MIME-Version: 1.0 Server: CERN/3.0 Date: Tuesday, 07-Jan-97 15:55:35 GMT Content-Type: text/html Content-Length: 1802 Last-Modified: Monday, 11-Dec-95 16:42:34 GMT Machine Learning Research Software

Machine Learning Research Software


We have compiled a group of Common Lisp files for various inductive classification algorithms. These algorithms are intended for research purposes and all use the same basic data format and interface. Also included is automatic testing software for running learning curves that compare multiple systems and utilities for plotting and statistically evaluating the results.

This software is all available via anonymous ftp.

Current Algorithms:

  1. AQ - An early DNF learner.
  2. Backprop - The standard multi-layer neural-net learning method.
  3. Bayes Indp - A simple naive or "idiot's" Bayesian classifier.
  4. Cobweb - A probabilistic clustering system.
  5. FOIL - A first-order Horn-clause learner (Prolog and Lisp versions).
  6. ID3 - A decision tree learner with a number of features.
  7. KNN - A k nearest neighbor (instance-based) algorithm.
  8. Perceptron - An early one-layer neural-net algorithm.
  9. PFOIL - A propositional version of FOIL for learning DNF.
  10. PFOIL-CNF - A propositional version of FOIL for learning CNF.
Some sample data sets included are "dna-standard.lisp" and "labor-neg.lisp". The file "data-utilties.lisp" should be loaded before any other code. Comments at the beginning of "universal-tester.lisp" help define the data format and interface standards used. The file "data-utilities.lisp" also includes a function for converting a data file suitable for Quinlan's C4.5 to a format usable by these algorithms.


estlin@cs.utexas.edu