def load_dataset_wavegan():
"""Load WaveGAN's dataset.
The loaded dataset consists of:
- original data (dataset_blob, train_data, train_label),
- encoded data from a pretrained model (train_mu, train_sigma), and
- index grouped by label (index_grouped_by_label).
Some of these attributes are not avaiable (set as None) but are left here
to keep everything aligned with returned value of `load_dataset`.
Returns:
An tuple of abovementioned components in the dataset.
"""
latent_dir = os.path.expanduser(FLAGS.wavegan_latent_dir)
path_train = os.path.join(latent_dir, 'data_train.npz')
train = np.load(path_train)
train_z = train['z']
train_label = train['label']
index_grouped_by_label = common.get_index_grouped_by_label(train_label)
dataset_blob, train_data = None, None
train_mu, train_sigma = train_z, None
return (dataset_blob, train_data, train_label, train_mu, train_sigma,
index_grouped_by_label)
def load_dataset_wavegan():
"""Load WaveGAN's dataset.
The loaded dataset consists of:
- original data (dataset_blob, train_data, train_label),
- encoded data from a pretrained model (train_mu, train_sigma), and
- index grouped by label (index_grouped_by_label).
Some of these attributes are not avaiable (set as None) but are left here
to keep everything aligned with returned value of `load_dataset`.
Returns:
An tuple of abovementioned components in the dataset.
"""
latent_dir = os.path.expanduser(FLAGS.wavegan_latent_dir)
path_train = os.path.join(latent_dir, 'data_train.npz')
train = np.load(path_train)
train_z = train['z']
train_label = train['label']
index_grouped_by_label = common.get_index_grouped_by_label(train_label)
dataset_blob, train_data = None, None
train_mu, train_sigma = train_z, None
return (dataset_blob, train_data, train_label, train_mu, train_sigma,
index_grouped_by_label)
def load_dataset(config_name, exp_uid):
"""Load a dataset from a config's name.
The loaded dataset consists of:
- original data (dataset_blob, train_data, train_label),
- encoded data from a pretrained model (train_mu, train_sigma), and
- index grouped by label (index_grouped_by_label).
Args:
config_name: A string indicating the name of config to parameterize the
model that associates with the dataset.
exp_uid: A string representing the unique id of experiment to be used in
model that associates with the dataset.
Returns:
An tuple of abovementioned components in the dataset.
"""
config = load_config(config_name)
if config_is_wavegan(config):
return load_dataset_wavegan()
model_uid = common.get_model_uid(config_name, exp_uid)
dataset = common.load_dataset(config)
train_data = dataset.train_data
attr_train = dataset.attr_train
path_train = os.path.join(dataset.basepath, 'encoded', model_uid,
'encoded_train_data.npz')
train = np.load(path_train)
train_mu = train['mu']
train_sigma = train['sigma']
train_label = np.argmax(attr_train, axis=-1) # from one-hot to label
index_grouped_by_label = common.get_index_grouped_by_label(train_label)
tf.logging.info('index_grouped_by_label size: %s',
[len(_) for _ in index_grouped_by_label])
tf.logging.info('train loaded from %s', path_train)
tf.logging.info('train shapes: mu = %s, sigma = %s', train_mu.shape,
train_sigma.shape)
dataset_blob = dataset
return (dataset_blob, train_data, train_label, train_mu, train_sigma,
index_grouped_by_label)
def load_dataset(config_name, exp_uid):
"""Load a dataset from a config's name.
The loaded dataset consists of:
- original data (dataset_blob, train_data, train_label),
- encoded data from a pretrained model (train_mu, train_sigma), and
- index grouped by label (index_grouped_by_label).
Args:
config_name: A string indicating the name of config to parameterize the
model that associates with the dataset.
exp_uid: A string representing the unique id of experiment to be used in
model that associates with the dataset.
Returns:
An tuple of abovementioned components in the dataset.
"""
config = load_config(config_name)
if config_is_wavegan(config):
return load_dataset_wavegan()
model_uid = common.get_model_uid(config_name, exp_uid)
dataset = common.load_dataset(config)
train_data = dataset.train_data
attr_train = dataset.attr_train
path_train = os.path.join(dataset.basepath, 'encoded', model_uid,
'encoded_train_data.npz')
train = np.load(path_train)
train_mu = train['mu']
train_sigma = train['sigma']
train_label = np.argmax(attr_train, axis=-1) # from one-hot to label
index_grouped_by_label = common.get_index_grouped_by_label(train_label)
tf.logging.info('index_grouped_by_label size: %s',
[len(_) for _ in index_grouped_by_label])
tf.logging.info('train loaded from %s', path_train)
tf.logging.info('train shapes: mu = %s, sigma = %s', train_mu.shape,
train_sigma.shape)
dataset_blob = dataset
return (dataset_blob, train_data, train_label, train_mu, train_sigma,
index_grouped_by_label)
