Newsgroups: comp.ai.neural-nets
From: jimmy@ecowar.demon.co.uk (Jimmy Shadbolt)
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!uunet!in1.uu.net!pipex!peernews.demon.co.uk!ecowar.demon.co.uk!jimmy
Subject: Re: bootstrapping 
Distribution: world
References: <3it3aa$h17@homer.alpha.net>
Organization: Econostat
Reply-To: jimmy@ecowar.demon.co.uk
X-Newsreader: Simple NEWS 1.90 (ka9q DIS 1.21)
Lines: 21
X-Posting-Host: ecowar.demon.co.uk
Date: Tue, 28 Feb 1995 12:49:11 +0000
Message-ID: <793975751snz@ecowar.demon.co.uk>
Sender: usenet@demon.co.uk

In article <3it3aa$h17@homer.alpha.net> nuwmfj5@tmgic7.mgic.com writes:

>A prominent characteristic of neural nets we have yet to see addressed
>in this forum is the effect of data splitting (into training, testing,
>and possibly cross-validation sets) on model results.  In a working 
...
>Informally, we see the same thing in our cross-sectional analyses.
...
        My research in econometrics suggests something even worse:
        variable *ordering* is important (even for back-to-basics
        linear models). 

>This raises the question of what to do about the variation in model 
>performance due to different data splits.  

        Does is mean that Bayesian framework cannot be avoided?

        Cheers

        Drago

