def alpharank_strategy(solver, return_joint=False, **unused_kwargs):
"""Returns AlphaRank distribution on meta game matrix.
This method works for general games.
Args:
solver: GenPSROSolver instance.
return_joint: a boolean specifying whether to return player-wise
marginals.
Returns:
marginals: a list, specifying for each player the alpharank marginal
distributions on their strategies.
joint_distr: a list, specifying the joint alpharank distributions for all
strategy profiles.
"""
meta_games = solver.get_meta_game()
meta_games = [
np.array([[1.1, -10], [1, -1], [-1, 1]]),
np.array([[-1.1, 10], [-1, 1], [1, -1]])
]
meta_games = [np.asarray(x) for x in meta_games]
if solver.symmetric_game:
meta_games = [meta_games[0]]
# Get alpharank distribution via alpha-sweep
joint_distr = alpharank.sweep_pi_vs_epsilon(meta_games)
joint_distr = remove_epsilon_negative_probs(joint_distr)
marginals = 2 * [joint_distr]
joint_distr = get_joint_strategy_from_marginals(marginals)
if return_joint:
return marginals, joint_distr
else:
return joint_distr
else:
joint_distr = alpharank.sweep_pi_vs_epsilon(meta_games)
joint_distr = remove_epsilon_negative_probs(joint_distr)
if return_joint:
marginals = get_alpharank_marginals(meta_games, joint_distr)
return marginals, joint_distr
else:
return joint_distr
def _alpharank(meta_game, per_player_repeats, ignore_repeats=False):
"""AlphaRank."""
if not ignore_repeats:
meta_game = _expand_meta_game(meta_game, per_player_repeats)
meta_dist = alpharank_lib.sweep_pi_vs_epsilon([mg for mg in meta_game])
meta_dist[meta_dist < DIST_TOL] = 0.0
meta_dist /= np.sum(meta_dist)
meta_dist = np.reshape(meta_dist, meta_game.shape[1:])
if not ignore_repeats:
meta_dist = _unexpand_meta_dist(meta_dist, per_player_repeats)
return meta_dist, dict()