def train_and_save_svr_dynamics(train_file_base_name, svr_output_path,
delta_t, n, m, epsilons, feature_names, target_names, xtra_names,
kernel_type='LINEAR',
kernel_params = {}):
# Read data from disk
print 'Reading training file:', train_file_base_name
(train_X, train_Y) = dynamics_learning.read_dynamics_learning_example_files(train_file_base_name)
# print 'train_X:\n',train_X
# print 'train_Y:\n',train_Y
print 'len(train_X)',len(train_X)
print 'len(train_Y)',len(train_Y)
# Build model class
if _LEARN_SVR_ERR_DYNAMICS:
svr_dynamics = SVRWithNaiveLinearPushDynamics(delta_t, n, m, epsilons=epsilons)
svr_dynamics.learn_model(train_X, train_Y, kernel_params)
elif _LEARN_GP_DYNAMICS:
svr_dynamics = GPPushDynamics(delta_t, n, m,
feature_names = feature_names,
target_names = target_names,
xtra_names = xtra_names,
mean_fnc = kernel_type.lower())
svr_dynamics.learn_model(train_X, train_Y)
else:
svr_dynamics = SVRPushDynamics(delta_t, n, m, epsilons=epsilons,
feature_names = feature_names,
target_names = target_names,
xtra_names = xtra_names,
kernel_type = kernel_type)
# Train and save model
svr_dynamics.learn_model(train_X, train_Y, kernel_params)
print 'Saving model to:', svr_output_path
svr_dynamics.save_models(svr_output_path)
return svr_dynamics
def test_svr_offline(model_param_file_name, test_data_example_base_name):
delta_t = 1./9.
n = 6
m = 3
if _LEARN_SVR_ERR_DYNAMICS:
svr_dynamics = SVRWithNaiveLinearPushDynamics(delta_t, n, m, param_file_name = model_param_file_name)
elif _LEARN_GP_DYNAMICS:
svr_dynamics = GPPushDynamics(param_file_name = model_param_file_name)
else:
svr_dynamics = SVRPushDynamics(param_file_name = model_param_file_name)
(test_X, test_Y) = dynamics_learning.read_dynamics_learning_example_files(test_data_example_base_name)
# Run batch testing
(Y_hat, Y_gt, _) = svr_dynamics.test_batch_data(test_X, test_Y)
return (Y_hat, Y_gt)
