def evaluate_players(p1: Player,
p2: Player,
games_per_battle=100,
num_battles=100):
board = Board()
p1_wins = []
p2_wins = []
draws = []
game_number = []
game_counter = 0
TFSessionManager.set_session(tf.Session())
TFSessionManager.get_session().run(tf.global_variables_initializer())
for i in range(num_battles):
p1win, p2win, draw = battle(p1, p2, games_per_battle, False)
p1_wins.append(p1win)
p2_wins.append(p2win)
draws.append(draw)
game_counter = game_counter + 1
game_number.append(game_counter)
TFSessionManager.set_session(None)
return game_number, p1_wins, p2_wins, draws
from tic_tac_toe.RndMinMaxAgent import RndMinMaxAgent
from tic_tac_toe.DirectPolicyAgent import DirectPolicyAgent
min_reward = -3
max_reward = 3
num_reward_steps = 1 + max_reward - min_reward
rewards = np.zeros((num_reward_steps, num_reward_steps))
for loss_reward in range(min_reward, max_reward):
for draw_reward in range(loss_reward + 1, max_reward + 1):
tf.reset_default_graph()
TFSessionManager.set_session(tf.Session())
sess = TFSessionManager.get_session()
nnplayer = DirectPolicyAgent("PolicyLearner1",
loss_value=loss_reward,
draw_value=draw_reward)
rm_player = RndMinMaxAgent()
sess.run(tf.global_variables_initializer())
game_number, p1_wins, p2_wins, draws = evaluate_players(
nnplayer, rm_player, num_battles=1000,
silent=True) # , num_battles = 20)
print("With loss reward {} and draw reward {} we get draws: {}".format(