game = '2h2o-v0' gen = 1 data = np.load('./champions/' + game + '/' + game + '_' + str(gen) + '.npz') cpus = 4 env = gym.make(game) s0 = env.reset() shape = s0.shape[0] num_actions = env.action_space.shape[0] a_bound = [env.action_space.low, env.action_space.high] hidden_units = data['h'] champion = Policy(shape, hidden_units, num_actions, a_bound, game) champion.W = data['w'] champion.B = data['b'] pool = Pool(processes=cpus) champions = [] for k in range(5): champions.append(champion) scores = pool.map(evaluate_policy_single, champions) scores = np.array(scores) score = np.mean(score) print('Champion Average Score = ' + str(score)) vis_policy(champion)