pybandit

A Python library for all popular multiarmed bandit algorithms.

Roadmap

The following algorithms are currently being explored before we set a roadmap for the first release.

1. Epsilon Bandit:
   1. Epsilon-greedy strategy (done)
   2. Epsilon-first strategy (in progress)
   3. Epsilon-decreasing strategy
   4. Epsilon-adaptive strategy
2. Bayesian Bandit
   1. Thompson Sampling
3. Contextual Bandit
   1. Linear Classifier
      1. LinUCB (Upper Confidence Bound) algorithm:
      2. LinRel (Linear Associative Reinforcement Learning) algorithm:
   2. Non-linear Classifier
      1. UCBogram algorithm
      2. NeuralBandit algorithm
      3. KernelUCB algorithm
      4. Bandit Forest algorithm
   3. Constrained
      1. UCB-ALP algorithm
   4. Greedy
      1. Contextual-Epsilon-greedy strategy
4. Adverserial Bandit
5. Dueling Bandit
6. Collaborative Bandit
   1. COFIBA