#poker #cards #card-game #no-std

no-std aya_poker

Fast poker hand evaluation with support for most popular poker variants

1 unstable release

0.1.0 Oct 31, 2023

#300 in Games

Zlib OR Apache-2.0 OR MIT

150KB
3K SLoC

AyaPoker

Latest Release Documentation Minimum Supported Rust Version 1.64

AyaPoker is a Rust library for fast poker hand evaluation based on the OMPEval C++ hand evaluator, with support for most popular poker variants.

Features

  • Can be used to rank hands from standard poker, ace-to-five lowball, deuce-to-seven lowball, six-or-better (short-deck), Omaha, Omaha Hi/Lo, Badugi or Baduci.
  • Can evaluate hands with 0 to 7 cards, with the missing cards counting as the worst possible kickers, allowing for use in stud poker games.
  • Uses compile-time generated perfect hash function lookup tables for excellent runtime performance and fast initialization.
  • Has extensive suite of tests to ensure correct implementation of the hand ranking rules for each variant.

Flags

This crate has the following Cargo features:

  • std: By default, aya_poker is a !#[no_std] crate, but can be compiled with the std feature in order to allow the initialization of Decks with system-generated random seeds.
  • colored: Use colored to display cards and hands in color.
  • colored-4color: Same as colored, but using a four-color deck.

Example

use std::cmp::Ordering;

use aya_poker::{base::*, deck::Deck, poker_rank};

const SEED: u64 = 42;
const SAMPLE_SIZE: usize = 100_000;

fn main() -> Result<(), ParseError> {
    // We can initialize cards by specifying the rank and suit,
    // and then use collect to combine the cards into a Hand
    let player = [
        Card::new(Rank::King, Suit::Hearts),
        Card::new(Rank::King, Suit::Clubs),
    ]
    .iter()
    .collect();

    // Or, first parse the cards from strings
    let opponent = ["Js".parse::<Card>(), "Qs".parse::<Card>()]
        .iter()
        .copied()
        .collect::<Result<Hand, ParseError>>()?;

    // Or, parse a space-separated string of cards as a Hand
    let board = "Ah 5s Ts".parse()?;

    let equity = equity_calculator(&player, &opponent, &board);
    println!(
        "{} has {:.1}% equity on {} against {}.",
        player,
        100.0 * equity,
        board,
        opponent
    );

    Ok(())

}

fn equity_calculator(
    player: &Hand,
    opponent: &Hand,
    board: &Hand,
) -> f64 {
    // To simulate board run-outs, we begin by preparing a deck
    // that doesn't contain the already dealt-out cards
    let available_cards = CARDS
        .iter()
        .filter(|c| !player.contains(c))
        .filter(|c| !opponent.contains(c))
        .filter(|c| !board.contains(c));
    let mut deck = Deck::with_seed(available_cards, SEED);

    let mut pots_won = 0.0;
    for _ in 0..SAMPLE_SIZE {
        // Then, for each run we draw cards to complete the board
        deck.reset();
        let missing = 5 - board.len();
        let complete_board = board
            .iter()
            .chain(deck.deal(missing).unwrap().iter())
            .collect::<Hand>();

        // Evaluate the player's hand given the completed board
        let mut player_hand = *player;
        player_hand.extend(complete_board.iter());
        let player_rank = poker_rank(&player_hand);

        // Evaluate the opponent's hand
        let mut opponent_hand = *opponent;
        opponent_hand.extend(complete_board.iter());
        let opponent_rank = poker_rank(&opponent_hand);

        // And record the player's share of the pot for the run
        match player_rank.cmp(&opponent_rank) {
            Ordering::Greater => pots_won += 1.0,
            Ordering::Less => {}
            Ordering::Equal => pots_won += 0.5,
        };
    }

    pots_won / SAMPLE_SIZE as f64
}

Performance

All of the following benchmarks were run on an AMD Ryzen 5 2600 and compiled with lto = "thin".

Standard / Ace-to-Five Lowball / Deuce-to-Seven Lowball / Six-plus

Standard poker, ace-to-five lowball, and six-or-better (short-deck) poker are the fastest since they represent the ideal scenario for the lookup table approach: they only require computing the Hand value, and then using it to perform a single lookup.

Deuce-to-seven lowball is slightly slower for more than 5 cards due to having to iterate over non-flush card combinations.

Badugi / Baduci

Badugi / Baduci evaluation is much slower since at the moment AyaPoker only has lookup tables for rank combinations with non-overlapping suits, and thus requires a number of lookups for each such combination of cards.

Omaha

AyaPoker does not include any lookup tables for Omaha since they are too large to compute at compile time, but, due to the excellent performance of the lookup tables for standard and ace-to-five lowball poker, is still able to achieve decent performance for Omaha by iterating over all 5-card combinations.

License

Licensed under any of:

by your choice.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be multi-licensed as above, without any additional terms or conditions.

Dependencies

~0–9.5MB
~42K SLoC