tf.keras.Initializer

View source on GitHub

Initializer base class: all Keras initializers inherit from this class.

Initializers should implement a __call__() method with one of two possible signatures. If your initializer does not handle distribution layouts, it should use this signature:

def __call__(self, shape, dtype=None):
    # returns a tensor of shape `shape` and dtype `dtype`
    # containing values drawn using a function of your choice.

If your initializer handles distribution layouts, it should use this signature. layout can either be a keras.distribution.TensorLayout or a backend-specific layout. If the given layout is not None, the returned value must be distributed according to layout.

def __call__(self, shape, dtype=None, layout=None):
    # returns a tensor of shape `shape`, dtype `dtype` and distributed
    # across devices according to the layout `layout`.
    # containing values drawn using a function of your choice.

Optionally, you can also implement the method get_config() and the class method from_config in order to support serialization, just like with any Keras object.

Here's a simple example: a random normal initializer.

class ExampleRandomNormal(Initializer):
    def __init__(self, mean, stddev):
        self.mean = mean
        self.stddev = stddev

    def __call__(self, shape, dtype=None):
        return keras.random.normal(
            shape, mean=self.mean, stddev=self.stddev, dtype=dtype
        )

    def get_config(self):  # To support serialization
        return {"mean": self.mean, "stddev": self.stddev}

Note that we don't have to implement from_config() in the example above since the constructor arguments of the class the keys in the config returned by get_config() are the same. In this case, the default from_config() works fine.

Methods

clone

View source

clone()

from_config

View source

@classmethod
from_config(
    config
)

Instantiates an initializer from a configuration dictionary.

Example:

initializer = RandomUniform(-1, 1)
config = initializer.get_config()
initializer = RandomUniform.from_config(config)

get_config

View source

get_config()

Returns the initializer's configuration as a JSON-serializable dict.

__call__

View source

__call__(
    shape, dtype=None
)

Returns a tensor object initialized as specified by the initializer.