def nodir_test(self):
""" Fail if logdir does not exist """
parser = argparse.ArgumentParser()
parser.add_argument('--a', type=int, nargs='*')
parser.add_argument('--b', type=int)
parser = argload.ArgumentLoader(parser, ['a'])
with self.assertRaises(ValueError) as context:
parser.parse_args(['--logdir', 'log', '--a', '3', '4', '--b', '2'])
self.assertIn("Logdir does", str(context.exception))
def overwrite_default_test(self):
""" You can overwrite defaults, and it behaves properly """
try:
parser = argparse.ArgumentParser()
parser.add_argument('--a', type=float, default=1e-3)
parser.add_argument('--b', type=float, default=2.)
parser = argload.ArgumentLoader(parser, ['a', 'b'])
mkdir('log')
parser.parse_args(['--logdir', 'log', '--a', '1e-2', '--b', '3.'])
args = parser.parse_args(
['--logdir', 'log', '--a', '1e-3', '--overwrite'])
self.assertEqual(args.a, 1e-3)
self.assertEqual(args.b, 3.)
finally:
shutil.rmtree('log')
def dump_no_overwrite_test(self):
""" Fail if dump is used without overwrite """
try:
parser = argparse.ArgumentParser()
parser.add_argument('--a', type=int, nargs='*')
parser.add_argument('--b', type=int)
parser = argload.ArgumentLoader(parser, ['a'])
mkdir('log')
parser.parse_args(['--logdir', 'log', '--a', '3', '4', '--b', '2'])
with self.assertRaises(ValueError) as context:
parser.parse_args(
['--logdir', 'log', '--a', '2', '4', '--b', '2', '--dump'])
self.assertIn("Dumping is not", str(context.exception))
finally:
shutil.rmtree('log')
def failed_overwrite_test(self):
""" Trying to overwrite without overwrite flag """
try:
parser = argparse.ArgumentParser()
parser.add_argument('--a', type=int, nargs='*')
parser.add_argument('--b', type=int)
parser = argload.ArgumentLoader(parser, ['a'])
mkdir('log')
parser.parse_args(['--logdir', 'log', '--a', '3', '4', '--b', '2'])
with self.assertRaises(ValueError) as context:
parser.parse_args(['--logdir', 'log', '--a', '4'])
self.assertTrue("Overwritting" in str(context.exception))
finally:
shutil.rmtree('log')
def overwrite_no_old_test(self):
""" Fail if overwrite or dump without first dump """
try:
parser = argparse.ArgumentParser()
parser.add_argument('--a', type=int, nargs='*')
parser.add_argument('--b', type=int)
parser = argload.ArgumentLoader(parser, ['a'])
mkdir('log')
with self.assertRaises(ValueError) as context:
parser.parse_args([
'--logdir', 'log', '--a', '2', '4', '--b', '2',
'--overwrite'
])
self.assertIn("No old", str(context.exception))
finally:
shutil.rmtree('log')
def dump_test(self):
""" Dumping modifies stored values """
try:
parser = argparse.ArgumentParser()
parser.add_argument('--a', type=int, nargs='*')
parser.add_argument('--b', type=int)
parser = argload.ArgumentLoader(parser, ['a'])
mkdir('log')
parser.parse_args(['--logdir', 'log', '--a', '3', '4', '--b', '2'])
parser.parse_args(
['--logdir', 'log', '--overwrite', '--dump', '--a', '3', '2'])
args = parser.parse_args(['--logdir', 'log'])
self.assertEqual([3, 2], args.a)
finally:
shutil.rmtree('log')
def normal_test(self):
""" Test normal use case """
try:
parser = argparse.ArgumentParser()
parser.add_argument('--a', type=int, nargs='*')
parser.add_argument('--b', type=int)
parser = argload.ArgumentLoader(parser, ['a'])
mkdir('log')
parser.parse_args(['--logdir', 'log', '--a', '3', '5', '--b', '2'])
args = parser.parse_args(['--logdir', 'log'])
self.assertTrue(args.a == [3, 5])
self.assertTrue(args.b is None)
self.assertTrue(args.logdir == 'log')
args = parser.parse_args(
['--logdir', 'log', '--overwrite', '--a', '4'])
self.assertTrue(args.a == [4])
self.assertTrue(args.b is None)
finally:
shutil.rmtree('log')
def setup_args():
parser = argparse.ArgumentParser()
parser.add_argument('--eval_gap', type=float, default=.1,
help='evaluation is performed every .1/dt epochs.')
parser.add_argument('--algo', type=str, default='approximate_advantage',
help='algorithm used.')
parser.add_argument('--dt', type=float, default=.02,
help='temporal discretization.')
parser.add_argument('--steps_btw_train', type=int, default=10,
help='number of environment steps between two training periods.')
parser.add_argument('--env_id', type=str, default='pendulum',
help='environment.')
parser.add_argument('--noise_type', type=str, default='coherent', choices=['coherent', 'independent'],
help='noise type used')
parser.add_argument('--batch_size', type=int, default=64,
help='training batch size')
parser.add_argument('--hidden_size', type=int, default=64,
help='number of hidden units per layer.')
parser.add_argument('--nb_layers', type=int, default=1,
help='number of layers (careful, the "true number of layers" is this number + 1).')
parser.add_argument('--gamma', type=float, default=.8,
help='discount factor.')
parser.add_argument('--n_step', type=int, default=20,
help='Number of steps in a2c')
parser.add_argument('--nb_true_epochs', type=float, default=50,
help='number of true epochs (epochs / dt) to train on.')
parser.add_argument('--nb_steps', type=int, default=100,
help='number of environment steps in an epoch')
parser.add_argument('--sigma', type=float, default=1.5,
help='OU noise parameter.')
parser.add_argument('--theta', type=float, default=7.5,
help='OU stiffness parameter.')
parser.add_argument('--c_entropy', type=float, default=1e-4,
help='entropy regularization')
parser.add_argument('--eps_clamp', type=float, default=0.2,
help='Clipping value for PPO, epsilon in the original paper')
parser.add_argument('--c_kl', type=float, default=0.,
help='KL regularization for PPO, beta in the original paper')
parser.add_argument('--nb_train_env', type=int, default=32,
help='number of parallel environments during training.')
parser.add_argument('--nb_eval_env', type=int, default=16,
help='number of parallel environments used to evaluate.')
parser.add_argument('--memory_size', type=int, default=1000000,
help='size of the memory buffer.')
parser.add_argument('--learn_per_step', type=int, default=50,
help='number of gradient step in one learning step')
parser.add_argument('--normalize_state', action='store_true',
help='is state normalization used.')
parser.add_argument('--lr', type=float, default=.03,
help='critic learning rate.')
parser.add_argument('--policy_lr', type=float, default=None,
help='policy learning rate (for approximate policies).')
parser.add_argument('--time_limit', type=float, default=None,
help='specify environment time limite (physical time).')
parser.add_argument('--redirect_stdout', action='store_true',
help='should we redirect stdout to a log file?')
parser.add_argument('--weight_decay', type=float, default=0,
help='actor weight decay.')
parser.add_argument('--alpha', type=float, default=None,
help='prioritized replay buffer alpha (untested).')
parser.add_argument('--beta', type=float, default=None,
help='prioritized replay buffer beta (untested).')
parser.add_argument('--tau', type=float, default=.99,
help='target network update rate (works for all '
'algo, do not expect it to work with dau).')
parser.add_argument('--epsilon', type=float, default=.1,
help='epsilon greedy coeficient')
parser.add_argument('--noscale', action='store_true',
help='use unscaled ddpg when set')
parser.add_argument('--optimizer', type=str, choices=['sgd', 'rmsprop', 'adam'],
default='sgd')
parser.add_argument('--noreload', action='store_true',
help='do not reload previously saved model when set.')
parser.add_argument('--snapshot', action='store_true',
help='if true, stores snapshot every once in a while')
parser = argload.ArgumentLoader(parser, to_reload=[
'algo', 'dt', 'steps_btw_train', 'env_id', 'noise_type',
'batch_size', 'hidden_size', 'nb_layers', 'gamma', 'n_step', 'nb_true_epochs', 'nb_steps',
'sigma', 'theta', 'c_entropy', 'eps_clamp', 'c_kl', 'nb_train_env', 'nb_eval_env', 'memory_size',
'learn_per_step', 'normalize_state', 'lr', 'time_limit',
'policy_lr', 'alpha', 'beta', 'weight_decay', 'optimizer',
'tau', 'eval_gap', 'noscale', 'epsilon', 'snapshot'
])
args = parser.parse_args()
# args translation
if args.algo == 'ddpg':
args.algo = 'approximate_value'
elif args.algo == 'dqn':
args.algo = 'discrete_value'
elif args.algo == 'ddau':
args.algo = 'discrete_advantage'
elif args.algo == 'cdau':
args.algo = 'approximate_advantage'
return args