<mod-ki 0778> FG 1.1.3 Maschinelles Lernen - Prolog Algorithmen
Hans-Werner Hein
To: mod...@unido.informatik.uni-dortmund.de
Date: Mon, 21 Oct 91 21:03:23 +0200
Subject: <mod-ki 0778> FG 1.1.3 Maschinelles Lernen - Prolog Algorithmen
Keywords: Software, Prolog, Machine Learning
mod-ki message 0778
************************************************************************
Gesellschaft fuer Informatik
Fachgruppe 1.1.3
Special Interest Group on 'Maschine Learing'
Prolog library of machine learning algorithms
Thomas Hoppe
Projektgruppe KIT
Technische Universitaet Berlin
Franklinstr. 28/29
1000 Berlin 10
Germany
The following algorithms are currently available from the library
(last up-date 21.10.91):
1. Learning Algorithms
AQ1
aq1.pro - Jeffrey M. Becker's AQ-PROLOG, a reimplementation
of Michalski's AQ for attribute vectors
aq1_1.pro - Simple Example for AQ
aq1_2.pro - Extension of AQ_1.PRO
ARCH1
arch1.pro - Stefan Wrobel's implementation of Winston's incre-
mental learning procedure for structural descrip-
tions
arch1_1.pro - Winston's example archs
ARCH2
arch2.pro - Ivan Bratko's minimal implementation of Winston's
incremental learning procedure for structural
descriptions
arch2_1.pro - simple example archs
ATTDSC
attdsc.pro - Ivan Bratko's simple algorithm for learning attri-
butional descriptions
attdsc_1.pro - Small example set for learning to recognize
objects from their silhouettes
COBWEB
cobweb.pro - Joerg-Uwe Kietz's implementation of Fisher's
incremental concept formation algorithm
cobweb_1.pro - a simple data set describing some hotels
DISCR
discr.pro - Pavel Brazdil's generation of discriminants from
derivation trees
discr_1.pro - Simple abstract example
discr_2.pro - Abstract example generating useful and not useful
discriminants
MULTAGNT
multagnt.pro - Yiu Cheung HO's implementation of Brazdil's tutor-
ing setting
teacher.pro - Teacher's knowledge base
learner1.pro - A correct Learner's knowledge base
learner2.pro - An erroneous Learner's knowledge base
calls_1.pro - Example calls concerning correct knowledge
calls_2.pro - Example calls concerning wrong knowledge
ID3
idt.pro - Luis Torgo's ID3-like program for induction of
decision trees based on the gain-ratio measure
idt_1.pro - Abstract example
idt_2.pro - Credit assignment example producing trivial clas-
sification
idt_3.pro - Credit assignment example producing non-trivial
classification
INVERS
invers.pro - Implementation of Steven Muggleton's absorption
and intra-construction operators for inverse reso-
lution based on the representation change idea of
Celine Rouveirol and Jean-Francois Puget
invers_1.pro - Example call's
LOGIC
logic.pro - Steven Muggleton's implementations of substitution
matching, Plotkin's term generalisations and Wray
Buntine's generalized subsumption
logic_1.pro - Example call's
EBG
ebg.pro - Basic algorithms for explanation based generalisa-
tion and partial evaluation based on Kedar-Cabelli
& McCarty's idea. Different kinds of simple PROLOG
meta-interpreters.
ebg_1.pro - Suicide example for EBG
ebg_2.pro - Safe_to_stack example for EBG
VS
vs.pro - Luc de Raeth's reimplementation of Mitchell's
version space strategy
vs_1.pro - A simple shape and color taxonomy
2. Updates
COBWEB
cobweb.pro - Joerg-Uwe Kietz has detected a fault in the
current implementation. In the near future a
corrected version will be available.
cobweb_1.pro - Will be substituted by a smaller and better
understandable example.
3. In preparation
MIS - Shapiro's Modell Inference System
AQ2 - Stefan Wrobel's reimplementation of Michalski's AQ
for structural descriptions
CLUSTER - Michalski and Stepp's Conceptual Clustering
HEBBSIM - Duwe & Grothaus connectionist association matrix
SPROUTER - Hayes Roth's Generalisation Algorithm
4. Ordering of files
Files can be ordered either via surface or electronic mail, or
accessed via anonymous ftp.
Surface Mail - Send a short notice to the address given at top of
this document. Files will be distributed via MS-
DOS formated 3.5 inch floppy, which should be send
with your request.
E-Mail - Send a short notice to the address given at top of
this document.
FTP - Files are accessible via anonymous FTP from
dione.rz.uni-osnabrueck.de (131.173.128.15)
userid: FTP or ANONYMOUS
password: your own userid
directory: PUB/UNIX/MLP
5. Documentation notes
All algorithms are written in Prolog with Edinburgh-syntax. Most of
the time we use only Clocksin/Mellish or DEC-10 build-in predicats.
However, sometimes it can happen that predicates with well-known
meaning are not build-in w.r.t. Clocksin/Mellish or DEC-10 (e.g.
append, member, bagof). Because they are so common, we rely that your
local Prolog system has access to their definitions. If you have
trouble with some paticular predicates in an algorithm send an e-mail
to the address given at top of this document. Either I will send you
the predicate definition I use or I will fix the problem in the source
code as soon as possible.
If a particular algorithm only consists out of some predicats I try to
document every predicat (e.g. EBG.PRO), otherwise I only document the
necessary predicats for running the algorithm and/or the special
changes I had to introduce (e.g. AQ1.PRO).
The sign '+' in front of an argument in a call means that the
argument, is an input argument to the function, which must be
instantiated. A sign '-' in front of an argument, means that this is
an output argument, which should not be instantiated. No sign in front
of an argument indicates that this argument works in both ways.
The property 'backtrackable' indicates that the predicate is re-
instantiated on backtracking. The property 'symmetric' means that this
predicat can also called with instantiated output arguments, deliver-
ing the appropriate input arguments.
6. Notes on bug detection
The algorithms are more or less tested. Sometimes bugs can occure
dependent on our more or less extensive testing. Because different
Prolog dialects are mixed up on my micro testing is a difficult task
for me. If you find a 'new feature' which you didn't expected, let it
me know, so that others can benefit from your experience. Send bug
reports to the address given at top of this document.
7. Notes on software delivery
Because any library has to rely on authors, we encourage everybody who
has implemented a machine learning algorithm to partitizipate in the
extension of the library.
Software should be send on 3.5 inch floppies formated either for Atari
ST or MSDOS 3.X, or by electronic mail to the address given at top of
this document.
For the purpose of a uniform documentation and easy maintainance,
please include a note with references about the algorithm and if
possible document the top-level calls to your program with the
following form:
/**********************************************************/
/* */
/* call : predicate (+ARG1,-ARG2) */
/* */
/* arguments : ARG1 = is of type xyz */
/* ARG2 = is something different in the */
/* following form */
/* */
/* side effects: which ever side effects the predicate */
/* produces */
/* */
/* properties : backtrackable, or whatever outstanding */
/* properties the predicate has */
/* */
/* references : actual reference of the algorithm */
/* */
/**********************************************************/
/* A short description of what the predicate does and how */
/* the input and output can be interpreted. */
/**********************************************************/
[Solche messages verschicke ich am liebsten :-) hwh]
************************************************************************