def __init__(self, cell_nb, lr=4e-3, nb_blocks=5, gamma=0.99):
self.cell_nb = cell_nb
self.gamma = gamma
self.ActorCritic = ActorCritic(lr, cell_nb**2, nb_blocks)
self.log_probs = None
env = gym.make(env_id)
env.seed(seed + rank)
return env
set_global_seeds(seed)
return _init
if __name__ == "__main__":
envs = SubprocVecEnv([make_env(env_name, i) for i in range(num_envs)])
env = gym.make(env_name)
num_inputs = envs.observation_space.shape
num_outputs = envs.action_space.shape
model = ActorCritic(num_inputs[0], num_outputs[0]).to(device)
if os.path.isfile(modelpath):
model.load_state_dict(torch.load(modelpath))
ppo = PPO(model=model,
envs=envs,
device=device,
lr=lr,
modelpath=modelpath)
if not play_mode:
ppo.ppo_train(num_steps,
mini_batch_size,
ppo_epochs,
max_frames,
max_pol_updates,
save_interval,
return env
set_global_seeds(seed)
return _init
if __name__ == "__main__":
envs = SubprocVecEnv([make_env(env_name, i) for i in range(num_envs)])
env = gym.make(env_name)
img_size = envs.observation_space[0].shape
sensor_size = envs.observation_space[1].shape
num_outputs = envs.action_space.shape
model = ActorCritic([img_size[1], img_size[0]], sensor_size[0],
num_outputs[0]).to(device)
if args.onnx_converter and os.path.isfile(modelpath):
model.load_state_dict(torch.load(modelpath))
model.export("gvsets_early_fusion.onnx")
exit(1)
if os.path.isfile(modelpath):
model.load_state_dict(torch.load(modelpath))
ppo = PPO(model=model,
envs=envs,
device=device,
lr=lr,
modelpath=modelpath,
tuple_ob=True)