Code for ICDL-Epirob 2015's paper, 'Diversity-Driven Selection of Exploration Strategies in Multi-Armed Bandits'
HTML Other
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
code
results
.gitignore
readme.md

readme.md

Code for ICDL-Epirob 2015 paper

This is the code for ICDL-Epirob 2015's paper, 'Diversity-Driven Selection of Exploration Strategies in Multi-Armed Bandits' by Fabien C. Y. Benureau and Pierre-Yves Oudeyer.

Usage

The code for reproducing the figures presented in the paper is available in the code/figures folder. This produces html files with interactive charts. Some precomputed charts are available in the results folder.

The fixed.py and ddmab.py allow to quickly reproduce a run of the experiment.

The fixed_cluster.py and ddmab_cluster.py may require a cluster to finish in a reasonable amount of time (if the qsub command is detected, jobs will automatically be submitted to the cluster). Before execution, one should adapt the paths.py file accordingly. Then:

python ddmab_cluster -w --hd
python ddmab_cluster -w --run

You should relaunch the python ddmab_cluster -w --run command until no more job is waiting to be run.

After execution, the fixed_graph.py and ddmab_graph.py will respectively produce interactive charts.

For any problem executing the code, send me a mail at fabien.benureau@gmail.com.

Dependencies

The code depends on several libraries: numpy, scipy, shapely, bokeh, forest, explorers, environments and learners packages.

pip install numpy scipy shapely bokeh explorers environments learners

forest needs to be installed manually.

git clone https://github.com/flowersteam/forest
cd forest
python setup.py install