def test_tree(env_path,
tree_pickle,
itr,
deterministic,
len,
episodes,
render=False):
env, _, _ = load_policy(env_path, itr, deterministic)
tree_data = joblib.load(tree_pickle)
tree_program = tree_data['tree']
label_data(env, lambda a: tree_program.predict([a]), len, episodes, render)
parser.add_argument('--deterministic', '-d', action='store_true')
parser.add_argument('--seed', type=int, default=301)
parser.add_argument('--save_file', type=str, default='')
parser.add_argument('--idProcess', type=int, default='0')
parser.add_argument('--num_env_interact', type=int, default=1000000)
parser.add_argument('--steps_per_epoch', type=int, default=30000)
parser.add_argument('--max_ep_len', type=int, default=1000)
args = parser.parse_args()
# Set the seed for reproducibility
tf.set_random_seed(args.seed)
np.random.seed(args.seed)
# Get the policies
env, get_action, sess = load_policy(args.fpath,
args.itr if args.itr >= 0 else 'last',
args.deterministic)
# Set the seed of the environment
print(args.seed, args.idProcess)
env.seed(args.seed + args.idProcess)
# Test the given policy
run_policy(env,
get_action,
render=args.render,
num_env_interact=args.num_env_interact,
steps_per_epoch=args.steps_per_epoch,
max_ep_len=args.max_ep_len,
save_file=args.save_file,
idProcess=args.idProcess)
def load(load_folder, env):
load_folder = str(load_folder)
gym_env, get_action, sess = load_policy(load_folder)
net = MultilayerPerceptron(env, get_action)
return net, gym_env