Multi-armed Bandit library. It provides implementation of main algorithms like:
- EpsionGreedy
- BetaTS
- Softmax
- UCB1
- AnnealingSoftmax
- AnnealingEpsionGreedy
- AB testing
- Random Select
Also, the package enables trial simulations:
- Motecarlo Simulation
with different types of rewards
- Bernoulli Arms (binary rewards)
- Uniform Arms (uniformly distributed rewards)
Two ways to install:
- Clone this repository
pip install mab-hakuinadvisors
The best usage example can be found in the file simulation.ipynb
The package can be used as
- A part of another library or software
- API (in progress)
- Theoretical learning by using simulations like in
simulation.ipynb
or with different types of rewards, algorythms and simulation parameters
The full documentation of the library and its functionality can be found here
If you have any questions email to Alexey Butyrev, Drew Maniglia