AdBandit: A New Algorithm For Multi-Armed Bandits
This repository contains the code for the algorithm described in the article: [AdBandit: A New Algorithm For Multi-Armed Bandits] (http://sites.poli.usp.br/p/fabio.cozman/Publications/Article/truzzi-silva-costa-cozman-eniac2013.pdf) published in ENIAC2013.
Most of the code comes from the pymabandits project, but some of them were refactored.
Unfortunately I didn't refactored everything, so you should encounter some javaish variable names all around.
I added the joblib in the Evaluator class, so the simulations can be parallelized.
I use a simple python file: configuration.py.
Example:
configuration = {
# Finite horizon of the simulation
"horizon": 10000,
# number of repetitions
"repetitions": 100,
# Number of cores for parallelization
"n_jobs": 4,
# Verbosity for the joblib
"verbosity": 5,
# Environment configuration, yeah you can set up more than one.
# I striped some code that were not published yet, but you can implement
# your own arms.
"environment": [
{
"arm_type": Bernoulli,
"probabilities": [0.02, 0.02, 0.02, 0.10, 0.05, 0.05, 0.05, 0.01, 0.01, 0.01]
}
],
# Policies that should be simulated, and their parameters.
"policies": [
{
"archtype": UCB,
"params": {}
},
{
"archtype": Thompson,
"params": {}
},
{
"archtype": klUCB,
"params": {}
},
{
"archtype": AdBandit,
"params": {
"alpha": 0.5,
"horizon": 10000
}
}
]
}It should be very straight forward. It will plot the results.
pip install -r requirements.txt
python main.py