def test_load_from_auto_save():
import pickle
path_to_dict = gps_path + '/gps/algorithm/policy_opt/tf_checkpoint/policy_checkpoint.ckpt_hyperparams'
state = pickle.load(open(path_to_dict, "rb"))
hyper_params = state['hyperparams']
deg_obs = state['dO']
deg_action = state['dU']
policy_opt = PolicyOptTf(hyper_params, deg_obs, deg_action)
policy_opt.__setstate__(state)
def test_load_from_auto_save():
import pickle
path_to_dict = gps_path + '/gps/algorithm/policy_opt/tf_checkpoint/policy_checkpoint.ckpt_hyperparams'
state = pickle.load(open(path_to_dict, "rb"))
hyper_params = state['hyperparams']
deg_obs = state['dO']
deg_action = state['dU']
policy_opt = PolicyOptTf(hyper_params, deg_obs, deg_action)
policy_opt.__setstate__(state)
def test_unpickle():
hyper_params = POLICY_OPT_TF
deg_obs = 14
deg_action = 7
policy_opt = PolicyOptTf(hyper_params, deg_obs, deg_action)
N = 20
T = 30
obs = np.random.randn(N, T, deg_obs)
obs_reshaped = np.reshape(obs, (N*T, deg_obs))
scale = np.diag(1.0 / np.std(obs_reshaped, axis=0))
bias = -np.mean(obs_reshaped.dot(scale), axis=0)
hyper_params['scale'] = scale
hyper_params['bias'] = bias
hyper_params['tf_iter'] = 100
policy_opt.__setstate__({'hyperparams': hyper_params, 'dO': deg_obs, 'dU': deg_action,
'scale': policy_opt.policy.scale, 'bias': policy_opt.policy.bias, 'tf_iter': 100})
def test_unpickle():
hyper_params = POLICY_OPT_TF
deg_obs = 14
deg_action = 7
policy_opt = PolicyOptTf(hyper_params, deg_obs, deg_action)
N = 20
T = 30
obs = np.random.randn(N, T, deg_obs)
obs_reshaped = np.reshape(obs, (N*T, deg_obs))
scale = np.diag(1.0 / np.std(obs_reshaped, axis=0))
bias = -np.mean(obs_reshaped.dot(scale), axis=0)
hyper_params['scale'] = scale
hyper_params['bias'] = bias
hyper_params['tf_iter'] = 100
policy_opt.__setstate__({'hyperparams': hyper_params, 'dO': deg_obs, 'dU': deg_action,
'scale': policy_opt.policy.scale, 'bias': policy_opt.policy.bias, 'tf_iter': 100})