def time_record_end(name):
end_time = time.time()
start_time = rc_start_time[name]
logger.record_tabular("time_used/{}".format(name), end_time - start_time)
logger.info("[test] func {0} time used {1:.2f}".format(
name, end_time - start_time))
del rc_start_time[name]
def recursion_util(inp_field):
if hasattr(inp_field, "__getitem__"):
for inp in inp_field:
recursion_util(inp)
elif hasattr(inp_field, 'simple_value'):
logger.record_tabular(name + '/' + inp_field.tag,
inp_field.simple_value)
else:
pass
def time_used_wrap(name, func, *args, **kwargs):
start_time = time.time()
output = func(*args, **kwargs)
end_time = time.time()
time_used = end_time - start_time
logger.info("[test] func {0} time used {1:.2f}".format(name, time_used))
logger.record_tabular("time_used/{}".format(name), time_used)
logger.dump_tabular()
return output
def update_fph(self, cum_epochs):
if self.last_record_fph_time is None:
self.last_record_fph_time = time.time()
else:
cur_time = time.time()
duration = (cur_time - self.last_record_fph_time) / 60 / 60
fph = cum_epochs / duration
logger.record_tabular('fph', fph)
# self.last_record_fph_time = cur_time
logger.dump_tabular()
def total_episode_reward_logger(rew_acc, rewards, masks,
# writer,
steps):
"""
calculates the cumulated episode reward, and prints to tensorflow log the output
:param rew_acc: (np.array float) the total running reward
:param rewards: (np.array float) the rewards
:param masks: (np.array bool) the end of episodes
# :param writer: (TensorFlow Session.writer) the writer to log to
:param steps: (int) the current timestep
:return: (np.array float) the updated total running reward
:return: (np.array float) the updated total running reward
"""
with tf.variable_scope("environment_info", reuse=True):
for env_idx in range(rewards.shape[0]):
dones_idx = np.sort(np.argwhere(masks[env_idx]))
if len(dones_idx) == 0:
rew_acc[env_idx] += sum(rewards[env_idx])
else:
rew_acc[env_idx] += sum(rewards[env_idx, :dones_idx[0, 0]])
time_step_holder.set_time(steps + dones_idx[0, 0])
# [RLA] phase 3: log
logger.record_tabular('perf/episode_reward', rew_acc[env_idx])
# summary = tf.Summary(value=[tf.Summary.Value(tag="episode_reward", simple_value=rew_acc[env_idx])])
# writer.add_summary(summary, steps + dones_idx[0, 0])
for k in range(1, len(dones_idx[:, 0])):
rew_acc[env_idx] = sum(rewards[env_idx, dones_idx[k - 1, 0]:dones_idx[k, 0]])
time_step_holder.set_time(steps + dones_idx[k, 0])
# [RLA] phase 3: log
logger.record_tabular('perf/episode_reward', rew_acc[env_idx])
# summary = tf.Summary(value=[tf.Summary.Value(tag="episode_reward", simple_value=rew_acc[env_idx])])
# writer.add_summary(summary, steps + dones_idx[k, 0])
rew_acc[env_idx] = sum(rewards[env_idx, dones_idx[-1, 0]:])
logger.dump_tabular()
return rew_acc
def print_custom_data(self, key, prefix=''):
assert key in self.custom_data
import numpy as np
mean_val = np.mean(self.custom_data[key])
logger.record_tabular(prefix + key, mean_val)
def target_func(x):
return np.tanh(np.mean(x, axis=-1, keepdims=True))
for i in range(start_epoch, 1000):
exp_manager.time_step_holder.set_time(i)
x_input = np.random.normal(0, 3, [64, kwargs["input_size"]])
y = target_func(x_input)
loss_out, y_pred = sess.run([loss, out, opt],
feed_dict={
X_ph: x_input,
y_ph: y
})[:-1]
logger.ma_record_tabular("perf/mse", loss_out, 10)
logger.record_tabular("y_out", np.mean(y))
logger.dump_tabular()
if i % 100 == 0:
exp_manager.save_checkpoint()
if i % 10 == 0:
def plot_func():
import matplotlib.pyplot as plt
testX = np.repeat(np.expand_dims(np.arange(-10, 10, 0.1), axis=-1),
repeats=16,
axis=-1)
testY = target_func(testX)
predY = sess.run(out, feed_dict={X_ph: testX})
plt.plot(testX.mean(axis=-1), predY.mean(axis=-1), label='pred')
plt.plot(testX.mean(axis=-1), testY.mean(axis=-1), label='real')