New fast and datamining

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

New fast and datamining

JRijnberk
Hi

I am using Weka mainly with a logistic classifier. Recently there has been
work from Komarek et al. on the dramatic improvement of logistic regression
for classification (IRLS with CG) which makes it look one of the best
classifiers fro many purposes. Moreover it is VERY FAST, is suitable for
large datasets with many attributes and instances, often doesn't require
explicite features selection and suffer overfitting.
The procedure depends on a balanced set of system parameters (Komarek 2004).
Is anybody aware if there is a weka implementation in the making?
With a lot of effort I am able to make a weka version myself if I can use it
for commercial purposes under the GNU license (as with weka). The software
is  available but not in Java and thus can be reworked for weka but not
under the GNU-license conditions . It is made available for use in public
institutions (hospitals, universities etc. ) only.

See http://komarix.org/ac/papers
eg
http://komarix.org/ac/papers/komarek_kdd2005_submission.pdf
http://www.ri.cmu.edu/pubs/download
http://www.autonlab.org/autonweb/documents/papers/komarek:lr_thesis.pdf

Looks very impressive!



Hans
van Rijnberk






Hans van Rijnberk

[hidden email]



_______________________________________________
Wekalist mailing list
[hidden email]
https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist