def main(unused_argv):
game = pyspiel.load_game(
FLAGS.game, {"players": pyspiel.GameParameter(FLAGS.num_players)})
oracle = optimization_oracle.EvolutionaryStrategyOracle(
n_evolution_tests=FLAGS.n_evolution_tests,
number_policies_sampled=FLAGS.number_policies_sampled,
number_episodes_sampled=FLAGS.number_episodes_sampled,
alpha=FLAGS.alpha,
beta=FLAGS.beta)
g_psro_solver = generalized_psro.GenPSROSolver(
game,
oracle,
sims_per_entry=FLAGS.sims_per_entry,
meta_strategy_method='nash',
rectify_training=FLAGS.rectify_training)
for a in range(FLAGS.gen_psro_iterations):
g_psro_solver.iteration()
nash_probabilities = g_psro_solver.get_and_update_meta_strategies()
logging.info("%s / %s", a + 1, FLAGS.gen_psro_iterations)
logging.info(nash_probabilities)
meta_game = g_psro_solver.get_meta_game
meta_probabilities = g_psro_solver.get_and_update_meta_strategies()
logging.info("%s meta probabilities", FLAGS.game)
logging.info(meta_probabilities)
logging.info("")
logging.info("%s Meta Game Values", FLAGS.game)
logging.info(meta_game)
logging.info("")
def test_gpsro(self, game_name, rnr_iterations, sims_per_entry,
number_players, rectify_training, training_strategy_selector,
meta_strategy_method):
game = pyspiel.load_game(game_name,
{"players": pyspiel.GameParameter(number_players)})
oracle = optimization_oracle.EvolutionaryStrategyOracle(
number_policies_sampled=2, number_episodes_sampled=2)
g_psro_solver = generalized_psro.GenPSROSolver(
game,
oracle,
sims_per_entry=sims_per_entry,
rectify_training=rectify_training,
training_strategy_selector=training_strategy_selector,
meta_strategy_method=meta_strategy_method)
for _ in range(rnr_iterations):
g_psro_solver.iteration()
meta_game = g_psro_solver.get_meta_game
meta_probabilities = g_psro_solver.get_and_update_meta_strategies()
logging.info("%s %sP - %s", game_name, str(number_players),
meta_strategy_method)
logging.info("Meta Strategies")
logging.info(meta_probabilities)
logging.info("")
logging.info("Meta Game Values")
logging.info(meta_game)
logging.info("")