dont_burn_my_cpu = pygame.time.Clock()
num = 0
mon = {'top': 0, 'left': 0, 'width': 200, 'height': 200}
sct = mss.mss()
write_to_file_step = 0.95
broj_ociscenih_linija_file = 0
broj_partija_file = 0
while 1:
app.screen.fill((0, 0, 0))
if app.gameover:
if br == None:
broj_ociscenih_linija_file += app.lines
broj_partija_file += 1
episodes = 0
app.gameover = True
agent.replay(800)
agent.target_train()
app.start_game()
tetris = Tetris()
# belezenje rezultata
if agent.epsilon < write_to_file_step:
with open("deep-q-learning-results.txt", "a") as myfile:
myfile.write(
"Broj ociscenih linija za exploration rate veci od "
+ str(write_to_file_step) + " je " +
str(broj_ociscenih_linija_file) +
" u broju partija " + str(broj_partija_file) +
"\n")
write_to_file_step = write_to_file_step - 0.05
broj_ociscenih_linija_file = 0
broj_partija_file = 0
batch_size = 16
loss_overall = []
error_avg = []
error = []
states = sim_env.state
for e in range(episodes):
sim_env.reset()
for k in range(training):
states = np.reshape(states, [1, state_size])
action = QN.act(states)
next_state, rewards, overall_err = sim_env.Assign_Cores(action)
next_state = np.reshape(next_state, [1, state_size])
QN.remember(states, action, rewards, next_state)
states = next_state
if len(QN.memory) > batch_size:
QN.replay(batch_size)
loss_overall = np.append(loss_overall, QN.loss_avg/training)
QN.loss_avg = 0
sim_env.reset()
for u in range(testing):
states = np.reshape(states, [1, state_size])
action = QN.act_test(states)
# print('SNR:', sim_env.SNR[-1])
# print('action:', sim_env.action[action])
next_state, rewards, overall_err = sim_env.Assign_Cores(action)
error = np.append(error, overall_err)
next_state = np.reshape(next_state, [1, state_size])
states = next_state
print(e)
batch_size = 32
n_episodes = 500
state_size = env.observation_space.shape[0]
action_size = env.action_space.n
agent = DQNAgent(state_size , action_size)
done = False
for i in range(n_episodes):
state = env.reset()
state = np.reshape(state,[1,state_size])
for time in range(5000):
if(i % 50 == 0):
env.render()
action = agent.act(state)
next_state,reward,done, _ = env.step(action)
reward = reward if not done else -10
next_state = np.reshape(next_state,[1,state_size])
if(np.random.rand() < 0.3):
agent.remember(state,action,reward,next_state,done)
state = next_state
if done:
print("episode: {}/{} , score: {}".format(i,n_episodes,time))
break
if(len(agent.memory) > batch_size):
agent.replay(batch_size)
if i % 100 == 0:
agent.save("weights_cartpole_{}.hdf5".format(i))
