def setUp(self):
self.env = GymEnv('Pendulum-v0')
self.env = SkillEnv(self.env, num_skill=4)
Feature extractor of the discriminator.
please see the paper 4.4.2 question 7.
e.g.:
def discrim_f(x): return x
f_dim = 9 # dimension of feature space
'''
def discrim_f(x): return x[:, 0:2]+x[:, 2:4]
f_dim = 2
env = gym.make(args.env_name)
env = SkillEnv(env, num_skill=args.num_skill)
obs = env.reset()
observation_space = env.real_observation_space
skill_space = env.skill_space
ob_skill_space = env.observation_space
action_space = env.action_space
ob_dim = ob_skill_space.shape[0] - args.num_skill
device_name = 'cpu' if args.cuda < 0 else "cuda:{}".format(args.cuda)
device = torch.device(device_name)
set_device(device)
# policy
pol_net = PolNet(ob_skill_space, action_space)
pol = GaussianPol(ob_skill_space, action_space, pol_net,
data_parallel=args.data_parallel, parallel_dim=0)