def test_load_config(self):
# Test that load config returns a config file and fails the config check
# when it should.
cfg = model_trainer.load_config(self.root_path + 'test.json')
del cfg['validation_params']
temp_cfg_path = self.root_path + 'temp.json'
with open(temp_cfg_path, 'w') as json_f:
json.dump(cfg, json_f, indent=4)
with self.assertRaises(RuntimeError):
model_trainer.load_config(temp_cfg_path)
os.remove(temp_cfg_path)
def setUp(self):
# Open up the config file.
# Initialize the class with a test baobab config
self.root_path = os.path.dirname(
os.path.abspath(__file__)) + '/test_data/'
self.config_path = self.root_path + 'test.json'
self.baobab_cfg_path = self.root_path + 'test_baobab_cfg.py'
# Initialize the config
self.cfg = model_trainer.load_config(self.config_path)
# Also initialize the baobab config
self.baobab_cfg = configs.BaobabConfig.from_file(self.baobab_cfg_path)
# A few bnn_inference testing things that need to be used again here.
self.lens_params = [
'external_shear_gamma_ext', 'external_shear_psi_ext',
'lens_mass_center_x', 'lens_mass_center_y', 'lens_mass_e1',
'lens_mass_e2', 'lens_mass_gamma', 'lens_mass_theta_E'
]
self.normalization_constants_path = self.root_path + 'norms.csv'
self.tf_record_path = self.root_path + 'tf_record_test'
# Create the normalization file that would have been made
# during training.
self.final_params = [
'external_shear_g1', 'external_shear_g2', 'lens_mass_center_x',
'lens_mass_center_y', 'lens_mass_e1', 'lens_mass_e2',
'lens_mass_gamma', 'lens_mass_theta_E_log'
]
model_trainer.prepare_tf_record(self.cfg, self.root_path,
self.tf_record_path, self.final_params,
'train')
def test_get_normed_pixel_scale(self):
# Test if get_normed_pixel scale rescales the pixel_scale as we would
# expect.
cfg = model_trainer.load_config(self.root_path + 'test.json')
# The original pixel scale
pixel_scale = 0.051
# Test if normalizing the lens parameters works correctly.
normalized_param_path = self.root_path + 'normed_metadata.csv'
normalization_constants_path = self.root_path + 'norms.csv'
train_or_test = 'train'
lens_params = [
'external_shear_gamma_ext', 'external_shear_psi_ext',
'lens_mass_center_x', 'lens_mass_center_y', 'lens_mass_e1',
'lens_mass_e2', 'lens_mass_gamma', 'lens_mass_theta_E'
]
lens_params_path = self.root_path + 'metadata.csv'
data_tools.normalize_lens_parameters(lens_params,
lens_params_path,
normalized_param_path,
normalization_constants_path,
train_or_test=train_or_test)
# New pixel scale
normed_pixel_scale = model_trainer.get_normed_pixel_scale(
cfg, pixel_scale)
lens_params_csv = pd.read_csv(lens_params_path, index_col=None)
norm_params_csv = pd.read_csv(normalized_param_path, index_col=None)
self.assertAlmostEqual(
np.std(lens_params_csv['lens_mass_center_x'] / pixel_scale -
norm_params_csv['lens_mass_center_x'] /
normed_pixel_scale['lens_mass_center_x']), 0)
self.assertAlmostEqual(
np.std(lens_params_csv['lens_mass_center_y'] / pixel_scale -
norm_params_csv['lens_mass_center_y'] /
normed_pixel_scale['lens_mass_center_y']), 0)
# Clean up the file now that we're done
os.remove(normalized_param_path)
os.remove(normalization_constants_path)
def test_model_loss_builder_diag(self):
# Test that the model and loss returned from model_loss_builder
# agree with what is expected.
cfg = model_trainer.load_config(self.root_path + 'test.json')
cfg['training_params']['dropout_type'] = 'concrete'
final_params = cfg['training_params']['final_params']
num_params = len(final_params)
tf.keras.backend.clear_session()
gc.collect()
cfg['training_params']['bnn_type'] = 'diag'
model, loss = model_trainer.model_loss_builder(cfg)
y_true = np.ones((1, num_params))
y_pred = np.ones((1, 2 * num_params))
yptf = tf.constant(y_pred, dtype=tf.float32)
yttf = tf.constant(y_true, dtype=tf.float32)
# Check that the loss function has the right dimensions. More rigerous
# tests of the loss function can be found in the test_bnn_alexnet.
loss(yttf, yptf)
self.assertEqual(len(model.layers), 13)
self.assertEqual(model.layers[-1].output_shape[-1], y_pred.shape[-1])
# First specify the config path
root_path = os.getcwd()[:-20]
config_path = args.config_path
# We also need the path to the baobab configs for the interim and target omega
interim_baobab_omega_path = os.path.join(
root_path, 'configs/baobab_configs/train_diagonal.py')
target_ovejero_omega_path = args.target_ovejero_omega_path
target_baobab_omega_path = args.target_baobab_omega_path
test_dataset_path = args.test_dataset_path
test_dataset_tf_record_path = args.test_dataset_tf_record_path
# Check that the config has what you need
cfg = model_trainer.load_config(config_path)
# If we're using the emprical config, we need the transformation dictionary
if 'empirical' in test_dataset_path:
train_to_test_param_map = dict(
orig_params=['lens_mass_e1', 'lens_mass_e2'],
transform_func=ellipticity2phi_q,
new_params=['lens_mass_phi', 'lens_mass_q'])
n_walkers = 200
else:
train_to_test_param_map = None
n_walkers = 50
# Correct any path issues.
def recursive_str_checker(cfg_dict):