RETIS - a Machine Learning System
RETIS is a system for automatic induction
of regression trees.
Regression trees are used to model a piecewise-linear relation between
discrete or continuous attributes and continuous class.
RETIS is a system for automatic knowledge acquisition from a given set
of examples. The knowledge, induced from examples, is represented
in a form of a regression tree. RETIS stands for
REgression
Tree Induction System. Besides creating regression
trees, RETIS also enables post-pruning, interpretation, printing,
plotting and
testing of the trees and classification of new examples by the induced
regression tree.
When learning regression trees with RETIS, examples have to be described
with a set of attributes. Each attribute has its possible set of values.
There are two kinds of attributes:
- continuous attributes - their value can be any real number,
- discrete attributes - they can have a value from some
predefined (finite, usually small) set of values.
Each example has also associated class value, which is
continuous and represents the quantity we want to learn.
So, RETIS actually learns a function
y(x1,..,xn) which
approximates the relationship between the values of the attributes and
the value of the class.
Each internal node of a regression tree contains a test on a value of an
attribute. According to the result of the test, interpretation of the tree
proceeds to the left or to the right subtree of the node. A leaf prescribes
a value to a function, approximated by the regression tree.
Further Reading
- L.Breiman, J.H.Friedman, R.A.Olshen, C.J.Stone, Classification and
Regression Trees, Wadsworth Int. Group, Belmont, California, 1984.
- Aram Karalic.
Employing linear regression in regression tree leaves.
In Proceedings of ECAI'92 (European Congress on Artificial
Intelligence), Vienna, Austria, August 1992.
- Aram Karalic.
Linear regression in regression tree leaves.
In Proceedings of ISSEK'92 (International School for Synthesis
of Expert Knowledge) Workshop, Bled, Slovenia, 1992.
- Aram Karalic and Bojan Cestnik.
The bayesian approach to tree-structured regression.
In Proceedings of ITI-91, Cavtat, Yugoslavia, 1991.
Estimation of the surface roughness of the workpiece in the steel
grinding process.
Described in:
Bogdan Filipic, Mihael Junkar, Ivan Bratko, and Aram Karalic.
An application of machine learning to a metal-working
process. In Proceedings of ITI-91, Cavtat, Yugoslavia, 1991.
Modelling of drug characteristics from its molecular properties.
Described in:
Dunja Mladenic and Aram Karalic.
Drug design by machine learning:
Modelling drug activity.
In Proceedings of ISSEK'92 (International School for
Synthesis of Expert Knowledge) Workshop, Bled, Slovenia, 1992.
Prediction of quality properties of polyester/cotton yarns from the
characteristics of its components.
Described in:
Zoran Stjepanovic, Anton Jezernik, M. Nikolic, and Aram Karalic.
A knowledge-based system for prediction of quality properties of
polyester/cotton yarns.
Fibres & Textiles in Eastern Europe, 1(1), 1993.
Classification of sub-atomic particles.
Described in:
Irma Sutlic, Aram Karalic, and Igor Mandic.
Machine learning in particle physics.
In Proceedings of the Second Electrotechnical and Computer
Science Conference ERK'93, Portoroz, Slovenia, 1993.
In Slovene.
RETIS is available as a commercial software. Please
download User's manual and free demo version (below).
For further information please contact Aram Karalic at
aram.karalic@ijs.si.
RETIS requires IBM PC compatible personal computer with at least 80286 CPU.
Download by selecting file link while holding down the SHIFT key.
Aram