def test_ddpg_workflow(self):
"""Run DDPG workflow to ensure no crashes, algorithm correctness
not tested here."""
params = {
"training_data_path":
os.path.join(
curr_dir,
"test_data/continuous_action/pendulum_training_data.json"),
"state_norm_data_path":
os.path.join(
curr_dir,
"test_data/continuous_action/state_features_norm.json"),
"action_norm_data_path":
os.path.join(curr_dir,
"test_data/continuous_action/action_norm.json"),
"model_output_path":
None,
"use_gpu":
False,
"epochs":
1,
"rl": {},
"rainbow": {},
"shared_training": {
"minibatch_size": 16
},
"actor_training": {},
"critic_training": {},
}
predictor = ddpg_workflow.train_network(params)
test_float_state_features = [{"0": 1.0, "1": 1.0, "2": 1.0, "3": 1.0}]
test_int_state_features = [{}]
action = predictor.actor_prediction(test_float_state_features,
test_int_state_features)
assert len(action) == 1
def _test_ddpg_workflow(self, use_gpu=False, use_all_avail_gpus=False):
"""Run DDPG workflow to ensure no crashes, algorithm correctness
not tested here."""
with tempfile.TemporaryDirectory() as tmpdirname:
params = {
"training_data_path": os.path.join(
curr_dir, "test_data/continuous_action/pendulum_training.json.bz2"
),
"eval_data_path": os.path.join(
curr_dir, "test_data/continuous_action/pendulum_eval.json.bz2"
),
"state_norm_data_path": os.path.join(
curr_dir, "test_data/continuous_action/state_features_norm.json"
),
"action_norm_data_path": os.path.join(
curr_dir, "test_data/continuous_action/action_norm.json"
),
"model_output_path": tmpdirname,
"use_gpu": use_gpu,
"use_all_avail_gpus": use_all_avail_gpus,
"epochs": 1,
"rl": {},
"rainbow": {},
"shared_training": {"minibatch_size": 128},
"actor_training": {},
"critic_training": {},
}
predictor = ddpg_workflow.train_network(params)
test_float_state_features = [{"0": 1.0, "1": 1.0, "2": 1.0, "3": 1.0}]
test_int_state_features = [{}]
action = predictor.actor_prediction(
test_float_state_features, test_int_state_features
)
assert len(action) == 1