Download BinClass: A Software Package for Classifying Binary Vectors User's

Transcript
4 Using BinClass
The BinClass software is designed to be machine independent, i.e. portable.
We have written the program to conform with the ANSI-C standard as far as
possible. Standards do not cover visual programming. Macintosh, Windows
and X-Windows are all totally dierent in architecture. Thus, we did not
develop a graphical user interface (GUI) for BinClass. BinClass is used via
the command prompt, which is usually available in all systems. The general
format of BinClass invocation is as follows:
binclass command switches le base
Most of the switches are optional and have default values.
The following call gives a list of available commands:
binclass help
and the following gives brief lists of available switches for the specic
commands.
binclass command
4.1 Classication methods
4.1.1 SC minimizer (GLA and Local Search)
The classication tool is the most complex part of the software. It uses
AMOLS 1 [27] and GLA 2 [32, 17] algorithms to minimize the stochastic
complexity [21].
GLA assigns vectors of the input set to their classes in the classication
by a nearest neighbor rule (distance measure, error measure). There are
six alternatives implemented in BinClass that can be chosen with the command line switch -f. These are: Hamming-distance to HMO, L1-distance
to centroid, L2-distance to centroid, Shannon-codelength, L1-distance with
Shannon-codelength and L2-distance with Shannon-codelength. The two last
variants work by running rst iterations with L1 or L2 and by nishing with
the Shannon-codelength. This combined version works best in most cases
because some numerical problems are avoided. Simple methods (Hammingdistance, L1-distance) are included for demonstration purposes.
1
2
Adaptive Multi-Operator Local Search
Generalized Lloyd Algorithm also known as GBL = Gray, Buzo, Linde and k-means
31