This is version 2.5 of Statistics::LTU, a module for Linear Threshold Units.
A linear threshold unit is a 1-layer neural network, also called a
perceptron. LTU's are used to learn classifications from examples. An LTU
learns to distinguish between two classes based on the data given to it.
After training on a number of examples, the LTU can then be used to
classify new (unseen) examples.
LTU.pm defines an (uninstantiable) base class, LTU, and four other
instantiable classes built on top of LTU. Each of the four classes uses a
different training method: ACR (absolute correction rule), TACR (a thermal
annealing version of the absolute correction rule), LMS (least-mean squares
fit) and RLS (recursive least-mean squares rule). Check out ltu.doc for
further information on these. You can use LTUs without understanding
exactly how they work.
REQUIREMENTS
Statistics::LTU needs Perl version 5, as it is object-oriented and uses
references extensively.
INSTALLATION
Run the following:
perl Makefile.PL
make
make install
LTU.pm has some tests at the end; try running "perl LTU.pm". Note that
this creates four LTU files with ".saved" extensions, which can be deleted
after the tests.
Note: Depending on the version of MakeMakefile you have, you may get an
error with ld when you run make. I don't know how to prevent this. The
Makefile is just used to copy LTU.pm, LTU.pod and weather.pl into the
Statistics subdirectory of your Perl library directory.
Then you probably want to:
make clean
FILES
ltu.doc has some useful information on LTUs, though it was written for the
C version.
weather.perl is a simple demo showing how examples are created, and how
LTUs are trained and tested. It should be more instructive than
LTU.man. Run "perl weather.perl | more".
The code itself is heavily documented.
AUTHOR / SUGGESTIONS / BUG MAGNET
Tom Fawcett (fawcett@nynexst.com).
LTU.pm is based on LTU.C, an implementation of LTUs written by James Callan
at the University of Massachusetts. I've tried to Perlify and objectify
the code completely, but some awkwardness remains.