axelbdt / abalone-agent

Abalone AI

This projects contains several agents to play abalone on the seahorse platform.

Algorithms implemented:

• Minimax
• AB pruning
• AB pruning with nodes sorted by heuristic
• AB pruning with Transposition Table

Heuristics available:

• Distance to center of the board
• Ejected pieces with distance as a tie-breaker
• Maximise push and ejections available

Run the agents

Run once:

python main_abalone.py -g -c simplified -t local my_player_ab_distance.py random_player_abalone.py 

More info on options with python main_abalone.py -h.

Run on every file change with watchexec:

watchexec -e py -- python main_abalone.py -g -c simplified -t local my_player_ab_distance.py random_player_abalone.py 

Smaller games

Simplified config

In addition to classical and alien starting position, there is an additionnal simplified config where each player has 3 pieces available.

Shorter games

To limit the number of turns, change MAX_STEP value in constants module.

To limit the goal score, change MAX_SCORE value in constants module.

Measure agent performance

Beside victory, an agent logs the following info :

• Time remaining at the end of the game
• Nodes where score was computed (as terminal node or from children)
• (for agents with a Transposition Table) the number of successful lookups

Using a heuristic for sorted AB pruning

1. Subclass MyPlayer from ab module.
2. Define the heuristic as a scalar function of a GameState in one of the following places:
   • as the self.heuristic attribute in the __init__ method
   • if your heuristic depends on element unkown at time of __init__, override the get_heuristic method to create the heuristic using the first GameState