def test_child_function_expected_behavior_for_sim_game(self):
"""Test expected behavior of child on simultaneous games."""
game = pyspiel.load_game("python_iterated_prisoners_dilemma")
parameter_state = game.new_initial_state()
actions = [1, 1]
new_state = policy.child(parameter_state, actions)
self.assertEqual(str(new_state), ("p0:D p1:D"))
def test_child_function_expected_behavior_for_seq_game(self):
"""Test expected behavior of child on sequential games."""
game = pyspiel.load_game("tic_tac_toe")
initial_state = game.new_initial_state()
action = 3
new_state = policy.child(initial_state, action)
self.assertNotEqual(new_state.history(), initial_state.history())
expected_new_state = initial_state.child(action)
self.assertNotEqual(new_state, expected_new_state)
self.assertEqual(new_state.history(), expected_new_state.history())
def decision_nodes(self, parent_state):
"""Yields a (state, cf_prob) pair for each descendant decision node."""
if not parent_state.is_terminal():
if (parent_state.current_player() == self._player_id
or parent_state.is_simultaneous_node()):
yield (parent_state, 1.0)
for action, p_action in self.transitions(parent_state):
for state, p_state in self.decision_nodes(
openspiel_policy.child(parent_state, action)):
yield (state, p_state * p_action)
def policy_value(state,
policies: Union[List[policy.Policy], policy.Policy],
probability_threshold: float = 0):
"""Returns the expected values for the state for players following `policies`.
Computes the expected value of the`state` for each player, assuming player `i`
follows the policy given in `policies[i]`.
Args:
state: A `pyspiel.State`.
policies: A `list` of `policy.Policy` objects, one per player for sequential
games, one policy for simulatenous games.
probability_threshold: only sum over entries with prob greater than this
(default: 0).
Returns:
A `numpy.array` containing the expected value for each player.
"""
if state.is_terminal():
return np.array(state.returns())
else:
return sum(prob * policy_value(policy.child(state, action), policies)
for action, prob in _transitions(state, policies)
if prob > probability_threshold)
def test_child_function_failure_behavior_for_sim_game(self):
"""Test failure behavior of child on simultaneous games."""
game = pyspiel.load_game("python_iterated_prisoners_dilemma")
parameter_state = game.new_initial_state()
with self.assertRaises(AssertionError):
policy.child(parameter_state, 0)