def testCreateRunner(self, mock_create_agent, mock_runner_constructor):
base_dir = '/tmp'
run_experiment.create_runner(base_dir)
self.assertEqual(1, mock_runner_constructor.call_count)
mock_args, _ = mock_runner_constructor.call_args
self.assertEqual(base_dir, mock_args[0])
self.assertEqual(mock_create_agent, mock_args[1])
def testCreateTrainRunner(self, mock_create_agent,
mock_runner_constructor):
base_dir = '/tmp'
run_experiment.create_runner(base_dir, schedule='continuous_train')
self.assertEqual(1, mock_runner_constructor.call_count)
mock_args, _ = mock_runner_constructor.call_args
self.assertEqual(base_dir, mock_args[0])
self.assertEqual(mock_create_agent, mock_args[1])
def main(unused_argv):
"""Main method.
Args:
unused_argv: Arguments (unused).
"""
tf.logging.set_verbosity(tf.logging.INFO)
run_experiment.load_gin_configs(FLAGS.gin_files, FLAGS.gin_bindings)
runner = run_experiment.create_runner(FLAGS.base_dir)
runner.run_experiment()
def main(unused_argv):
"""Main method.
Args:
unused_argv: Arguments (unused).
"""
logging.set_verbosity(logging.INFO)
tf.compat.v1.disable_v2_behavior()
run_experiment.load_gin_configs(FLAGS.gin_files, FLAGS.gin_bindings)
runner = run_experiment.create_runner(FLAGS.base_dir)
runner.run_experiment()
def main(unused_argv):
"""Main method.
Args:
unused_argv: Arguments (unused).
"""
tf.logging.set_verbosity(tf.logging.INFO)
print("levin: test ok")
run_experiment.load_gin_configs(FLAGS.gin_files, FLAGS.gin_bindings)
FLAGS.base_dir = FLAGS.base_dir + '-' + str(int(time.time()))
runner = run_experiment.create_runner(FLAGS.base_dir)
runner.run_experiment()
RainbowAgent.epsilon_train = 0.01
RainbowAgent.epsilon_eval = 0.001
RainbowAgent.epsilon_decay_period = 250000 # agent steps
# IQN currently does not support prioritized replay.
RainbowAgent.replay_scheme = 'uniform'
RainbowAgent.tf_device = '/gpu:0' # '/cpu:*' use for non-GPU version
RainbowAgent.optimizer = @tf.train.AdamOptimizer()
tf.train.AdamOptimizer.learning_rate = 0.00005
tf.train.AdamOptimizer.epsilon = 0.0003125
atari_lib.create_atari_environment.game_name = 'Pong'
# Sticky actions with probability 0.25, as suggested by (Machado et al., 2017).
atari_lib.create_atari_environment.sticky_actions = True
create_agent.agent_name = 'implicit_quantile'
Runner.num_iterations = 50
Runner.training_steps = 1000
Runner.evaluation_steps = 1000
Runner.max_steps_per_episode = 200 # Default max episode length.
WrappedPrioritizedReplayBuffer.replay_capacity = 50000
WrappedPrioritizedReplayBuffer.batch_size = 128
"""
gin.parse_config(iqn_config, skip_unknown=False)
# @title Train IQN
iqn_runner = run_experiment.create_runner(IQN_PATH,
schedule='continuous_train')
print('Will train IQN agent, please be patient, may be a while...')
iqn_runner.run_experiment()
print('Done training!')
def testCreateRunnerUnknown(self):
base_dir = '/tmp'
with self.assertRaisesRegexp(ValueError, 'Unknown schedule'):
run_experiment.create_runner(base_dir, 'Unknown schedule')
#TrainRunner.create_environment_fn = @gym_lib.create_gym_environment
Runner.num_iterations = 100
Runner.training_steps = 1000
Runner.evaluation_steps = 1000
Runner.max_steps_per_episode = 1000 # Default max episode length.
WrappedReplayBuffer.replay_capacity = 100000
WrappedReplayBuffer.batch_size = 64
"""
gin.parse_config(dqn_config, skip_unknown=False)
for run in range(10):
DQN_PATH = os.path.join(BASE_PATH,
'llv2_standard_eval_decay/run_' + str(run))
dqn_runner = run_experiment.create_runner(
DQN_PATH, schedule='continuous_train_and_eval')
print('Will train DQN agent, please be patient, may be a while...')
dqn_runner.run_experiment()
print('Done training!')
data = colab_utils.read_experiment(
DQN_PATH,
verbose=True,
summary_keys=[
'train_episode_returns', 'train_episode_actual_returns',
'train_episode_lengths', 'eval_episode_returns',
'eval_episode_actual_returns', 'eval_episode_lengths'
])
data['agent'] = 'DQN'
data['run'] = run
from __future__ import absolute_import from __future__ import division from __future__ import print_function from dopamine.discrete_domains import run_experiment import tensorflow as tf gin_files = ['./my_tests/fourier_pong.gin'] base_dir = './tmp/fourier/' gin_bindings = [] tf.logging.set_verbosity(tf.logging.INFO) run_experiment.load_gin_configs(gin_files, gin_bindings) runner = run_experiment.create_runner(base_dir) runner.run_experiment()
