def play_n_plot():
r1 = RandomPlayer()
r2 = RandomPlayer()
game_number, p1_wins, p2_wins, draws = evaluate_players(r1,
r2,
num_battles=100)
p = plt.plot(game_number, draws, 'r-', game_number, p1_wins, 'g-',
game_number, p2_wins, 'b-')
plt.show()
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(
loss_reward, draw_reward, draws[-1]))
rewards[loss_reward - min_reward, draw_reward - min_reward] = draws[-1]
TFSessionManager.set_session(None)
fig, ax = plt.subplots()
im = ax.imshow(rewards)
reward_range = np.arange(num_reward_steps + 1)
# We want to show all ticks...
# nnplayer = EGreedyNNQPlayer("QLearner1")#, learning_rate=0.001, win_value=10.0, loss_value=-10.0)
# nn2player = EGreedyNNQPlayer("QLearner2")#, learning_rate=0.001, win_value=10.0, loss_value=-10.0)
mm_player = MinMaxAgent()
rndplayer = RandomPlayer()
rm_player = RndMinMaxAgent()
TFSessionManager.set_session(tf.Session())
sess = TFSessionManager.get_session()
writer = tf.summary.FileWriter(TENSORLOG_DIR, sess.graph)
nnplayer.writer = writer
sess.run(tf.global_variables_initializer())
# game_number, p1_wins, p2_wins, draws = evaluate_players(rndplayer, nnplayer, num_battles=10000) #, num_battles = 20)
# game_number, p1_wins, p2_wins, draws = evaluate_players(rndplayer, nnplayer) #, num_battles = 20)
# game_number, p1_wins, p2_wins, draws = evaluate_players( mm_player, nnplayer, num_battles=300) # , num_battles = 20)
game_number, p1_wins, p2_wins, draws = evaluate_players(
nnplayer, rm_player, num_battles=1000,
writer=writer) # , num_battles = 20)
# game_number, p1_wins, p2_wins, draws = evaluate_players(nnplayer, rndplayer, num_battles=100) # , num_battles = 20)
# game_number, p1_wins, p2_wins, draws = evaluate_players(mm_player, nn2player, num_battles=100) # , num_battles = 20)
writer.close()
p = plt.plot(game_number, draws, 'r-', game_number, p1_wins, 'g-', game_number,
p2_wins, 'b-')
plt.show()
TFSessionManager.set_session(None)
if not train:
TFSessionManager.load_session('models/SimpleNNQPlayer')
sess = TFSessionManager.get_session()
if train:
sess.run(tf.global_variables_initializer())
# num battles
nb = 500
# games per battle
gpb = 100
game_number, p1_wins, p2_wins, draws = evaluate_players(dddplayer,
rmmplayer,
num_battles=nb,
games_per_battle=gpb)
if train:
TFSessionManager.save_session('models/models_session2')
plt.plot(game_number, draws, color=(0.7, 0.7, 0.7), label='draws')
plt.plot(game_number, p1_wins, 'r-', label='player 1')
plt.plot(game_number, p2_wins, 'y-', label='player 2')
plt.xlabel('battle iterations ({} games per battle)'.format(gpb))
plt.ylabel('battle winning ratio (%)')
plt.legend(loc='best')
plt.show()
TFSessionManager.set_session(None)
import os
import random
print('start')
pickle_in = open('player1.pickle', 'rb')
qvals = pickle.load(pickle_in)
print(os.stat('player1.pickle').st_size / (1024 * 1024))
print(qvals[(4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4)])
total = 0
for key in qvals.keys():
total += qvals[key].count(0)
print(total / len(qvals))
discreetPlayer = DiscreetQValuesPlayer(training=False, qValues=qvals)
rndplayer = RandomPlayer()
game_number, p1_wins, p2_wins, draws, allPos = evaluate_players(
discreetPlayer, rndplayer, games_per_battle=100000, num_battles=1)
total = 0
numSeen = 0
numNotSeen = 0
for key in allPos:
if key in qvals:
total += qvals[key].count(0)
numSeen += 1
else:
numNotSeen += 1
print(total / numSeen)
print(numSeen)
print(numNotSeen)
print(numSeen / (numNotSeen + numSeen))
for i in range(3):
print(random.choice(tuple(allPos)))
from tic_tac_toe.ExpDoubleDuelQPlayer import ExpDoubleDuelQPlayer
tf.reset_default_graph()
nnplayer = ExpDoubleDuelQPlayer(
"QLearner1") # , win_value=100.0, loss_value=-100.0)
# nn2player = EGreedyNNQPlayer("QLearner2", win_value=100.0, loss_value=-100.0)
# nnplayer = EGreedyNNQPlayer("QLearner1")#, learning_rate=0.001, win_value=10.0, loss_value=-10.0)
# nn2player = EGreedyNNQPlayer("QLearner2")#, learning_rate=0.001, win_value=10.0, loss_value=-10.0)
mm_player = MinMaxAgent()
rndplayer = RandomPlayer()
rm_player = RndMinMaxAgent()
TFSessionManager.set_session(tf.Session())
TFSessionManager.get_session().run(tf.global_variables_initializer())
# game_number, p1_wins, p2_wins, draws = evaluate_players(rndplayer, nnplayer, num_battles=10000) #, num_battles = 20)
# game_number, p1_wins, p2_wins, draws = evaluate_players(rndplayer, nnplayer) #, num_battles = 20)
# game_number, p1_wins, p2_wins, draws = evaluate_players( mm_player, nnplayer, num_battles=300) # , num_battles = 20)
game_number, p1_wins, p2_wins, draws = evaluate_players(
rm_player, nnplayer, num_battles=300) # , num_battles = 20)
# game_number, p1_wins, p2_wins, draws = evaluate_players(nnplayer, rndplayer, num_battles=100) # , num_battles = 20)
# game_number, p1_wins, p2_wins, draws = evaluate_players(mm_player, nn2player, num_battles=100) # , num_battles = 20)
p = plt.plot(game_number, draws, 'r-', game_number, p1_wins, 'g-', game_number,
p2_wins, 'b-')
plt.show()
TFSessionManager.set_session(None)
self.random_move_prob *= self.random_move_decrease
if self.writer is not None:
self.writer.add_summary(summary, self.game_counter)
summary = tf.Summary(value=[
tf.Summary.Value(tag='Random_Move_Probability',
simple_value=self.random_move_prob)
])
self.writer.add_summary(summary, self.game_counter)
TFSN.get_session().run(self.graph_copy_op)
tf.reset_default_graph()
nnplayer = deepPlayer("QLearner1")
rndplayer = RandomPlayer()
TFSN.set_session(tf.Session())
TFSN.get_session().run(tf.global_variables_initializer())
game_number, p1_wins, p2_wins, draws = evaluate_players(rndplayer,
nnplayer,
games_per_battle=1000,
num_battles=5)
p = plt.plot(game_number, p1_wins, 'g-', game_number, p2_wins, 'b-')
plt.show()
TFSN.set_session(None)