pyker-engine

pyker_engine is a general purpose python poker engine for running arbitrary poker configurations.


Keywords
reinforcment learning, poker, AI, gym, blinds, clubs, community-cards, machine-learning, poker-game, python, raise-sizes, reinforcement-learning
License
Other
Install
pip install pyker-engine==0.1.0

Documentation

Logo

clubs

PyPI Status PyPI Status codecov CodeFactor Documentation Status

clubs is a python library for running arbitrary configurations of community card poker games. This includes anything from simple Leduc or Kuhn poker to full n-player No Limit Texas Hold'em or Pot Limit Omaha.

Install

Install using pip install clubs. To enable webserver rendering (see example below), install using pip install clubs[render].

Example

import random

import clubs

config = clubs.configs.NO_LIMIT_HOLDEM_SIX_PLAYER
dealer = clubs.poker.Dealer(**config)
obs = dealer.reset()

while True:
    call = obs['call']
    min_raise = obs['min_raise']
    max_raise = obs['max_raise']

    rand = random.random()
    if rand < 0.1:
        bet = 0
    elif rand < 0.80:
        bet = call
    else:
        bet = random.randint(min_raise, max_raise)

    obs, rewards, done = dealer.step(bet)
    if all(done):
        break

print(rewards)

Configuration

The type of poker game is defined using a configuration dictionary. See configs.py for some example configurations. A configuration dictionary has to have the following key value structure:

  • num_players
    • int: maximum number of players
  • num_streets
    • int: number of streets
  • blinds
    • int or list of ints: the blind distribution starting from the button e.g. [0, 1, 2, 0, 0, 0] for a 6 player 1-2 game
    • a single int is expanded to the number of players, settings blinds=0 will result in no blinds [0, 0, 0, 0, 0, 0]
  • antes
    • int or list of ints: the ante distribution starting from the button, analog to the blind distribution
    • single ints are expanded to the number of players
  • raise_sizes
    • float or str or list of floats or str: the maximum raise size as a list of values, one for each street
    • options are ints (for fixed raise sizes), float('inf') (for no limit raise sizes) or 'pot' for pot limit raise sizes
    • single values are expanded to the number of streets
  • num_raises
    • float or list of floats: the maximum number of raises for each street
    • options are ints (for a fixed number of bets per round) or float('inf') for unlimited number of raises
    • single values are expanded to the number of streets
  • num_suits
    • number of suits in the deck
  • num_ranks
    • number of ranks in the deck
  • num_hole_cards
    • number of hole cards for each player
  • num_community_cards
    • number of community cards per street
  • num_cards_for_hand
    • number of cards for a valid poker hand
  • mandatory_num_hole_cards
    • number of hole cards which must be used for a poker hand
  • start_stack
    • initial stack size

API

clubs adopts the Open AI gym interface. See clubs gym for a full clubs gym environment. To deal a new hand, call dealer.reset(), which returns a dictionary of observations for the current active player. To advance the game, call dealer.step({bet}) with an integer bet size. Invalid bet sizes are always rounded to the nearest valid bet size. When the bet lies exactly between 2 valid bet sizes, it is always rounded down. For example, if the minimum raise size is 10 and the bet is 5, the bet is rounded down to 0, i.e. call or fold.

Universal Deuces

The hand evaluator is heavily inspired by the deuces library. The basic logic is identical, but the evaluator and lookup table are generalized to work for any deck configuration with number of ranks <= 13 and number of suits <= 4 and poker hands with 5 or less cards. See the poker README for further details.

Some speed was sacrificed for the sake of better usability. Nonetheless, the evaluator is still quite fast on modern machines (Intel(R) Core(TM) i5-8250U):

>>> import clubs
>>> evaluator = clubs.poker.Evaluator(4, 13, 5)
>>> avg_time = evaluator.speed_test()
>>> print(f"Average time per evaluation: {avg_time}")
Average time per evaluation: 1.3986515504075214e-06
>>> print(f"Evaluations per second = {1.0/avg_time}")
Evaluations per second = 714974.362062254

Visualize

3 different render modes are available via dealer.render(). The default render mode uses a web front which gets exposed on localhost on a specified or random port.

Render Example

Limitations

Even though the library is designed to be as general as possible, it currently has a couple limitations:

  • Only integer chip denominations are supported
  • The evaluator only works for sub decks of the standard 52 card deck as well as a maximum of 5 card poker hands
  • Draw and stud poker games are not supported