Machine Learning Tools

People

RJ Honicky

Numpy, Numeric Python and Fortran

In the most recent release of Maemo, there is no FORTRAN compiler, due to a switch from the ARM to ARMEL architecture between os2007 and os2008. Since Debian does not currently support FORTRAN on ARMEL, neither does Maemo. However, recent unstable releases of Debian do have a FORTRAN compiler (gfortran). Therefore, we should be able to use a FORTRAN compiler with some minimal package work.

Numpy, or NUMerical PYthon, wraps several FORTRAN libraries for matrix manipulation and numerical computation. Unfortunately, some of these libraries need to be tweeked to each fortran compiler, so there we can't just compile the fortran into C and then compile with GCC. Thus, to get the whole Numpy suite running, gfortran will need to be working.

There is, however, an older version of Numpy, called Numeric, available as 'python-numeric', which can be fetched with

apt-get install python-numeric

also, we need python 2.5, which can be installed with

apt-get install python2.5

libsvm

Support Vector Macines

Support Vector Machines (SVM) are a common tool in statistical machine learning for numerically efficient classification of vectors into discreet classes, and doing regression (SVR).

LibSVM on the Phone

I have built and and packaged libsvm, a common and robust tool with python bindings. It is available under the class repository. You can get it by doing:

apt-get install libsvm-python

after adding the class repository to your sources.list file:

http://phone.cs.berkeley.edu/apt

Training your model

To train and develop your model, download the package at http://www.csie.ntu.edu.tw/~cjlin/cgi-bin/libsvm.cgi?+http://www.csie.ntu.edu.tw/~cjlin/libsvm+tar.gz or install scipy on your desktop for a robust, pre-packaged python binding. Do

import sandbox.svm.libsvm

or the other classes. There is good documentation at http://cours-info.iut-bm.univ-fcomte.fr/docs/python/scipy/scipy.sandbox.svm-module.html

 
shortproject/ml.txt · Last modified: 2008/03/18 16:47 by honicky
 
Recent changes RSS feed Creative Commons License Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki