[19019] in s-news-athena
[S] summary: estimates for nodes of tree object
daemon@ATHENA.MIT.EDU (Hannoever)
Tue Aug 24 02:47:41 1999
Date: Tue, 24 Aug 1999 08:29:39 +0200 (MET DST)
From: Hannoever <hann@Psyres-Stuttgart.DE>
To: splus mailinglist <s-news@wubios.wustl.edu>
Message-Id: <Pine.GSO.3.96.990824081609.13521A-100000@sunserv>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Dear S+ users,
last week I posted the question, if and how to estimate the
misclassification rates for every single node of a tree. You can find the
original question below.
Since I promised to summarize to the list but did not get any response,
this is my summary.
I still appreciate any help or suggestions and of course will summarize
these to the list.
All the best!
Wolfgang
*********************************************
Wolfgang Hannoever
Forschungsstelle fuer Psychotherapie
Christian-Belser-Str. 79a
70597 Stuttgart
Tel.: 0711 / 6781-407 Fax:0711 / 6876902
e-mail:hann@psyres-stuttgart.de
*********************************************
# original question from Aug. 17. #
Dear S+ users,
this question, regarding estimates for nodes of tree(or rpart) objects,
is somewhat beyond S+, but may be of interest to other members of the
group as well.
When crossvalidating a classification tree, the crossvalidation results
(for instance, number of misclassifications) refer only to the tree as a whole.
How do I obtain these results for every node of the tree? CV-estimates
for the pruning sequence often prune back whole branches instead of single
nodes.
Could these serve as estimates for the true misclassification
of this node and how useful would this estimation be?
Another approach might be, to reconstruct the tree and bootstrap the
proportions of each class for every node and leaf. With very small nodes
or leafs, the results of course approach nonsense, but does this make
sense for large to moderately large nodes?
Any suggestions and comments are appreciated and of course I will
summarize answers to the list.
Thanks in advance and all the best.
wolfgang
*********************************************
Wolfgang Hannoever
Forschungsstelle fuer Psychotherapie
Christian-Belser-Str. 79a
70597 Stuttgart
Tel.: 0711 / 6781-407 Fax:0711 / 6876902
e-mail:hann@psyres-stuttgart.de
*********************************************
-----------------------------------------------------------------------
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
