def OnReset(self, event):
toolbar = self.GetToolBar()
toolbar.EnableTool(self.start_btn.GetId(), True)
toolbar.EnableTool(self.reset_btn.GetId(), False)
toolbar.EnableTool(self.enable_blue.GetId(), True)
pub.sendMessage('Reset')
server.reset()
def run(self):
total_step = 1
while self.g_ep.value < MAX_EP:
s = reset()
print("img_array", s)
s = feature_vec(s)
print("feat", s)
# s = self.env.reset()
# feature_vec
buffer_s, buffer_a, buffer_r = [], [], []
ep_r = 0.
while True:
# if self.name == 'w0':
# self.env.render()
# feature_vec
a = self.lnet.choose_action(s)
# a = self.lnet.choose_action(v_wrap(s[None, :]))
s_, r, done = step(a)
# s_, r, done, _ = self.env.step(a)
# feature_vec
print("a", a)
print("s_", s_)
print("r", r)
print("done", done)
if done:
r = -1
ep_r += r
buffer_a.append(a)
buffer_s.append(s)
buffer_r.append(r)
# update global and assign to local net
if total_step % UPDATE_GLOBAL_ITER == 0 or done:
# sync
push_and_pull(self.opt, self.lnet, self.gnet, done, s_,
buffer_s, buffer_a, buffer_r, GAMMA)
buffer_s, buffer_a, buffer_r = [], [], []
if done: # done and print information
record(self.g_ep, self.g_ep_r, ep_r, self.res_queue,
self.name)
break
s = s_
total_step += 1
self.res_queue.put(None)
def setUp(self):
global server
server.reset()
def setUp(self):
global server
server.reset()
NUM_MINIBATCH)
break
observation_stats = np.reshape(
observation_buffer, (-1, input_normalizer.num_inputs))
input_normalizer.update(
sess,
mean=observation_stats.mean(axis=0, keepdims=True),
var=observation_stats.var(axis=0, keepdims=True),
count=observation_stats.shape[0])
if not stats_ready:
progress.set_postfix_str(
"Initial statistics gathered, commencing training session"
)
sess.run([sync_op, set_stats_ready_op])
utility.apply(lambda server: server.reset(), servers)
else:
progress.set_postfix_str("Update %d completed" %
update_idx)
progress.close()
rollout_queue, stats_ready, progress = queue.Queue(
), True, None
break
if step_count > MAX_UPDATE_STEPS:
print("Completed training process, terminating session")
utility.apply(lambda server: server.stop(), servers)
rollout_queue = None
utility.apply(lambda thread: thread.join(), server_threads)