def get(identifier, **kwargs):
# Sanity check
assert len(kwargs) == 0
# Return identifier directly if it is callable
if callable(identifier): return identifier
elif isinstance(identifier, six.string_types):
# Parse identifier
p = Parser.parse(identifier)
identifier = p.name.lower()
if identifier in ['relu']: return relu
elif identifier in ['lrelu', 'leakyrelu', 'leaky-relu']:
leak = p.get_arg(float, default=0.1)
return lambda x: leaky_relu(x, leak=leak)
elif identifier in ['softmax']:
return softmax
elif identifier in ['cumax']:
return cumax
elif identifier in ['sigmoid']:
return lambda x: sigmoid(x, **kwargs)
else:
# Try to find activation in tf.nn
activation = getattr(tf, identifier, None)
if activation is None:
raise ValueError('Can not resolve {}'.format(identifier))
return activation
else:
raise TypeError('identifier must be callable or a string')
def get_global_constraint(self):
if self.global_constraint in [None, '']: return None
p = Parser.parse(self.global_constraint)
if p.name in ['max_norm']:
max_value = p.get_arg(float, default=2.0)
axis = p.get_kwarg('axis', int, default=0)
return tf.keras.constraints.max_norm(max_value=max_value, axis=axis)
else: KeyError('Unknown constraint name `{}`'.format(p.name))
def __init__(self, identifier, set_logits=False):
p = Parser.parse(identifier)
self._id = p.name
self.abbreviation = (p.name if isinstance(identifier, six.string_types)
else identifier.__name__)
self.full_name = self.abbreviation
self._activation = activations.get(identifier)
self._set_logits = set_logits
def get(identifier):
if identifier is None or callable(identifier): return identifier
if not isinstance(identifier, six.string_types):
raise TypeError('identifier must be a function or a string')
p = Parser.parse(identifier)
key = p.name.lower()
if key in ['l1']: return L1(penalty=p.get_arg(float))
elif key in ['l2']: return L2(penalty=p.get_arg(float))
else: raise ValueError('Can not resolve "{}"'.format(key))
def image_augmentation_processor(data_batch: DataSet, is_training: bool):
# Get hub
th = tfr.hub
if not is_training or th.aug_config is None: return data_batch
# Parse augmentation setting
assert isinstance(th.aug_config, str)
configs = [Parser.parse(s) for s in th.aug_config.split('|')]
if len(configs) == 0: return data_batch
# Apply each method according to configs
for cfg in configs:
# Find method
if cfg.name == 'rotate': method = _rotate
elif cfg.name == 'flip': method = _flip
else:
raise KeyError('!! Unknown augmentation option {}'.format(
cfg.name))
# Do augmentation
data_batch.features = method(data_batch.features, *cfg.arg_list,
**cfg.arg_dict)
return data_batch
def developer_options(self): if not self.developer_args: return None from tframe.utils.arg_parser import Parser parser = Parser.parse(self.developer_args) return parser