8000 GitHub - pilotCapp/shallowstack: simplified version of Deepstack used to solve imperfect information games, specifically 24 card Texas holdem poker
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

simplified version of Deepstack used to solve imperfect information games, specifically 24 card Texas holdem poker

Notifications You must be signed in to change notification settings

pilotCapp/shallowstack

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

requirements:

numpy tensorflow phevaluator ++?

How to run

open main and run the program

Marameters

you can change the parameters in the config file, however make sure NN are trained with new parameters in mind

Modules

Game Manager

Creates the players, sets up the game, and controls the current players and flow of the game, together with deciding what legal actions a player can take

Poker Oracle

Creates the decks, picking cards based on current cards on table, computes win probability using rollout which is used to compute the cheat sheet(which is not beeing used). Also more importantly it can rank any two collections of hands, currently using an extension phyevaluator, because my own was too slow. It uses this to create the utility matrix used in resolving.

State Manager

creates and modifies states based on legal actions from the game manager and selection of the players or other components that need to change the states.

Resolver

takes a state and generates a tree of nodes with different states, which is used to update ranges and evaluations that generates regret matrixes which is used to optimalize strategies. This is used for resolving and by the resolving AI agents.

NN manager

generates training data using the resolver. It bootstraps training where the utility matrixes are used to train generate the river data, and then the river model is used for the turn data etc...

How to generate data and train models

go to the NN_creation file and run the program. Make sure lower tier models are created before creating the data. The sequence is (flop-turn-river) (data generation takes a long time so the current models have only about 100 instances, however training data can be accumulated over time, simply generate more data and then a new model)

Additional components

Players- User and AI

Nodes- Chance and Action

State

About

simplified version of Deepstack used to solve imperfect information games, specifically 24 card Texas holdem poker

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

0