def setUp(self):
network_args = {
'network_id': 'grid',
'horizon': 999,
'demand_type': 'constant',
'tls_type': 'rl'
}
self.network = Network(**network_args)
def get_lanes_lengths(train_args):
network_args = {
'network_id': train_args['network'],
'horizon': int(train_args['experiment_time']),
'demand_type': train_args['demand_type'],
'demand_mode': train_args['demand_mode'],
'tls_type': train_args['tls_type']
}
network = Network(**network_args)
lanes_lengths = {
(edge['id'], int(lane['index'])): float(lane['length'])
for edge in network.edges
for lane in sorted(edge['lanes'], key=lambda x: int(x['index']))
}
return lanes_lengths
def setUp(self):
network_args = {
'network_id': 'grid',
'horizon': 999,
'demand_type': 'constant',
'tls_type': 'rl'
}
self.network = Network(**network_args)
# with open('tests/unit/data/grid_kernel_data.dat', "rb") as f:
# kernel_data = pickle.load(f)
# self.kernel_data = kernel_data
with open('tests/unit/data/grid_kernel_data_new_1.dat', "rb") as f:
kernel_data_1 = pickle.load(f)
self.kernel_data_1 = kernel_data_1
with open('tests/unit/data/grid_kernel_data_new_2.dat', "rb") as f:
kernel_data_2 = pickle.load(f)
self.kernel_data_2 = kernel_data_2
def main(train_config_path=None):
# Setup config parser path.
if train_config_path is not None:
print(f'Loading train parameters from: {train_config_path}')
config_parser.set_config_path(train_config_path)
else:
print('Loading train parameters from: configs/train.config [DEFAULT]')
# Parse train parameters.
train_args = config_parser.parse_train_params(print_params=True)
network_args = {
'network_id': train_args.network,
'horizon': train_args.experiment_time,
'demand_type': train_args.demand_type,
'demand_mode': train_args.demand_mode,
'tls_type': train_args.tls_type
}
network = Network(**network_args)
# Create directory to store data.
experiment_path = EMISSION_PATH / network.name
os.makedirs(experiment_path, exist_ok=True)
print(f'Experiment: {str(experiment_path)}\n')
sumo_args = {
'render': train_args.sumo_render,
'print_warnings': False,
'sim_step': 1, # Do not change.
'restart_instance': True,
'teleport_time': 120
}
# Setup seeds.
if train_args.experiment_seed is not None:
random.seed(train_args.experiment_seed)
np.random.seed(train_args.experiment_seed)
sumo_args['seed'] = train_args.experiment_seed
# Setup emission path.
if train_args.sumo_emission:
sumo_args['emission_path'] = experiment_path.as_posix()
sim_params = SumoParams(**sumo_args)
additional_params = {}
additional_params['tls_type'] = train_args.tls_type
env_args = {
'evaluate': True,
'additional_params': additional_params
}
env_params = EnvParams(**env_args)
# Load MDP parameters from file (train.config[mdg_args]).
mdp_params = config_parser.parse_mdp_params(network.tls_ids)
env = TrafficLightEnv(
env_params=env_params,
sim_params=sim_params,
mdp_params=mdp_params,
network=network,
exp_path=experiment_path.as_posix(),
seed=train_args.experiment_seed,
)
exp = Experiment(
env=env,
exp_path=experiment_path.as_posix(),
train=True,
save_agent=train_args.experiment_save_agent,
save_agent_interval=train_args.experiment_save_agent_interval,
tls_type=train_args.tls_type
)
# Store train parameters (config file).
config_parser.store_config(experiment_path / 'config')
# Store a copy of the tls_config.json file.
tls_config_path = NETWORKS_PATH / train_args.network / 'tls_config.json'
copyfile(tls_config_path, experiment_path / 'tls_config.json')
# Store a copy of the demands.json file.
demands_file_path = NETWORKS_PATH / train_args.network / 'demands.json'
copyfile(demands_file_path, experiment_path / 'demands.json')
# Run the experiment.
info_dict = exp.run(train_args.experiment_time)
# Store train info dict.
os.makedirs(experiment_path/ 'logs', exist_ok=True)
train_log_path = experiment_path / 'logs' / "train_log.json"
with train_log_path.open('w') as f:
json.dump(info_dict, f)
return str(experiment_path)
def main(config_file_path=None):
args = get_arguments(config_file_path)
print_arguments(args)
print('Loading from train run: {0}\n'.format(args.run_path))
# Setup parser with custom path (load correct train parameters).
config_path = Path(args.run_path) / 'config' / 'train.config'
config_parser.set_config_path(config_path)
# Parse train parameters.
train_args = config_parser.parse_train_params(print_params=False)
network_args = {
'network_id': train_args.network,
'horizon': args.rollout_time,
'demand_type': train_args.demand_type,
'demand_mode': train_args.demand_mode,
'tls_type': train_args.tls_type
}
network = Network(**network_args)
# Create directory to store evaluation script data.
experiment_path = Path(args.run_path) / 'eval' / f'{network.name}'
os.makedirs(experiment_path, exist_ok=True)
print(f'Experiment: {str(experiment_path)}')
sumo_args = {
'render': args.sumo_render,
'print_warnings': False,
'sim_step': 1, # Do not change.
'restart_instance': True,
'teleport_time': 120
}
# Setup seeds.
if args.seed is not None:
random.seed(args.seed)
np.random.seed(args.seed)
sumo_args['seed'] = args.seed
# Setup emission path.
if args.sumo_emission:
sumo_args['emission_path'] = experiment_path.as_posix()
sim_params = SumoParams(**sumo_args)
additional_params = {}
additional_params['tls_type'] = train_args.tls_type
env_args = {
'evaluate': True,
'additional_params': additional_params
}
env_params = EnvParams(**env_args)
# Load MDP parameters from file (train.config[mdg_args]).
mdp_params = config_parser.parse_mdp_params(network.tls_ids)
env = TrafficLightEnv(
env_params=env_params,
sim_params=sim_params,
mdp_params=mdp_params,
network=network,
exp_path=experiment_path.as_posix(),
seed=args.seed
)
# Setup checkpoints.
checkpoints_dir_path = Path(args.run_path) / 'checkpoints'
env.tsc.load_checkpoint(checkpoints_dir_path, args.chkpt_number)
# Stop training.
env.stop = True
exp = Experiment(
env=env,
exp_path=experiment_path.as_posix(),
train=False, # Stop training.
save_agent=False
)
# Run the experiment.
info_dict = exp.run(args.rollout_time)
info_dict['id'] = args.chkpt_number
return info_dict
def main(preprocess_config_path=None):
# Setup config parser path.
if preprocess_config_path is not None:
print(f'Loading train parameters from: {preprocess_config_path}')
config_parser.set_config_path(preprocess_config_path)
else:
print('Loading train parameters from: configs/train.config [DEFAULT]')
# Parse pre-process parameters.
# Since there is not a preprocess file we just use config/train.config
preprocess_args = config_parser.parse_train_params(print_params=True)
network_args = {
'network_id': preprocess_args.network,
'horizon': preprocess_args.experiment_time,
'demand_type': preprocess_args.demand_type,
'demand_mode': preprocess_args.demand_mode,
'tls_type': preprocess_args.tls_type
}
network = Network(**network_args)
# Create directory to store data.
# experiment_path = EMISSION_PATH / network.name
# os.makedirs(experiment_path, exist_ok=True)
# print(f'Experiment: {str(experiment_path)}\n')
sumo_args = {
'render': preprocess_args.sumo_render,
'print_warnings': False,
'sim_step': 1, # Do not change.
'restart_instance': True,
'teleport_time': 120
}
# Setup seeds.
if preprocess_args.experiment_seed is not None:
random.seed(preprocess_args.experiment_seed)
np.random.seed(preprocess_args.experiment_seed)
sumo_args['seed'] = preprocess_args.experiment_seed
sim_params = SumoParams(**sumo_args)
additional_params = {}
additional_params['tls_type'] = preprocess_args.tls_type
env_args = {'evaluate': True, 'additional_params': additional_params}
env_params = EnvParams(**env_args)
# Load MDP parameters from file (train.config[mdg_args]).
mdp_params = config_parser.parse_mdp_params(network.tls_ids)
env = TrafficLightEnv(
env_params=env_params,
sim_params=sim_params,
mdp_params=mdp_params,
network=network,
exp_path=None,
seed=preprocess_args.experiment_seed,
)
exp = Experiment(
env=env,
train=True,
exp_path=None,
save_agent=preprocess_args.experiment_save_agent,
save_agent_interval=preprocess_args.experiment_save_agent_interval,
tls_type=preprocess_args.tls_type)
# Run the experiment.
info_dict = exp.run(preprocess_args.experiment_time)
return info_dict