def run(method, output_dir, fake_data=False, fake_training=False):
"""Trains a model and records its predictions on configured datasets.
Args:
method: Modeling method to experiment with.
output_dir: Directory to record the trained model and output stats.
fake_data: If true, use fake data.
fake_training: If true, train for a trivial number of steps.
Returns:
Trained Keras model.
"""
tf.io.gfile.makedirs(output_dir)
model_opts = hparams_lib.model_opts_from_hparams(
hparams_lib.HPS_DICT[method], method, fake_training=fake_training)
if fake_training:
model_opts.batch_size = 32
model_opts.examples_per_epoch = 256
model_opts.train_epochs = 1
experiment_utils.record_config(model_opts,
output_dir + '/model_options.json')
dataset_train = data_lib.build_dataset(image_data_utils.DATA_CONFIG_TRAIN,
is_training=True,
fake_data=fake_data)
dataset_test = data_lib.build_dataset(image_data_utils.DATA_CONFIG_TEST,
fake_data=fake_data)
model = models_lib.build_and_train(model_opts, dataset_train, dataset_test,
output_dir)
logging.info('Saving model to output_dir.')
model.save_weights(output_dir + '/model.ckpt')
# TODO(yovadia): Figure out why save_model() wants to serialize ModelOptions.
return model
def test_static_cifar_c(self, split):
if not flags.FLAGS.fake_data:
config = image_data_utils.DataConfig(
split, corruption_static=True, corruption_level=3,
corruption_type='pixelate')
if split in ['train', 'valid']:
with self.assertRaises(ValueError):
data_lib.build_dataset(config)
else:
dataset = data_lib.build_dataset(config)
image_shape = next(iter(dataset))[0].numpy().shape
self.assertEqual(image_shape, data_lib.CIFAR_SHAPE)
def run(dataset_name,
model_dir,
predictions_per_example,
max_examples,
output_dir,
fake_data=False):
"""Runs predictions on the given dataset using the specified model."""
tf.io.gfile.makedirs(output_dir)
data_config = image_data_utils.get_data_config(dataset_name)
dataset = data_lib.build_dataset(data_config, fake_data=fake_data)
if max_examples:
dataset = dataset.take(max_examples)
model = models_lib.load_model(model_dir)
logging.info('Starting predictions.')
predictions = experiment_utils.make_predictions(model,
dataset.batch(_BATCH_SIZE),
predictions_per_example)
logging.info('Done computing predictions; recording results to disk.')
array_utils.write_npz(output_dir, 'predictions_%s.npz' % dataset_name,
predictions)
del predictions['logits_samples']
array_utils.write_npz(output_dir,
'predictions_small_%s.npz' % dataset_name,
predictions)
def test_value_cifar_c(self, split):
if not flags.FLAGS.fake_data:
config = image_data_utils.DataConfig(
split, corruption_value=.25, corruption_type='brightness')
dataset = data_lib.build_dataset(config)
image_shape = next(iter(dataset))[0].numpy().shape
self.assertEqual(image_shape, data_lib.CIFAR_SHAPE)
def test_array_cifar_c(self, split):
if not flags.FLAGS.fake_data:
config = image_data_utils.DataConfig(
split, corruption_level=4, corruption_type='glass_blur')
dataset = data_lib.build_dataset(config)
image_shape = next(iter(dataset))[0].numpy().shape
self.assertEqual(image_shape, data_lib.CIFAR_SHAPE)
def test_roll_pixels(self, split):
config = image_data_utils.DataConfig(split, roll_pixels=5)
if not flags.FLAGS.fake_data:
dataset = data_lib.build_dataset(config)
image_shape = next(iter(dataset))[0].numpy().shape
self.assertEqual(image_shape, data_lib.CIFAR_SHAPE)
def test_fake_data(self, split):
# config is ignored for fake data
config = image_data_utils.DataConfig(split)
dataset = data_lib.build_dataset(config, fake_data=True)
image_shape = next(iter(dataset))[0].numpy().shape
self.assertEqual(image_shape, data_lib.CIFAR_SHAPE)
