# Deep Deterministic Policy Gradient (DDPG)
print("Training algorithm: Deep Deterministic Policy Gradient (DDPG)")
trainer = DDPGTrainer(
env=env_title,
config={
"num_workers": num_workers,
"num_cpus_per_worker": num_cpus_per_worker,
"num_gpus": num_gpus,
"num_gpus_per_worker": num_gpus_per_worker,
"model": nw_model,
"lr": lr,
"gamma": gamma,
"actor_hiddens": [hidden_layer_size, hidden_layer_size],
"critic_hiddens": [hidden_layer_size, hidden_layer_size],
"multiagent": {
"policy_graphs": policy_graphs,
"policy_mapping_fn": policy_mapping_fn,
"policies_to_train": ["agent_policy{}".format(i) for i in range(n_agents)],
},
"callbacks": {
"on_episode_start": tune.function(on_episode_start),
"on_episode_step": tune.function(on_episode_step),
"on_episode_end": tune.function(on_episode_end),
},
"log_level": "ERROR",
})
elif(train_algo == "A3C"):
net_file=
'/home/sonic/Desktop/sumo-rl-research-offset/sumo-rl-research/experiments/nets/Research/case04/intersection.net.xml',
route_file=
'/home/sonic/Desktop/sumo-rl-research-offset/sumo-rl-research/experiments/nets/Research/case04/intersection.rou.xml',
out_csv_path='outputs/case04/',
out_csv_name='DDPG',
use_gui=False,
num_seconds=12240612,
time_to_load_vehicles=612,
max_depart_delay=0))
trainer = DDPGTrainer(
env="2TLS",
config={
"multiagent": {
"policy_graphs": {
'offset_agent': (DDPGTFPolicy,
spaces.Box(low=np.zeros(2),
high=np.array(['inf'] * 2)),
spaces.Box(low=np.array([0, 0]),
high=np.array([+1, +1])), {})
},
"policy_mapping_fn":
policy_mapping # Traffic lights are always controlled by this policy
},
"lr": 0.0001,
})
while True:
result = trainer.train()
# /home/sonic/Desktop/sumo-rl-research-offset/sumo-rl-research/experiments/