def configurable_module(module):
if not issubclass(module, nn.Module):
raise ValueError(
"this decorator can only be used on flax.nn.Module class.")
def wrapper(**kwargs):
return module.partial(**kwargs)
wrapper.__name__ = module.__name__
return gin.configurable(wrapper)
def register_module_with_gin(module, module_name=None):
"""
Register all the callables in a single module with gin.
A useful way to add gin configurability to a codebase without explicilty using the @gin.configurable decorator.
"""
module_name = module.__name__ if module_name is None else module_name
for attr in dir(module):
if callable(getattr(module, attr)):
setattr(module, attr, gin.configurable(getattr(module, attr), module=module_name))
import gin
from tensorflow.keras import applications
from tensorflow.compat.v1.keras.layers import BatchNormalization
from thin.models import resnet
EfficientNetB0 = gin.configurable(applications.EfficientNetB0,
module='tf.keras.applications')
EfficientNetB1 = gin.configurable(applications.EfficientNetB1,
module='tf.keras.applications')
EfficientNetB2 = gin.configurable(applications.EfficientNetB2,
module='tf.keras.applications')
EfficientNetB3 = gin.configurable(applications.EfficientNetB3,
module='tf.keras.applications')
EfficientNetB4 = gin.configurable(applications.EfficientNetB4,
module='tf.keras.applications')
EfficientNetB5 = gin.configurable(applications.EfficientNetB5,
module='tf.keras.applications')
EfficientNetB6 = gin.configurable(applications.EfficientNetB6,
module='tf.keras.applications')
EfficientNetB7 = gin.configurable(applications.EfficientNetB7,
module='tf.keras.applications')
ResNet18 = gin.configurable(resnet.ResNet18, module='thin.models')
ResNet34 = gin.configurable(resnet.ResNet34, module='thin.models')
ResNet50 = gin.configurable(resnet.ResNet50, module='thin.models')
ResUNet18 = gin.configurable(resnet.ResUNet18, module='thin.models')
ResUNet34 = gin.configurable(resnet.ResUNet34, module='thin.models')
ResUNet50 = gin.configurable(resnet.ResUNet50, module='thin.models')
# See the License for the specific language governing permissions and # limitations under the License. # Lint as: python3 """Module with all the global configurable models for training.""" from ddsp.training.models.autoencoder import Autoencoder from ddsp.training.models.vae import VAE from ddsp.training.models.inverse_synthesis import InverseSynthesis from ddsp.training.models.midi_autoencoder import MidiAutoencoder from ddsp.training.models.midi_autoencoder import ZMidiAutoencoder from ddsp.training.models.model import Model import gin _configurable = lambda cls: gin.configurable(cls, module=__name__) Autoencoder = _configurable(Autoencoder) VAE = _configurable(VAE) InverseSynthesis = _configurable(InverseSynthesis) MidiAutoencoder = _configurable(MidiAutoencoder) ZMidiAutoencoder = _configurable(ZMidiAutoencoder) @gin.configurable def get_model(model=gin.REQUIRED): """Gin configurable function get a 'global' model for use in ddsp_run.py. Convenience for using the same model in train(), evaluate(), and sample(). Args:
# limitations under the License.
# Lint as: python3
"""Tests for ddsp.dags.py."""
from absl.testing import parameterized
from ddsp import dags
import gin
import tensorflow as tf
# Make dense layers configurable for this test.
gin.external_configurable(tf.keras.layers.Dense, 'tf.keras.layers.Dense')
# Make dag_layers configurable for this test.
gin.enter_interactive_mode()
gin.configurable(dags.DAGLayer)
class DAGLayerTest(parameterized.TestCase, tf.test.TestCase):
def setUp(self):
"""Create some dummy input data for the chain."""
super().setUp()
# Create inputs.
self.n_batch = 4
self.x_dims = 5
self.z_dims = 2
self.x = tf.ones([self.n_batch, self.x_dims])
self.inputs = {'test_data': self.x}
self.gin_config_kwarg_modules = f"""
import ddsp
dags.run_dag.verbose = True