Rock-Paper-Scissors-master
说明： rock paper scissors algorithm
文件列表:
a.png (55566, 2019-06-19)
b.png (55775, 2019-06-19)
regret_matching.ipynb (166935, 2019-06-19)
b.png (55775, 2019-06-19)
regret_matching.ipynb (166935, 2019-06-19)
## Regret Matching for Rock-Paper-Scissors
Regret Matching algorithm for computing Rock-Paper-Scissors Nash Equilibrium
### Algorithm
**Regret Matching algorithm** 
will maintain the **vector of weights assigned to experts** 
. After loss vector is revealed we can compute cumulative regret with respect to an expert 
at time 
(it expresses how we regret not listening particular expert ):
Having that, experts weights are updated with the formula:
_+&space;=&space;\max(0,&space;R_{i,t}) "w_{i,t} = (R_{i,t})_+ = \max(0, R_{i,t})")
and finally components of our vector 
(a probability distribution vector over 
experts) are given by:
You can find a more comprehensive explanation of this algorithm [here](https://int8.io/counterfactual-regret-minimization-for-poker-ai/).
### Uasge
Check `regret_matching.ipynb` for more infomation.
### Result
Both players find the Nash equilibrium of the game and play rock/paper/scissors with roughly the same probability.
 |
 |
| strategy of player a | strategy of player b |
|---|
近期下载者：
相关文件：
收藏者：