Yaniv

Yaniv is a highly addictive backpacker's card game.

For rules see the Wikipedia page.

The main objectives for this ongoing project:

• Calculate the game statistics
• Train bots with the end goal to learn strategy
• Make Interactive

In the current format yaniv can run a game with N players from an overlooker's point of view.
One nice feature added is calculating the statistics of a Yaniv declaration from the declarer's point of view. It is currently correct for two players, and requires more work for N>2.

To run a game go to the yaniv/yaniv directory and do:

python run.py

This will yield the game results, round for round with a mention of the Yaniv success statistics, e.g:

Albert (bot) strategy: picks if min pile top min value  <= 3
----------

Roland (bot) strategy: picks if min pile top min value  <= 5
----------

Amos (bot) strategy: picks if min pile top min value  <= 5
----------

Claude (bot) strategy: picks if min pile top min value  <= 5
----------
Deck of 54 cards
['W☻', 'W☺', 'A♦', 'A♥', 'A♣', 'A♠', '2♦', '2♥', '2♣', '2♠', '3♦', '3♥', '3♣', '3♠', '4♦', '4♥', '4♣', '4♠', '5♦', '5♥', '5♣', '5♠', '6♦', '6♥', '6♣', '6♠', '7♦', '7♥', '7♣', '7♠', '8♦', '8♥', '8♣', '8♠', '9♦', '9♥', '9♣', '9♠', '10♦', '10♥', '10♣', '10♠', 'J♦', 'J♥', 'J♣', 'J♠', 'Q♦', 'Q♥', 'Q♣', 'Q♠', 'K♦', 'K♥', 'K♣', 'K♠']
Round 7:
	Lucky Claude! Aggregated 100 points reduced to 50
Claude is the winner with 194 points

For more output information change verbose (values 0 to 3).

Statistics

Within stats you will find code that can calculates at a given round:
Given a player has declared victory (yaniv!), what is the probability that they truly have the lowest hand.

For this you will need to update within run.py:

do_stats = True
verbose = 1
players = ['Albert','Roland'] # currently limited to two players

Example output of a round

Round: 23
Player starting the round: Roland
n_j=4, h_i=7, n_jokers=1 yields
t_nj=6
N: 36
['3c', '8c', '3h', 'Kh', 'Qd', 'Ac', '10d', '5d', 'Ks', 'Jd', 'Qc', '3s', 'Qs', 'Kd', '5h', '8h', '5s', 'Wb', '4d', '10c', '2d', 'Ad', '9h', 'As', '4h', 'Jc', 'Kc', '7d', '7c', '4c', '2c', '2h', '6d', '6c', '3d', '10s']
n: 17
['3c', '3h', 'Ac', '5d', '3s', '5h', '5s', 'Wb', '4d', '2d', 'Ad', 'As', '4h', '4c', '2c', '2h', '3d']
total: 2380
successes: 2179
P(h_j>h_i|U)=0.916
N=36, K=17, n_j=4, yields:
p(U)=0.040
P(h_j>h_i=7|n_j, cards)=0.997
~~~~~~~~~~
Round Conclusion
Albert declared Yaniv with 7
Albert 0 ['2♠', 'W☻', '4♠', 'A♥'] 185
Roland 15 ['2♥', '5♥', '3♠', '5♣'] 138

From here we learn that:

• Roland started the round,
• Albert declared yaniv
  • with a hand of h_i=7
  • was unaware of N=36 cards
  • (of which Roland had n_j=4 cards in hand).
• Given those facts from Albert's POV the probability of the decleration being successful is P(h_j>h_i=7|n_j, cards)=0.997
• Roland has h_j=15 from cards ['2♥', '5♥', '3♠', '5♣']
• Resulting in Albert not getting any points where Roland got 15 (all points are bad).

Reinforcement Learning

This is a work in progress.
Currently there is a working Classifier that trains on the probability that it should pick up from the deck of cards (or top of pile):

Simulates two players, where both have basic deterministic policies (see rl_utils.basic_policy), and a DNN trains on the policy of the first player.

To test do:

python run_train_deck_pickup.py

Human play

There is some preliminary human playing mode, but still requires more work.

For a quick peek set one of the players to human, like this:

players = {'<Your Name>': 'human','Roland':'bot', 'Amos':'bot','Claude':'bot'}

Requirements

python 3

• numpy
• scipy
• pandas
• tensorflow 2 (for reinforcement learning)

Testing

Run bash command

py.test

(or pytest)

Strategies Currently Used

• Pull card:
  • bots are assigned a maximum value used to pull from pile, otherwise default from deck (currently between 3 and 5; see game.Player.pile_pull_strategy, game.Round.pull_card)
  • if choosing from pile and there is more than one accessible card, bots hypothesise the best combination for the next round (game.Round.pull_card_from_pile_top)
• Throw card: bots throw the highest valued valid combination (game.Round.throw_cards_to_pile)

Comments

• Jokers values are currently 0, but not used yet a s wild cards for streaks (i.e, no use in cards.cards_to_consecutive_combinations)

Credits

Image credit: Barak Edry from his excellent App Store game of Yaniv