params-flow 0.1.8

Tensorflow Keras utilities for reducing boilerplate code.

TensorFlow Keras utilities and helpers for building custom layers by reducing boilerplate code.

LICENSE

MIT. See License File.

Install

params-flow is on the Python Package Index (PyPI):

pip install params-flow

Usage

params-flow provides a Layer class that helps reducing common boilerplate code when writing custom Keras layers.

Instead of defining your layer parameters in __init(), define them in a Params class like this:

class MyLayer(params_flow.Layer):
    Params(params_flow.Layer.Params):
       hidden_size = 128
       activation  = "gelu"

After extending the params.flow.Layer like above, the base class will take care for serializing your layer configuration, and will spare you from coding comon keras boilerplate code like:

class MyLayer(keras.Layer):
    def __init__(self, hidden_size=128, activation="gelu"):
        super(MyLayer, self).__init__()
        self.hidden_size = hidden_size
        self.activation  = activation
    def get_config(self):
        config = {
        "hidden_size": self.hidden_size,
        "activation":  self.activation,
    }
    base_config = super(MyLayer, self).get_config()
    return dict(list(base_config.items()) + list(config.items()))