def test_nested_action_spaces(self):
config = DEFAULT_CONFIG.copy()
config["env"] = RandomEnv
# Write output to check, whether actions are written correctly.
tmp_dir = os.popen("mktemp -d").read()[:-1]
if not os.path.exists(tmp_dir):
# Last resort: Resolve via underlying tempdir (and cut tmp_.
tmp_dir = ray._private.utils.tempfile.gettempdir() + tmp_dir[4:]
assert os.path.exists(tmp_dir), f"'{tmp_dir}' not found!"
config["output"] = tmp_dir
# Switch off OPE as we don't write action-probs.
# TODO: We should probably always write those if `output` is given.
config["input_evaluation"] = []
# Pretend actions in offline files are already normalized.
config["actions_in_input_normalized"] = True
for _ in framework_iterator(config):
for name, action_space in SPACES.items():
config["env_config"] = {
"action_space": action_space,
}
for flatten in [False, True]:
print(f"A={action_space} flatten={flatten}")
shutil.rmtree(config["output"])
config["_disable_action_flattening"] = not flatten
trainer = PGTrainer(config)
trainer.train()
trainer.stop()
# Check actions in output file (whether properly flattened
# or not).
reader = JsonReader(
inputs=config["output"],
ioctx=trainer.workers.local_worker().io_context,
)
sample_batch = reader.next()
if flatten:
assert isinstance(sample_batch["actions"], np.ndarray)
assert len(sample_batch["actions"].shape) == 2
assert sample_batch["actions"].shape[0] == len(
sample_batch)
else:
tree.assert_same_structure(
trainer.get_policy().action_space_struct,
sample_batch["actions"],
)
# Test, whether offline data can be properly read by a
# BCTrainer, configured accordingly.
config["input"] = config["output"]
del config["output"]
bc_trainer = BCTrainer(config=config)
bc_trainer.train()
bc_trainer.stop()
config["output"] = tmp_dir
config["input"] = "sampler"
MARWIL_agent = MARWILTrainer(config=marwil_config, env=SSA_Tasker_Env)
MARWIL_agent.restore(marwil_checkpoint)
MARWIL_agent.get_policy().config['explore'] = False
pg_config = PG_CONFIG.copy()
pg_config['batch_mode'] = 'complete_episodes'
pg_config['train_batch_size'] = 2000
pg_config['lr'] = 0.0001
pg_config['evaluation_interval'] = None
pg_config['postprocess_inputs'] = True
pg_config['env_config'] = env_config
pg_config['explore'] = False
PGR_agent = PGTrainer(config=pg_config, env=SSA_Tasker_Env)
PGR_agent.restore(pgr_checkpoint)
PGR_agent.get_policy().config['explore'] = False
PGRE_agent = PGTrainer(config=pg_config, env=SSA_Tasker_Env)
PGRE_agent.restore(pgre_checkpoint)
PGRE_agent.get_policy().config['explore'] = False
OLR_agent = PGTrainer(config=pg_config, env=SSA_Tasker_Env)
OLR_agent.restore(olr_checkpoint)
OLR_agent.get_policy().config['explore'] = False
def ppo_agent(obs, env):
return PPO_agent.compute_action(obs)
def marwil_agent(obs, env):
MARWIL_trainer.restore(
'/home/ash/ray_results/MARWIL_SSA_Tasker_Env_2020-09-10_08-24-14dv96mkld/checkpoint_5000/checkpoint-5000'
) # 20 SSA Complete
elif env_config['rso_count'] == 40:
MARWIL_trainer.restore(
'/home/ash/ray_results/MARWIL_SSA_Tasker_Env_2020-09-10_08-46-52wmigl7hj/checkpoint_10000/checkpoint-10000'
) # 40 SSA Complete
else:
print(str(env_config['rso_count']) + ' is not a valid number of RSOs')
with MARWIL_trainer.get_policy()._sess.graph.as_default():
with MARWIL_trainer.get_policy()._sess.as_default():
MARWIL_trainer.get_policy().model.base_model.save_weights(
"/tmp/pgr/weights3.h5")
with PG_trainer.get_policy()._sess.graph.as_default():
with PG_trainer.get_policy()._sess.as_default():
PG_trainer.get_policy().model.base_model.load_weights(
"/tmp/pgr/weights3.h5")
result = {'timesteps_total': 0}
num_steps_train = 5000000
best_athlete = 480
episode_len_mean = []
episode_reward_mean = []
episode_reward_max = []
episode_reward_min = []
start = datetime.datetime.now()
num_steps_trained = []
clock_time = []
training_iteration = []
