def test_icm_cmaes(encoder, inverse_model, forward_model, env, policy, sess): policy = CMAESPolicy(env.spec, encoder, inverse_model, forward_model, sess=sess) o = env.reset() while True: a, _ = policy.get_action([o], env=env) next_o, r, d, env_info = env.step(a) if d: o = env.reset() else: o = next_o env.render() time.sleep(0.05)
pass # else: # raise ValueError("Unsupported snapshot!") env = data['env'] if args.test_state_hist: test_state_hist(env) elif args.get_max_reward: get_max_reward(env, policy, num_trajs=200) elif args.icm: if args.cmaes: from railrl.policies.cmaes_icm import CMAESPolicy policy = CMAESPolicy(env.spec, encoder, inverse_model, forward_model, sess=sess) test_icm_cmaes(encoder, inverse_model, forward_model, env, policy, sess) elif args.test_inverse_loss: investigate_inverse_loss(encoder, inverse_model, forward_model, env, policy, sess, img_path=args.data_path, num_trajs=100, animate=args.render) elif args.test_forward_loss: