BayesMAB

Bayesian Multi-Armed Bandit with adaptive Thompson Sampling, suitable for A/B testing and real-time decision-making.

Install

pip install bayesmab

Usage Example

An example script is included in the examples/ directory.

You can run it with:

python examples/run_bernoulli_bandit.py

The example initializes a bandit with three arms, each simulating a different true conversion rate. It uses:

• Thompson Sampling to allocate "traffic" to each arm based on uncertainty

• Updates posterior beliefs as binary rewards are observed

• Tracks and visualizes:

  • Posterior distributions

  • Posterior mean estimates over time

  • Cumulative regret

  • Traffic allocation per arm

This simulates an A/B/n test where better-performing variants gradually receive more attention, showing how Bayesian bandits adaptively optimize decisions under uncertainty.