def play_game(self):
while True:
players = self.pick_players()
game = LandlordGame(players=players)
# play a meaningful game
game.play_round()
if game.has_winners():
players_to_record = game._winners
if random.random() < self.record_everyone_pct:
players_to_record = list(TurnPosition)
for pos in players_to_record:
player = game.get_ai(pos)
self.record_player(game, player)
self.track_stats(game)
break
# clear out in case a full game wasn't played
for player in players:
player._reset_records()
def test_features_v2(self):
players = [
LearningPlayer_v2(name='random',
epsilon=0,
estimation_mode=LearningPlayer.ACTUAL_Q,
learning_rate=1) for _ in range(3)
]
game = LandlordGame(players=players)
while not game.is_round_over():
curr_player = game.get_current_player()
curr_features = curr_player._derive_features(game)
best_move, best_move_q = curr_player.decide_best_move(game)
curr_move_vector = game.get_current_player().compute_move_vector(
game.get_current_position(), game.get_landlord_position(),
best_move)
curr_hand_vector = game.get_current_player(
).compute_remaining_hand_vector(game, curr_move_vector,
game.get_current_position())
curr_player.record_move(game, best_move, best_move_q,
game.get_current_position())
game.play_move(best_move)
self.assertTrue(
np.allclose(curr_features,
curr_player.record_history_matrices[-1]))
self.assertTrue(
np.allclose(curr_move_vector,
curr_player.record_move_vectors[-1]))
self.assertTrue(
np.allclose(curr_hand_vector,
curr_player.record_hand_vectors[-1]))
players[0].compute_future_q(game)
if game.has_winners():
print(np.max(np.abs(players[0].get_estimated_qs())))
self.assertTrue(np.max(np.abs(players[0].get_estimated_qs())) == 1)
self.assertTrue(
players[0].record_history_matrices[0][0].dtype == np.int8)
def human_game(player_names, perspective):
perspective_hand = None
players = []
for player_name in player_names:
if player_name == perspective:
perspective_hand = manual_hand()
player_is_perspective = (player_name == perspective)
players.append(
HumanPlayer(name=player_name,
reference_player=reference_player,
known_hand=player_is_perspective,
ai_before=player_is_perspective))
game = LandlordGame(players, kitty_callback=manual_kitty)
first_player = get_first_player(game)
game.force_current_position(first_player)
game.force_hand(perspective_position(game, perspective), perspective_hand)
while not game.is_round_over():
current_player = game.get_current_player()
current_position = game.get_current_position()
best_move, best_move_q = current_player.decide_best_move(game)
print(current_player.get_name(), "(" + game.get_position_role_name(current_position) + ", " \
+ str(len(game.get_hand(current_position))) + "):", best_move, '(' + str(best_move_q) + ')')
# play with known hand if it matches perspective
game.play_move(best_move,
hand_known=current_player.get_name() == perspective)
if type(game.get_last_played()) == KittyReveal:
print(game.get_last_played())
if game.has_winners():
for winner in game.get_winners():
print('WINNERS:', game.get_ai_players()[winner].get_name())
def play_against_two(players, show_q=True):
game = LandlordGame(players)
while not game.is_round_over():
current_player = game.get_current_player()
current_position = game.get_current_position()
best_move, best_move_q = current_player.decide_best_move(game)
if show_q:
best_move_q_str = '(' + str(best_move_q) + ')'
else:
best_move_q_str = ''
print(current_player.get_name(), "(" + game.get_position_role_name(current_position) + ", " \
+ str(len(game.get_hand(current_position))) + "):", best_move, best_move_q_str)
game.play_move(best_move)
if type(game.get_last_played()) == KittyReveal:
print(game.get_last_played())
if game.has_winners():
for winner in game.get_winners():
print('WINNERS:', game.get_ai_players()[winner].get_name())
if __name__ == "__main__":
#players = [load_net('4_11_actualq4_model20'),
# load_net('4_11_actualq4_model20'),
# load_net('4_11_actualq4_model20')]
players = [
load_v2_net('4_13_stream2_model3_170', '../stream_models/'),
load_v2_net('4_13_stream2_model2_194', '../stream_models/'),
load_v2_net('4_13_stream2_model1_141', '../stream_models/')
]
while True:
game = LandlordGame(players=players)
game.play_round(debug=True)
if game.has_winners():
break
def printout_floats(array):
print(', '.join(["%.3f" % val for val in array]))
print('\n')
for i in range(3):
players[i].compute_future_q(game)
print(players[i].get_name())
printout_floats(players[i].get_estimated_qs())
printout_floats(players[i]._record_state_q)
print('\n')
print('')