Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!pipex!lyra.csx.cam.ac.uk!sunsite.doc.ic.ac.uk!yama.mcc.ac.uk!cf-cm!C.M.Sully
From: C.M.Sully@cm.cf.ac.uk (Chris Sully)
Subject: Re: bias
Message-ID: <1995May23.092011.19036@cm.cf.ac.uk>
Sender: C.M.Sully@cm.cf.ac.uk (Chris Sully)
Nntp-Posting-Host: aquamarine.cm.cf.ac.uk
Organization: University of Wales College of Cardiff, Cardiff, WALES, UK.
References: <1995May11.113124.15048@cm.cf.ac.uk> <D8G3DD.53z@unx.sas.com>
Date: Tue, 23 May 1995 09:20:09 +0000
Lines: 29

In article <D8G3DD.53z@unx.sas.com>, saswss@hotellng.unx.sas.com (Warren Sarle) writes:
|> In article <1995May11.113124.15048@cm.cf.ac.uk>, C.M.Sully@cm.cf.ac.uk (Chris Sully) writes:
|> |> Following on from the recent discussion: I tried a linear activation
|> |> function for my output unit. This resulted in a 8 fold increase in bias in the residuals when compared to the results with a logistic activation function.
|> |>
|> |> Any thoughts?
|> 
|> It is difficult to say anything constructive with so little information.
|> If you would post your data and a precise description of the network
|> architecture, then we might be able to say something useful.
|> 
|> -- 
|> 
|> Warren S. Sarle       SAS Institute Inc.   The opinions expressed here
|> saswss@unx.sas.com    SAS Campus Drive     are mine and not necessarily
|> (919) 677-8000        Cary, NC 27513, USA  those of SAS Institute.

Apologies, but I assumed people had been following the thread and had read the earlier posts. I guess a little presumptious on memory grounds given the 
volume of valuable contributions (to me at least:)) Warren makes to this newsgroup. A summary:

MLP. 11-6-1. Prediction of birthweight of a baby from 11 input vars. 2000 records from c.25000 (so I don't have to wait a week to train one network
on a Sun) split 1000:500:500 training, validation, testing. I'm using PlaNet
with default activation function for all units (logistic) except for the output unit which I changed to linear on received advice (from Warren amongst others).

Any other info required?

Cheers.

Chris.
