PokerShark is An Open Source Texas no limit Hold'em player, based on HTN Planning techniques including risk-awareness.
PokerShark is part of my bachelor thesis at university stuttgart. Project goals were described as follows:
The game of poker is interesting in that it combines elements of stochasticity, reasoning, risk awareness, opponent modeling, and decision making with hidden information. The goal of this project is to study components that are necessary to build a successful AI poker player. Most notably:
- HTN Planning techniques that include risk-awareness
- Accurate opponent modeling based on game hand history
- Dynamic evaluation of the current state of the game based on hand history, stack values, and position
- Adaptation of strategies to type of game played (number of tables, speed, limits, etc.)
The case study will be online No-Limit Hold 'em Multi-Table Tournaments.
PokerShark is based on HTN Planning techniques including risk-awareness. In the preflop stage it uses expert knowledge to form a recommendation, which is then evaluated based on the current game situation to form a decision. Factors such as position, number of callers, number of raisers, opponents playing style, pot size ..etc are all taken into consideration when formaing the decision.
In the postflop stages of the game it uses an oracel to evaluate the strength of its hand. The hand evaluation is built using a combination of metrics such as hand strength (described in Aaron Davidson's masters thesis), hand potential, and Monte Carlo simulation.
PokerShark builds also a model for each opponent keeping track of thier action history, playing stats and playing style in addition to a weight table that describes the probability of the opponent holding every possible set of pocket cards. Using all of this information plus a dynamic risk attitude, PokerShark will choose the move with the best utility.
You can check GameAnalyzer folder to see how did PokerShark perform heads-up against dummy bots, human players, using dynamic and static risk attitude and the self-play experiments results.
PokerShark is C#-based and it requires .NET runtime to run.
To run Poker server you need Docker to be installed on your machine.
First you will need to clone the project:
git clone [email protected]:mikeashi/PokerShark.git
cd PokerShark/
git submodule init
git submodule update
start the provided server image:
cd PokerServer
./start.ps1
This will start a web server on port 80, which you can visit be navigating to http://localhost. You should wait until the RabbitMQ server has started, after that you can navigate to the v0.2.2/ folder where you can start PokerShark.exe
We have adopted the PyPoker project to work with PokerShark you can play using the web interface or using the console. We have provided easy to use docker image that can be started in matter of seconds. For more information please check PokerServer. Make sure to start the server, before starting PokerShark.
Just make sure to start PokerShark and dont cheat using its informative console :)
Currently PokerShark is only compatible with PyPokerEngine, but we have developed it to be able to support multiple formats/protocols. The current state of online poker does not provide a standard protocol to use, that is why PokerShark has its own built-in model to describe game state, it is by no means a poker engine, but a simple model that can track game state, board, player pocket and players actions.
One has to create a new interface to make PokerShark able to participate in servers with a non-supported format. The Bot class has six methods that represent the different events that can happen during a poker game:
- StartGame
- StartRound
- StartStreet
- DeclareAction
- ReceiveAction
- EndRound
If your interface manages to call these functions with the correct set of parameters, PokerShark should be able to participate in any server.
One thing to be noted here PokerShark is written using C#, and you might need to provide the game server with a java or a python client/player in this case we found that using RPC is a good solution PyPokerEngine and PyPokerGUI are python based and they require a python agent, but we were easily able to use RabbitMQ to create RPC queue.
Sadly we could not find a standard logging protocol for poker games such as PGN for chess games and we found it more helpful for development purposes to log the entire game state including round history, player positions, stacks, models including action history and the odds calculated by PokerShark.
The Bot will log the entire state of the game after the game is finished (when a new game starts). The logs will be located in a log folder with the current date and if the game has only one player a new folder with player name will be created. The game log file is a json file signed with the bot hash which is a md5 hash of the current version of the bot (the actual executable), so it is easy to verify which version of the bot had played the game.
will come soon x.0
- Pål Trefall for the amazing HTN planner fluid-hierarchical-task-network.
- Keith Rule for the great library Holdem Hand.
- @ishikota for the great projects PyPokerEngine, PyPokerGUI.
- Alan Hemmings for the pretty console library konsole.