The game of Yaniv implemented as a Gymnasium environment to be used for Reinforcement Learning. Ray RLLib is used for training agents using PPO. ParametricActions are used for action embeddings.
The goal is to win a round by getting to 5 points or less and call Yaniv.
| Card Rank | Points Value |
|---|---|
| Joker | 0 |
| Ace | 1 |
| 2-10 | same as card |
| Face Card | 10 |
- Traditional Imperfect information game
- Non-deterministic
- 2-6 players
- Deck type (52 card deck, 54 card deck with 2 jokers)
- Open/Closed Deck
- Suit (♠♥♦♣)
- Rank (A23456789TJQK)
- Points
- Cardinality
- Unicode
Moves that a player can take in a turn
- Call Yaniv
- Throw a card on the pile
- Pickup a card from the top of the deck
- Pickup a card from the top of the pile
- End its turn
The game is implemented in pure Python (no dependencies), it is wrapped to a environment encoding
0: Call Yaniv (if 5 point or lower)
1 - 54: Throw card
Spades (A2-10JQK),
Hearts (..),
Diamonds (..),
Clubs (..),
53: Throw Red Joker
54: Throw Black Joker
55: Pickup card from deck
56: Pickup card from pile
57: End turn
Box(2, 56) Binary masks
- The player's cards
- Card on top of the pile (1-Hot)