def objective(params):
config = {'model_name': model_name, 'num_epochs': num_epochs}
for (key, value) in params.items():
config[key] = value
print("Parameter: {} | Value: {}".format(key, value))
run = experiment.run(config_updates=config)
return run.result
def test_train_cartpole_td3_tdg(self):
from main import ex
config = dict(env_name='GYMMB_CartPole-v1',
agent_alg='td3',
tdg_error_weight=5,
dump_dir=None,
print_config=False,
seed=123,
n_total_steps=200,
n_warm_up_steps=50,
model_training_n_batches=10,
policy_training_n_iters=5,
policy_actors=32,
eval_freq=100,
n_eval_episodes_per_policy=1,
neptune_project=None)
ret, abs_action = ex.run('train', config_updates=config).result
self.assertAlmostEqual(ret, 42.)
self.assertAlmostEqual(abs_action, 0.39846721291542053)
def test_train_test_ddpg(self):
from main import ex
config = dict(env_name='GYMMB_Test-v0',
agent_alg='ddpg',
tdg_error_weight=0,
dump_dir=None,
print_config=False,
seed=123,
n_total_steps=200,
n_warm_up_steps=50,
model_training_n_batches=50,
policy_training_n_iters=5,
policy_actors=32,
eval_freq=100,
n_eval_episodes_per_policy=1,
neptune_project=None)
ret, abs_action = ex.run('train', config_updates=config).result
self.assertAlmostEqual(ret, -0.02898537367582321)
self.assertAlmostEqual(abs_action, 0.06166519597172737)