bandit 算法资料大全

算法介绍：

1.课程两节   Tutorial: Introduction to Bandits: Algorithms and Theory

    http://techtalks.tv/talks/54451/

    http://techtalks.tv/talks/54455/

2.博文介绍  Multi_armed bandit 

  https://mpatacchiola.github.io/blog/2017/08/14/dissecting-reinforcement-learning-6.html

toolbox:

1. Project details for pymabandits

    http://mloss.org/software/view/415/

2.Multi-Armed Bandit project  (version0.2  2005)  C#

   http://bandit.sourceforge.net/

3. bandit lib  (github  C++)

     https://github.com/jkomiyama/banditlib

      这个作者还有两个bandit算法库

    没有优化算法速度，支持 linux/GNU C++ environment. 不支持windows/MacOSX

   

• Arms:

• Binary and Normal distribution of rewards (arms) are implemented.

• Policies:

• DMED for binary rewards [1]
• Epsilon-Greedy
• KL-UCB [2]
• MOSS [3]
• Thompson sampling for binary rewards [4]
• UCB [5]
• UCB-V [6]

4.https://github.com/bgalbraith/bandits

Bandits

Python library for Multi-Armed Bandits

Implements the following algorithms:

• Epsilon-Greedy
• UCB1
• Softmax
• Thompson Sampling (Bayesian)
  • Bernoulli, Binomial <=> Beta Distributions

5. MOE 里面包括了bandit, 好像挺大的一个工具包

http://yelp.github.io/MOE/why_moe.html

6.libbandit

https://github.com/tor/libbandit

#LibBandit

LibBandit is a C++ library designed for efficiently simulating multi-armed bandit algorithms.

Currently the following algorithms are implemented:

• UCB
• Optimally confident UCB
• Almost optimally confident UCB
• Thompson sampling (Gaussian prior)
• MOSS
• Finite-horizon Gittins index (Gaussian/Gaussian model/prior)
• An approximation of the finite-horizon Gittins index
• Bayesian optimal for two arms (Gaussian/Gaussian model/prior)

算法程序（不是工具包）

算法图形化展示：

1.Bayesian Bandit Explorer

  https://learnforeverlearn.com/bandits/

2.Multi-armed Bandit Demo (steven用过）