Skip to content

Latest commit

 

History

History
38 lines (23 loc) · 877 Bytes

README.md

File metadata and controls

38 lines (23 loc) · 877 Bytes
Raw

Bandits

implement famous bandit algorithms

Requirements

I checked algorithms in the following environment:

  • OS X (10.9+), CPU
  • python (2.7.11)
  • numpy (1.11.3)
  • matplotlib (1.4.3)

K-armed Bernoulli Bandits

  • Epsilon-greedy
  • UCB
  • KL-UCB
  • Thompson sampling

Simple test can be done as follows:

python models.py

The result is shown as follows:

Contextual Bandits (incomplete)

1: Thompson sampling action selection with online logistic regression , Algorithm 3 at Chapelle, O., & Li, L. (2011). An empirical evaluation of thompson sampling. In Advances in neural information processing systems (pp. 2249-2257).

2: Epsilon greedy action selection with online logistic regression

3: Random action selection with online logistic regression (for comparison)