lukose@cm.deakin.oz.au (Dickson Lukose) (06/13/91)
Dear Colleague,
There has been only two positive reply to my request on public
domain ID3 software, as listed below:-
====================11111================11111================11111====
From: moskowit@paul.rutgers.edu (Len Moskowitz)
Subject: Re: Public Domain ID3 source code
I'm appending a net post I had filed away. Hope it helps.
Len Moskowitz
----
>From jdooley@neutrino.urbana.mcd.mot.com (John Dooley) Fri Oct 26 14:00:39 1990
Path: paul.rutgers.edu!aramis.rutgers.edu!rutgers!cs.utexas.edu!asuvax!mcdphx!udc!neutrino.urbana.mcd.mot.com!jdooley
From: jdooley@neutrino.urbana.mcd.mot.com (John Dooley)
Newsgroups: comp.ai
Subject: Re: ID3 and CN2 Source Code
Message-ID: <1421@urbana.mcd.mot.com>
Date: 26 Oct 90 18:00:39 GMT
Sender: netnews@urbana.mcd.mot.com
Reply-To: jdooley@urbana.mcd.mot.com
Organization: Motorola, Inc. Urbana Design Center
Lines: 18
There are Common LISP versions of ID3 and ID5 (an incremental ID3) available
from Dr. Dan StClair at the University of Missouri Graduate Engineering Center
in St. Louis, MO. (c0567@umrvmb.umr.edu). These LISP versions have been tested
on LISP Machines, Vaxes, Sun-4, Mac II (running Allegro), and TI Explorers and
work fine on all. There is also a C language version of ID3.
A versionof CN2 in C was in the works this past summer, but I don't know
if it's available yet. CN2 is another non-incremental inductive learning
algorithm which produces rules instead of decision trees like ID3 does.
=======22222========22222=========222222===========22222==========22222==
Subject: Re: Public Domain ID3 source code
From: spratt@hawk.cs.ukans.edu
I wrote the following program in Franz Lisp for a course on knowledge
acquisition. I also have it in XLISP (the code is essentially identical).
;;; ID3 by Lindsey Spratt (spratt@hawk.cs.ukans.edu).
;;; Written circa 1990 at the University of Kansas.
;;;
;;; The main function is id3-invoke, called with no arguments. This calls
;;; id3 in a loop until the user says she is no longer interested in running
;;; id3.
;;;
;;; [Because this implementation of ID3 can be very slow, various messages
;;; are printed out during execution to comfort the user that progress is
;;; being made. For the second test file, ID3.in2, this program takes over
;;; 2 hours of real execution time.]
;;; The implementation has two major parts, one for building a tree given a
;;; set of examples and the other for determining the decision given an
;;; example and a tree.
;;;
;;; The main function for building a tree is find-tree. If the examples all
;;; have the same decision, then its job is trivially done. Otherwise, the
;;; attribute with the highest information gain (lowest conditional entropy)
;;; with respect to the decision is found and made the root of the tree.
;;; In the process, the examples are classified according to values in the
;;; chosen attribute. Each of these classifications is further analyzed to
;;; build the child subtrees for the new root.
################### REMAINING PARTS NOT INCLUDED ###########################
For all those who have requested information from me, I have forwarded
your email address to spratt@hawk.cs.ukans.edu for direct reply.
Once again, thank to all who have responded to my request for information.
regards
lukose@deakin.oz.au