iboost

Overview

Itemset boosting performs linear regression in the complete space of power sets of mutations. It implements a forward feature selection procedure where, in each iteration, one combination is found by an efficient branch-and-bound search. The method uses all possible combinations, and salient combinations are explicitly shown.

Dataset

HIV1_Comp (in matlab format) is compiled from Stanford HIVDB in Dec 2006. Encoding of mutations: NRTI NNRTI PI

Software Requirements

- Linux
- Python
- Matlab R14
- CVX Matlab Optimization Kit (not included): Available from author's web page .
- LCM (included): Please note that the latest code and news about LCM is available from author's web page .

Download

iboost-0.1.tar.gz

Install

% make
Then please adjust the lcm_path environment in src/run.py around line 50, and lcm_path environment in bin/lcm.m around line 21. Start matlab, and executing example.m will show an example.

Disclaimer

The author is not responsible for implications from the use of this software.

Copyright information

This software is distributed under the GNU General Public License 2.0.

References

[Demiriz2002], Ayhan Demiriz, Kristin P. Bennett and John Shawe-Taylor, "Linear Programming Boosting via Column Generation", 2002, Journal of Machine Learning, Vol. 46, pages 225-254,
[Uno2005], Takeaki Uno, Masashi Kiyomi, Hiroki Arimura, "LCM ver.3: Collaboration of Array, Bitmap and Prefix Tree for Frequent Itemset Mining" Open Source Data Mining Workshop on Frequent Pattern Mining Implementations 2005
[Nowozin07], Sebastian Nowozin, Koji Tsuda, Takeaki Uno, Koji Tsuda and Goekhan Baklr, "Weighted Substructure Mining for Image Analysis", CVPR 2007
[Saigo07], Hiroto Saigo, Takeaki Uno and Koji Tsuda "Mining complex genotypic features for predicting HIV-1 drug resistance" Bioinformatics 2007 (in press)

Link

graph boosting package: gboost

Authors

Sebastian Nowozin
Koji Tsuda
Hiroto Saigo
Takeaki Uno

Contact: hiroto.saigo@tuebingen.mpg.de