Newsgroups: comp.ai.neural-nets
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!news.sprintlink.net!redstone.interpath.net!sas!mozart.unx.sas.com!saswss
From: saswss@hotellng.unx.sas.com (Warren Sarle)
Subject: Re: *Where did hid. formula come from?
Originator: saswss@hotellng.unx.sas.com
Sender: news@unx.sas.com (Noter of Newsworthy Events)
Message-ID: <D4q1x7.JH0@unx.sas.com>
Date: Tue, 28 Feb 1995 17:58:19 GMT
X-Nntp-Posting-Host: hotellng.unx.sas.com
References: <gjHot5i00iV6I1DF4a@andrew.cmu.edu>  <Pine.SOL.3.91.950227052811.20654B-100000@bingsun1> <3it4td$qa5@cantaloupe.srv.cs.cmu.edu>
Organization: SAS Institute Inc.
Lines: 37


In article <3it4td$qa5@cantaloupe.srv.cs.cmu.edu>, sef@CS.CMU.EDU (Scott Fahlman) writes:
|>
|> In article <Pine.SOL.3.91.950227052811.20654B-100000@bingsun1> Scott Hackett <br00372@bingsuns.cc.binghamton.edu> writes:
|>
|>    There is another equation for the number if first layer and second layer
|>    hidden neurons in Timothy Masters' book"Neural Network Recipes in C++".
|>    His equation appears on page 177.
|>
|>    r = cube root(n/m)
|>
|>    num hidden on first layer = m * r^2
|>    num hidden on second layer = m * r
|>
|>    n = number of input neurons
|>    m = number of output neurons
|>
|> Actually, instead of the cube root, I think you should take (n/m) to
|> the power of 1/Pi.  The theory behind this is too complex to explain
|> here, since it involves circular cropping of the data and the
|> influence of higher powers in the expansion.  Suffice it to say that
|> this formula works better for all the real-world test cases I have
|> tried.

There's a typo in Scott's post: that should be i/Pi where i=sqrt(-1).

Also note the section on the next page (178) in Masters' book starting,
"The best approach to finding the optimal number of hidden neurons
is time-consuming, but should always be followed for important tasks."

|> P.S.  :-)

-- 

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.
