: calzola@mailhost.apa.com
Subject: Re: [S] rpart
From: "Atkinson, Beth" <atkinson@mayo.edu> (Beth Atkinson)
Date: Wed, 24 Jun 1998 16:17:42 -0500
Cc: s-news@wubios.wustl.edu
Sender: owner-s-news@wubios.wustl.edu
 A few of suggestions:

1) Turn off the cross-validation and surrogate options (maxsurrogate=0, xval=0).
   That should speed things up.

2) Try using a smaller subset just to make sure you've got the code working
   (say n=500).

3)  There is a stand-alone version of rpart out on statlib in the 'general'
    section.  This may work best for extremely large datasets.   


~> I am trying to use rpart for a classification problem with 10,000
~> observations.  The dependent variable has 11 levels and I have 4
~> predictors: 2 continuous and 2 categorical (with 5 and 4 levels
~> respectively). It runs fine if I take the defaults, but if I use a loss
~> matrix (with off-diagonal elements equal to the absolute value of the
~> misclassification error it just runs forever and never finishes).
~> 
~> Am I being too ambitious here and the problem is just too big? (10,000
~> is 25% of the full dataset). 
~> 
~> Carlos Alzola
~> calzola@apa.com
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu.  To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message:  unsubscribe s-news

<Prev in Thread] Current Thread [Next in Thread>
  • [S] rpart, Carlos Alzola
    • Re: [S] rpart, Atkinson, Beth <=