def get_metric(metric_name):
if metric_name is None:
return None
metric_modules = ['trident.optims.tensorflow_metrics']
try:
metric_fn = get_function(camel2snake(metric_name), metric_modules)
except Exception:
metric_fn = get_function(metric_name, metric_modules)
return metric_fn
def get_metric(metric_name):
if metric_name is None:
return None
metric_modules = ['trident.optims.pytorch_metrics']
if metric_name in __all__:
metric_fn = get_function(metric_name, metric_modules)
else:
try:
metric_fn = get_function(camel2snake(metric_name), metric_modules)
except Exception:
metric_fn = get_function(metric_name, metric_modules)
return metric_fn
def get_activation(fn_name,only_layer=False):
"""
Args:
fn_name ():
Returns:
Examples:
>>> print(get_activation('swish'))
"""
if fn_name is None:
return None
fn_modules = ['trident.layers.pytorch_activations', 'trident.backend.pytorch_ops', 'torch.nn.functional']
trident_fn_modules = ['trident.layers.pytorch_activations', 'trident.backend.pytorch_ops']
if only_layer:
fn_modules = ['trident.layers.pytorch_activations']
trident_fn_modules = ['trident.layers.pytorch_activations']
try:
if isinstance(fn_name, str):
if not only_layer and (camel2snake(fn_name)== fn_name or fn_name.lower()== fn_name):
if fn_name == 'p_relu' or fn_name == 'prelu':
return PRelu()
activation_fn = get_function(fn_name, trident_fn_modules)
return activation_fn
else:
try:
activation_fn = get_class(snake2camel(fn_name), fn_modules)
return activation_fn()
except Exception:
activation_fn = get_class(fn_name, fn_modules)
return activation_fn()
elif getattr(fn_name, '__module__', None) == 'trident.layers.pytorch_activations':
if inspect.isfunction(fn_name):
return partial(fn_name)
elif inspect.isclass(fn_name) and fn_name.__class__.__name__=="type":
return fn_name()
elif isinstance(fn_name, Layer):
return fn_name
elif inspect.isfunction(fn_name) and getattr(fn_name, '__module__', None) == 'trident.backend.pytorch_ops':
if only_layer:
activation_layer = get_class(snake2camel(fn_name.__name__), trident_fn_modules)
return activation_layer()
else:
return fn_name
else:
if callable(fn_name):
result = inspect.getfullargspec(fn_name)
if 1 <= len(result.args) <= 2:
return fn_name if inspect.isfunction(fn_name) else fn_name()
else:
raise ValueError('Unknown activation function/ class')
except Exception as e:
print(e)
return None
def get_reg(reg_name):
if reg_name is None:
return None
if '_reg' not in reg_name:
reg_name = reg_name + '_reg'
reg_modules = ['trident.optims.tensorflow_regularizers']
reg_fn = get_function(reg_name, reg_modules)
return reg_fn
def get_reg(reg_name):
if reg_name is None:
return None
if '_reg' not in reg_name:
reg_name = reg_name + '_reg'
reg_modules = ['trident.optims.pytorch_regularizers']
reg_fn = get_function(camel2snake(reg_name), reg_modules)
return reg_fn
def get_initializer(initializer, **kwargs):
if isinstance(initializer, str):
initializer_fn = get_function(camel2snake(initializer),
['trident.backend.pytorch_initializers'])
initializer_fn = partial(initializer_fn, **
kwargs) if len(kwargs) > 0 else initializer_fn
return initializer_fn
elif inspect.isfunction(initializer) and getattr(
initializer, '__module__',
None) == 'trident.backend.pytorch_initializers':
initializer = partial(initializer, **
kwargs) if len(kwargs) > 0 else initializer
return initializer