def test_that_the_output_is_of_the_correct_scale(self):
# create random weights
self.make_model()
self.weights_path = os.path.join(self.weights_folder,
'testweights_E001_.hdf5')
self.rdn.save_weights(self.weights_path)
# test the network with random weights
self.create_random_dataset('correct', dataset_size=1)
cl_args = ['--no_verbose', '--test']
test_args = {
'test_folder': self.dataset_folder['correct']['LR'],
'results_folder': self.results_folder,
'weights_path': self.weights_path,
}
parser = get_parser()
cl_args = parser.parse_args(cl_args)
cl_args = vars(cl_args)
load_configuration(cl_args, '../config.json')
cl_args.update(test_args)
cl_args.update(self.model_params)
run.main(cl_args)
for file in os.listdir(self.results_folder):
img = imageio.imread(os.path.join(self.results_folder, file))
self.assertTrue(img.shape == (self.img_size['HR'],
self.img_size['HR'], 3))
def test_that_configuration_is_loaded_correctly(self):
parser = get_parser()
cl_args = parser.parse_args(['--train', '--custom-data'])
cl_args = vars(cl_args)
load_configuration(cl_args, '../config.json')
self.assertTrue(isinstance(cl_args['G'], int))
self.assertTrue(cl_args['log_dir'] == './logs')
self.assertTrue(cl_args['weights_dir'] == './weights')
self.assertTrue(cl_args['data_name'] == 'CUSTOM')
def test_if_train_is_executed(self):
self.create_random_dataset('correct', dataset_size=14)
train_arguments = {
'validation_labels': self.dataset_folder['correct']['HR'],
'validation_input': self.dataset_folder['correct']['LR'],
'training_labels': self.dataset_folder['correct']['HR'],
'training_input': self.dataset_folder['correct']['LR'],
}
cl_args = ['--pytest', '--no_verbose']
parser = get_parser()
cl_args = parser.parse_args(cl_args)
cl_args = vars(cl_args)
load_configuration(cl_args, '../config.json')
cl_args.update(train_arguments)
run.main(cl_args)
self.assertTrue(1 != 0)
def test_if_trainable_weights_update_with_one_step(self):
self.scale = self.model_params['scale']
self.img_size = {'HR': 10 * self.scale, 'LR': 10}
self.dataset_size = 8
self.create_random_dataset(type='correct')
before_step = []
for layer in self.model.rdn.layers:
if len(layer.trainable_weights) > 0:
before_step.append(layer.get_weights()[0])
train_arguments = {
'validation_labels': self.dataset_folder['correct']['HR'],
'validation_input': self.dataset_folder['correct']['LR'],
'training_labels': self.dataset_folder['correct']['HR'],
'training_input': self.dataset_folder['correct']['LR'],
}
cl_args = ['--pytest', '--no_verbose']
parser = get_parser()
cl_args = parser.parse_args(cl_args)
cl_args = vars(cl_args)
load_configuration(cl_args, '../config.json')
cl_args.update(train_arguments)
trainer = Trainer(train_arguments=cl_args)
i = 0
for layer in self.model.rdn.layers:
if len(layer.trainable_weights) > 0:
self.assertTrue(
np.all(before_step[i] == layer.get_weights()[0]))
i += 1
trainer.train_model(self.model)
i = 0
for layer in self.model.rdn.layers:
if len(layer.trainable_weights) > 0:
self.assertFalse(
np.all(before_step[i] == layer.get_weights()[0]))
i += 1
from models import VanillaCNN, SpectralDropoutCNN, SpectralDropoutEasyCNN
from utils.data_utils import get_data_loaders
from utils.utils import get_parser
# from utils.visualization import plot_hidden
# System
path_to_home = os.environ['HOME']
path_to_proj = os.path.join(path_to_home, 'spectral-dropout')
path_save = os.path.join(path_to_proj, 'saved_models')
# Model parameters
GREYSCALE = True
WEIGHT_DECAY = 1e-3
# Get a parser for the training settings
parser = get_parser()
def signal_handler(sig, frame):
print('Aborting, trying to save the model')
try:
final_model_save_name = ''.join([model_save_name, 'final', '.pt'])
torch.save(
model,
os.path.join(path_save, model_save_name, final_model_save_name))
print('Model saved!')
except NameError:
print('Model not saved, probably not existing yet')
finally:
sys.exit(0)
def test_that_parser_returns_the_correct_namepace(self):
parser = get_parser()
cl_args = parser.parse_args(['--train'])
namespace = cl_args._get_kwargs()
self.assertTrue(('train', True) in namespace)
self.assertTrue(('test', False) in namespace)
from utils import utils, keystoneutils, novautils
# Create a parser and parse the args and/or the environment variables.
p = utils.get_parser()
args = p.parse_args()
# Create a keystone client from the args and run supported operations.
kc = keystoneutils.get_client(args)
# List all projects.
print kc.projects.list()
# Show details of a project.
print kc.projects.get(kc.projects.list()[0].id)
# Create a project.
print kc.projects.create('my_project', 'default',
description='This is my_project.')
# List all users with admin roles.
print kc.role_assignments.list(role=kc.roles.list(name='admin')[0].id)
# Create a nova client and run supported operations.
nc = novautils.get_client(kc.session)
# OR
# session = keystoneutils.get_session(args)
# nc = novautils.get_client()
# List all instances. Like `nova list --all-tenants`.