CMU Artificial Intelligence Repository
ToolDiag: Feature selection software for improving
classifiers.
areas/neural/systems/tooldiag/
ToolDiag is a feature selection program that increases the accuracy of
classifiers and reduces their complexity by providing them with a
subset containing only the most relevant features. It has interfaces
to LVQ_PAK and SNNS, and uses a data file format that is compatible
with that of LVQ_PAK. The 2-d graphics can be displayed using the
GNUPLOT plotting package. ToolDiag implements many concepts from
Devijver and Kittler's book "Pattern Recognition -- A Statistical
Approach" (Prentice Hall, 1982), including the optimal branch and
bound search strategy, together with several different selection
criteria. ToolDiag can also perform an error estimation using the
leave-one-out method and a K-nearest-neighbor classifier. It also
includes a learning module (Q*) that has the same functionality as
LVQ. ToolDiag cannot handle missing values and requires continuous or
ordered discrete numerical features.
Origin:
ftp.fct.unl.pt:/pub/di/packages/tooldiag-1.4.tar.Z
Version: 1.4.1 (3-DEC-93)
Requires: C
Ports: Test on IBM, DEC, NeXT, Sun, and DOS.
Copying: Copyright (C) 1992, 1993 Thomas W. Rauber
CD-ROM: Prime Time Freeware for AI, Issue 1-1
Author(s): Thomas Rauber
Universidade Nova de Lisboa
2825 Monte Caparica
PORTUGAL
Tel: (+351) (1) 295-7787
Fax: (+351) (1) 295-7786
Keywords:
Authors!Rauber, Branch and Bound Search, C!Code,
Classification, Feature Selection, K Nearest Neighbor,
Leave One Out, Machine Learning!Neural Networks,
Neural Networks!Classification, ToolDiag
References: ?
Last Web update on Mon Feb 13 10:25:38 1995
AI.Repository@cs.cmu.edu