		MDMC

		Home
		Consultancies
		Education
		Research
		Facilities
		Software
			CaMML
			DTree
			Snob
			Random Number Generators
		People
		Contacts
		Bibliography
		Seminars
		Member Login

MDMC Software - DTree & DGraf

Supervised learning (classification) using MML methods. DTree and DGraf attempt to learn the best decision trees (or decision graphs) to predict the class of each "thing". They allow fast automatic categorization, but must be trained. (For unsupervised classification, use Snob.)

Both DTree and DGraf allow the target class to be either discrete (multi-state) or continuous. DGraf has several other benefits.

DTree

Authors: Chris Wallace
Other contributors: Sarah George
License: Academic License
Documentation: [ Quickstart | DTree documentation ]

Download: dtree.tgz
File Size: (143405 bytes)
(Last updated: Thursday, 29 March 2001)

The archive has the FORTRAN source code, user documentation, and source-code documentation in the style of JavaDoc. It compiles under g77.

Use of this program constitutes agreement to the Academic license.

DGraf

Authors: Chris Wallace
License: Academic License
Documentation: [ See DTree documentation for now. ]

Download: dgraf.tgz
File Size: (467391 bytes)
(Last updated: Friday, 7 April 2004)

A decision graph is a decision tree that allows joins. It has the same expressive power, but is statistically more powerful because wherever it uses a join, it has twice as much data from which to estimate parameters. DGraf implements decision graphs in MML.

Decision graphs were invented and first implemented by Jon Oliver. This version was implemented by Chris Wallace in C. In addition to joins, Chris has allowed leaf nodes to have Naive Bayes estimators.

The archive file has Chris' C source code, as well as some version of the earlier FORTRAN source, and apparently Jon Oliver's source. It does not appear to have any documentation.

Decision Graphs with Multi-Way Joins etc.

Oliver-style decision graphs only allow binary joins. Peter Tan and David Dowe have generalized them to allow multi-way joins. This results in a more efficient coding of the models and provides better results. In addition, they have also implemented dynamic attributes, which give further gains.

Tan & Dowe 2002. MML Inference of Decision Graphs with Multi-Way Joins.

Tan & Dowe 2003. MML Inference of Decision Graphs with Multi-Way Joins and Dynamic Attributes

For now the code is available only from the authors.