lteu / sunny-fs

a branch of sunny project - experiments with a subset of instances filtered by feature selection algorithms

sunny-fs

Feature selection for SUNNY (closed).

This project has generated the results for the work:

Roberto Amadini, Fabio Biselli, Maurizio Gabbrielli, Tong Liu, Jacopo Mauro. Feature Selection for SUNNY: a Study on the Algorithm Selection Library. ICTAI, Nov 2015, Vietri sul Mare, Italy.

This is the first time that we executed a systematic study for the impact of FS techniques on SUNNY performance.

Technical notes:

• To change feature selection algorithm, please change the value of 'selection_algorithm' in the fs.py file.
• Sequential execution: Add scenario names to directories.txt file and run main.py, you will test all selected scenarios in one time.
• Parallel execution: create run_(scenario_name).py files(follow the example of run_PREMARSHALLING-ASTAR-2013.py) and then call each file from a different machine.
• Results are generated in data/results/(scenario_name).txt

NEW results

General results

Ranker FS results