tfm.utils.get_activation

View source on GitHub

Maps an identifier to a Python function, e.g., "relu" => tf.nn.relu.

tfm.utils.get_activation(
 identifier, use_keras_layer=False, **kwargs
)

It checks string first and if it is one of customized activation not in TF, the corresponding activation will be returned. For non-customized activation names and callable identifiers, always fallback to tf.keras.activations.get.

Prefers using keras layers when use_keras_layer=True. Now it only supports 'relu', 'linear', 'identity', 'swish', 'mish', 'leaky_relu', and 'gelu'.

Args

identifier String name of the activation function or callable.
use_keras_layer If True, use keras layer if identifier is allow-listed.
**kwargs Keyword arguments to use to instantiate an activation function. Available only for 'leaky_relu' and 'gelu' when using keras layers. For example: get_activation('leaky_relu', use_keras_layer=True, alpha=0.1)

Returns

A Python function corresponding to the activation function or a keras activation layer when use_keras_layer=True.

Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies. Java is a registered trademark of Oracle and/or its affiliates. Some content is licensed under the numpy license.

Last updated 2024年02月02日 UTC.