def __init__(self, sess, env, network, log_dir, goal_checker=None):
self.sess = sess
self.env = env
self.network = network
self.goal_checker = goal_checker
if network.summaries_op is not None:
self.summary_writer = tf.summary.FileWriter(log_dir, flush_secs=1)
self.logger = easy_tf_log.Logger()
self.logger.set_writer(self.summary_writer.event_writer)
else:
self.summary_writer = None
self.logger = None
self.updates = 0
self.last_state = self.env.reset()
self.goal_checker.reset()
self.last_goal_inputs = self.last_state[1]
if len(self.last_goal_inputs) % 3 != 0:
raise Exception(
'Proprioceptions must be multiples of three. Current size of Proprioceptions: {0}'
.format(len(self.last_goal_inputs)))
self.goals = self.make_goals(self.last_goal_inputs)
self.last_state = self.last_state[0]
self.episode_values = []
def __init__(self, sess, env, network, log_dir):
self.sess = sess
self.env = env
self.network = network
self.bank_ops = []
self.banks_template = OrderedDict()
for key in network.memory_bank:
self.banks_template[key] = []
b = network.memory_bank[key]
self.bank_ops += b.update
if network.summaries_op is not None:
self.summary_writer = tf.summary.FileWriter(log_dir, flush_secs=1)
self.logger = easy_tf_log.Logger()
self.logger.set_writer(self.summary_writer.event_writer)
else:
self.summary_writer = None
self.logger = None
self.updates = 0
self.last_state = self.env.reset()
if type(self.last_state) is tuple:
self.last_extra_inputs = self.last_state[1]
self.last_state = self.last_state[0]
else:
self.last_extra_inputs = None
self.episode_values = []
def __init__(self, sess, env, network, log_dir, goal_checker=None):
if goal_checker is None:
raise Exception("goal_checker must be a class with __call__(goal, pos, min, max, default_value) method implemented!!!")
self.sess = sess
self.env = env
self.network = network
self.goal_checker = goal_checker
self.current_goal = 0
if network.summaries_op is not None:
self.summary_writer = tf.summary.FileWriter(log_dir, flush_secs=1)
self.logger = easy_tf_log.Logger()
self.logger.set_writer(self.summary_writer.event_writer)
else:
self.summary_writer = None
self.logger = None
self.updates = 0
self.last_state = self.env.reset()
self.goal_checker.reset()
self.last_goal_inputs = self.last_state[1]
if len(self.last_goal_inputs) % 3 != 0:
raise Exception('Proprioceptions must be multiples of three. Current size of Proprioceptions: {0}'.format(len(self.last_goal_inputs)))
self.goals = self.make_goals(self.last_goal_inputs)
self.current_goal = np.random.choice(len(self.goals))
self.last_state = self.last_state[0]
self.episode_values = []
def init_logging(self):
create_dirs()
# $ tensorboard --logdir=logs --port=6006
# then go to http://acai.local:6006/#scalars&run=.
self.rewards_dir_path = os.path.join(curr_dir_path, 'logs', 'rewards')
self.log_dir_path = os.path.join(curr_dir_path, 'logs', 'train')
self.save_historical_logs()
#using easy_tf_log
self.logger = easy_tf_log.Logger()
self.logger.set_log_dir(self.rewards_dir_path)
#using std tensorboard method
self.log_every_n = 10
self.train_writer = tf.summary.FileWriter(
"logs/train/", flush_secs=5) #self.sess.graph)
def __init__(self, sess, env, network, log_dir):
self.sess = sess
self.env = env
self.network = network
if network.summaries_op is not None:
self.summary_writer = tf.summary.FileWriter(log_dir, flush_secs=1)
self.logger = easy_tf_log.Logger()
self.logger.set_writer(self.summary_writer.event_writer)
else:
self.summary_writer = None
self.logger = None
self.updates = 0
self.last_state = self.env.reset()
self.episode_values = []
def __init__(self, env, log_prefix="", log_dir=None):
Wrapper.__init__(self, env)
if log_prefix:
self.log_prefix = log_prefix + ": "
else:
self.log_prefix = ""
if log_dir is not None:
self.logger = easy_tf_log.Logger()
self.logger.set_log_dir(log_dir)
else:
self.logger = None
self.episode_rewards = None
self.episode_length_steps = None
self.episode_n = -1
self.episode_done = None
def test_measure_rate(self):
with tempfile.TemporaryDirectory() as temp_dir:
logger = easy_tf_log.Logger(log_dir=temp_dir)
logger.measure_rate('foo', 0)
time.sleep(1)
logger.measure_rate('foo', 10)
time.sleep(1)
logger.measure_rate('foo', 25)
event_filename = list(os.scandir(temp_dir))[0].path
event_n = 0
rates = []
for event in tf.train.summary_iterator(event_filename):
if event_n == 0: # metadata
event_n += 1
continue
rates.append(event.summary.value[0].simple_value)
event_n += 1
np.testing.assert_array_almost_equal(rates, [10., 15.], decimal=1)
import time
import easy_tf_log
for i in range(3):
logger = easy_tf_log.Logger()
logger.set_log_dir(f'run-seed{i}')
logger.logkv('foo', 0)
time.sleep(1.0)
logger.logkv('foo', 1 + i / 3)
# Logging using the global logger
# Will log to automatically-created 'logs' directory
for i in range(10):
easy_tf_log.tflog('foo', i)
for j in range(10, 20):
easy_tf_log.tflog('bar', j)
easy_tf_log.set_dir('logs2')
for k in range(20, 30):
easy_tf_log.tflog('baz', k)
for l in range(5):
easy_tf_log.tflog('qux', l, step=(10 * l))
# Logging using a Logger object
logger = easy_tf_log.Logger(log_dir='logs3')
for i in range(10):
logger.log_key_value('quux', i)
logger.log_list_stats('quuz', [1, 2, 3, 4, 5])
logger.measure_rate('corge', 10)
time.sleep(1)
logger.measure_rate('corge', 20) # Logged rate: (20 - 10) / 1
time.sleep(2)
logger.measure_rate('corge', 30) # Logged rate: (30 - 20) / 2