Module: tf_agents.keras_layers

Keras Layers Module.


bias_layer module: Keras layer mirroring tf.contrib.layers.bias_add.

dynamic_unroll_layer module: Tensorflow RL Agent RNN utilities.

inner_reshape module: Keras layer to reshape inner dimensions (keeping outer dimensions the same).

permanent_variable_rate_dropout module: A keras layer that applies dropout both in training and serving.

rnn_wrapper module: Wrapper for tf.keras.layers.RNN subclasses.

squashed_outer_wrapper module: SquashedOuterWrapper Keras Layer.


class BiasLayer: Keras layer that only adds a bias to the input.

class DynamicUnroll: Process a history of sequences that are concatenated without padding.

class RNNWrapper: Wraps a Keras RNN/LSTM/GRU layer to make network state more consistent.

class SquashedOuterWrapper: Squash the outer dimensions of input tensors; unsquash outputs.


InnerReshape(...): Returns a Keras layer that reshapes the inner dimensions of tensors.