Grapical Models — Methods for Data Analysis and Mining
The links listed in this section refer to the corresponding sections
on Christian Borgelt's Software
Page. The programs are general in the sense that they work for
arbitrary datasets, not only those used in the book.
The programs listed in this section are not general purpose, but
have been written for specific examples in the book.
- Expectation Maximization
An expectation maximization example program, which was used to
do the computations for the example discussed in Section 5.3
(pages 137-142). The computations for this example can be followed
by calling the program with "em 0" (standard expectation
maximization) or "em -a1 0" (expectation maximization with
a momentum term). The value of epsilon (limit for termination) can
be specified with "-e<value>". Calling the
program without any arguments produces a list of other options.
em.c (23 kb)
(source file, version 1.2, 05.01.2002)
em
(Linux executable, 23 kb)
em.exe
(Windows console executable, 60 kb)
- Kullback-Leibler Information Divergence
A program to compute the Kullback-Leibler information divergence
as well as the log-likelihood of a database for the geometrical
objects example (Figure 7.5, Section 7.1.2, page 173).
kldiv.c (8 kb)
(source file, version 1.1, 08.01.2002)
kldiv
(Linux executable, 18 kb)
kldiv.exe
(Windows console executable, 60 kb)
- Specificity Divergence
A program to compute the specificity divergence as well as
the weighted sum of possibility degrees of a database for
the geometrical objects example (Figure 7.8, Section 7.1.3,
page 180).
spcdiv.c (10 kb)
(source file, version 1.1, 08.01.2002)
spcdiv
(Linux executable, 21 kb)
spcdiv.exe
(Windows console executable, 64 kb)
With the shell scripts listed in this section the experimental
results reported in the book can be checked.
- Computing Maximum Projections
This shell script generates Table 5.7 (Section 5.4.4, page 149),
which contains the number of tuples in the support and the closure
of three databases. Note that in order to execute this script the
data table utilities referred to above (general
programs) as well as the datasets djc,
soybean, and vote made available below
(data files) must also be retrieved.
proj.sh (Bourne shell script, 1kb)
proj.zip (archive with Bourne shell
script and necessary data files; 13kb)
- Bayes Classifiers for the Iris Data
This shell script generates the naive and full Bayes classifiers
referred to in Section 6.1.3 (pages 155-156). Note that in order
to execute this script the data table utilities and the Bayes
classifier programs referred to above (general
programs) as well as the dataset iris2d made available
below (data files) must also be retrieved.
iris.sh (Bourne shell script, 1kb)
iris.zip (archive with Bourne shell
script and necessary data file, 2kb)
Colored versions of the two diagrams of Figure 6.2 (Section 6.1.3,
page 156) are available here. They were
generated from the induced classifiers with the Bayes classifier
visualization program listed
here.
- Naive Classifier Induction
This shell script generates Table 6.2 (Section 6.4, page 159),
which contains the evaluation of different types of classifiers
for for datasets. Note that in order to execute this script the
data table utilities, the Bayes classifier programs, and the
decision tree programs referred to above
(general programs) as well as the
datasets audio, horse, soybean,
and vote0/vote1 made available below
(data files) must also be retrieved.
Note also that some of the error percentages in the table
produced by this script can differ slightly from those reported
in the book due to rounding problems.
nclass.sh (Bourne shell script, 5kb)
nclass.zip (archive with Bourne shell
script and necessary data files, 30kb)
- Learning Graphical Models with INES
These shell scripts generate Table 7.4 (Section 7.4.1, page 248),
Table 7.5 (Section 7.4.2, page 251), and Table 8.1 (Section 8.3,
page 259). Note that in order to execute these scripts the data
table utilities and the ines program referred to above
(general programs) as well as the dataset
djc made available below (data files)
must also be retrieved.
Note also that the results for the BDeu metric as well as for
simulated annealing learning of possibilistic networks differ
from those reported in the book due to program bugs in an earlier
version of the INES program. The results of local structure
learning differ due to a complete reimplementation of this part
of the INES program, which was done after the publication of
the book.
djc_prob
(Bourne shell script for probabilistic network learning, 7kb)
djc_poss
(Bourne shell script for possibilistic network learning, 6kb)
djc_local
(Bourne shell script for probabilistic network learning with
local structure, 5kb)
djc.zip
(archive with Bourne shell scripts and dataset, 14kb)
The datasets listed here (except the geometric objects dataset)
can also be found either at the
UCI Machine Learning Repository or at the
Bayesian Network Repository,
where additional information about these datasets is available, too.
The versions listed here have been adapted to meet the default input
format of the programs referred to above.
Last updated:
Mon Aug 28 16:10:20 CEST 2006
- christian@borgelt.net
