dummy-link

MAB

A Julia Package for providing Multi Armed Bandit Experiments

Readme

MAB.jl - A Package for Bandit Experiments

This package provide a framework for developing and comparing various Bandit algorithms

Available Algorithms

  1. Uniform Strategy (Randomly picking some arm)
  2. ϵ-greedy
    1. ϵ-greedy
    2. ϵ_n greedy
  3. Upper Confidence Bound Policies
    1. UCB1
    2. UCB-Normal
    3. UCB-V
    4. Bayes-UCB (For Bernoulli Rewards)
    5. KL-UCB
    6. Discounted-UCB
    7. Sliding Window UCB
  4. Thompson Sampling
    1. Thompson Sampling
    2. Dynamic Thompson Sampling
    3. Optimistic Thompson Sampling
    4. TSNormal (Thompson Sampling for Gaussian distributed rewards)
    5. Restarting Thompson Sampling
    6. TS With Gaussian Prior
  5. EXP3
    1. EXP3
    2. EXP3.1
    3. EXP3-IX
  6. SoftMax
  7. REXP3
  8. Gradient Bandit

Available Arm Models

  1. Bernoulli
  2. Beta
  3. Normal
  4. Sinusoidal (without noise)
  5. Pulse (without noise)
  6. Square
  7. Variational (without noise)

First Commit

10/19/2016

Last Touched

3 months ago

Commits

122 commits

Used By: