Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.3 4.3bsd-beta 6/6/85; site vu-vlsi.UUCP
Path: utzoo!watmath!clyde!cbosgd!ukma!psuvm.bitnet!psuvax1!vu-vlsi!perry
From: perry@vu-vlsi.UUCP (Rick Perry)
Newsgroups: net.math,net.micro,net.micro.pc
Subject: Re: simplex algorithsm for curve fitting---any disadvantage ??
Message-ID: <236@vu-vlsi.UUCP>
Date: Fri, 14-Feb-86 01:04:29 EST
Article-I.D.: vu-vlsi.236
Posted: Fri Feb 14 01:04:29 1986
Date-Received: Sun, 16-Feb-86 04:23:05 EST
References: <1217@princeton.UUCP>
Distribution: net
Organization: Villanova Univ. EE Dept.
Lines: 20
Xref: watmath net.math:2843 net.micro:13799 net.micro.pc:6978
Summary: Nelder-Mead simplex is pretty good

> I recently ported the curve fitting program, listed on BYTE 1984 May p.340,
> The authors claimed that the program could fit *any* equation with *any*
> number of parameters and variables to experimental data.

  I've implemented that (or similiar) simplex method in both APL and Pascal
and have great success with it.  The Simplex algorithm is very robust and
stable if programmed properly (taking proper account of machine precision,
roundoff's and convergence criterion).  It has some advantages over other
methods in that it tends to avoid local minima if started over a big enough
region.  But like any global minimization algorithm, there is always some
chance that it could converge to a local minima.

   The only disadvantage that I would note is that it has more or less linear
convergence, so it may take more cpu time to get the answers with this method.
But I like it for its stability and easy understandability - that is, you
don't need 10 Ph.D.'s in math to comprehend how it works, and it's fairly
easy to program in a straightforward, bug-free (if there is such a thing!)
way.

...Rick         ...{pyrnj,psuvax1}!vu-vlsi!perry