This repository contains the code to setup the final evaluation of the course "Machine Learning: Project" (KU Leuven, Faculty of engineering, Department of Computer Science, DTAI research group).

The easiest setup is to simply copy the module tournaments.py to your own code repository. You will need Python>=3.6, openspiel, pandas, numpy and click.

You will have to submit your agent's policy (both as 1st and 2nd player) as a CSV file. These CSV files should follow the <teamname>_p<pid>.csv naming convention. For example, if you are team robberechts_meert, you should deliver two csv files: robberechts_meert_p1.csv and robberechts_meert_p2.csv. You can use the code below to create these CSV files:

from tournament import policy_to_csv

game = pyspiel.load_game("leduc_poker")
policy = <your policy for agent 1>
policy_to_csv(game, policy, '/models/agent_p1.csv')

Note that you can use PolicyFromCallable to create a policy from any function f(state) : action.

def _random_policy(state):
  actions = state.legal_actions()
  return np.choice(actions)

game = pyspiel.load_game("leduc_poker")
callable_policy = policy.PolicyFromCallable(game, _random_policy)

For the final evaluation of your agent we will play a round robin (pairwise) tournament, in which each agent will play many games (pairwise) against all other agents.

python tournament.py 'leduc_poker' <modeldir> <outputdir> --rounds 20

This script will generate two files ranking.csv (containing the final rankings) and results.csv (containing the result of each individual game). On these outcomes we will do a game-theoretic analysis to determine the winners.