Mancala board game written in python.

• Mancala playable on CLI
• Cmpatible with the gym API
• Can train RL agents
• Mancala playable on GUI

$ pip install mancala

$ mancala play --player0 human --player1 random

The values are player1's (second move) win rates in percentage

$ mancala arena
              p0_random  p0_exact  p0_max  p0_minimax  p0_negascout
p1_random          50.0      53.0     3.0         0.0           0.0
p1_exact           42.0      48.0     4.0         1.0           1.0
p1_max             95.0      91.0    41.0         0.0           3.0
p1_minimax        100.0      96.0    87.0        30.0          39.0
p1_negascout      100.0      97.0    84.0        19.0          32.0

Mancala is a game with perfect information. マンカラは完全情報ゲームです。

Mini-max is an algorithm for n-player zero-sum games. The concept is to assume the opponent will take their best move and try to minimize them.

• MiniMax https://en.wikipedia.org/wiki/Minimax
• Alpha-beta pruning https://en.wikipedia.org/wiki/Alpha–beta_pruning
• Negamax https://en.wikipedia.org/wiki/Negamax
• NegaScout (PVS, principal variation search) https://en.wikipedia.org/wiki/Principal_variation_search

Using Dynamic Programming (DP), calculate value for states and memorize them. Use the value to plan future actions.

Other implementations

• OpenSpiel value_iteration algorithm https://github.com/deepmind/open_spiel/blob/master/open_spiel/python/algorithms/value_iteration.py example https://github.com/deepmind/open_spiel/blob/master/open_spiel/python/examples/value_iteration.py

Using Dynamic Programming (DP), calculate value for states and memorize them. Use the value and policy for planning.

• https://github.com/mdavolio/mancala

• https://github.com/deepmind/open_spiel
• https://github.com/PettingZoo-Team/PettingZoo