From uhog.mit.edu!news.mathworks.com!newsfeed.internetmci.com!usenet.eel.ufl.edu!psgrain!nntp.teleport.com!usenet 4 Jan 1996 00:51:06 GMT
Path: uhog.mit.edu!news.mathworks.com!newsfeed.internetmci.com!usenet.eel.ufl.edu!psgrain!nntp.teleport.com!usenet
From: fawcett@nynexst.com (Tom Fawcett)
Newsgroups: comp.lang.perl.announce,comp.lang.perl.misc
Subject: New module:  Statistics::LTU  - Implements linear threshold units
Followup-To: comp.lang.perl.misc
Date: 4 Jan 1996 00:51:06 GMT
Organization: The practical transmission of speech
Lines: 30
Approved: merlyn@stonehenge.com (comp.lang.perl.announce)
Message-ID: <4cf89q$o3h@maureen.teleport.com>
Reply-To: fawcett@nynexst.com
NNTP-Posting-Host: linda.teleport.com
X-Disclaimer: The "Approved" header verifies header information for article transmission and does not imply approval of content.
Xref: uhog.mit.edu comp.lang.perl.announce:47 comp.lang.perl.misc:5933


I have uploaded to CPAN a new module, Statistics::LTU, which defines
methods for managing Linear Threshold Units.  A linear threshold unit is
basically a 1-layer neural network, also called a perceptron.

LTU's are used to learn classifications from examples.  Given a set of
examples from two classes, the LTU will learn to distinguish one class from
the other by adjusting weights associated with the features.  After
training on a number of examples, the LTU can then be used to classify new
(unseen) examples.

The Statistics::LTU module defines an uninstantiable base class (LTU) and
four instantiable classes built on top of it.  The four individual classes
differ in the training rule used.  I have implemented the absolute
correction rule (ACR), Mark Frean's thermal absolute correction rule
(TACR), least-mean squares (LMS) and Young's recursive least-mean squares
(RLS).

The module is object-oriented and requires Perl 5.
Here is the line for the module list:

Statistics::
::LTU		RdpO	Implements Linear Threshold Units	TOMFA

(I have put this under Statistics rather than Math because the latter seems
to contain mostly data structures.)

-Tom
fawcett@nynexst.com