https://github.com/mathisve/k-armed-bandit-ucb

K-armed Bandit with Upper-Confidence-Bound action selection
https://github.com/mathisve/k-armed-bandit-ucb

Last synced: about 2 months ago
JSON representation

K-armed Bandit with Upper-Confidence-Bound action selection

• Host: GitHub
• URL: https://github.com/mathisve/k-armed-bandit-ucb
• Owner: mathisve
• Created: 2020-01-06T15:31:31.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2020-07-23T21:07:25.000Z (over 4 years ago)
• Last Synced: 2023-03-07T20:10:04.190Z (almost 2 years ago)
• Language: Go
• Size: 22 MB
• Stars: 0
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# K-armed-Bandit-UCB

K-armed Bandit with Upper-Confidence-Bound action selection based on this formula.

![formula](https://github.com/Mathisco-01/K-armed-Bandit-UCB/blob/master/imgs/formula.png?raw=true)

Where *Qt(a)* is the quality of action *a* at timestep *t* (or expected reward based on the cumulative average of previous results), *c* is the **exploration** constant (where a higher *c* means a larger bias towards exploring rather than exploiting) and where *Nt(a)* is the amount of times action *a* has been picked. The last part of the algorithm incentivises the exploration of actions that haven't been explored that much.