tf_agents.networks.NestMap

The NestMap network processes nested inputs via nested layers.

Inherits From: Network

tf_agents.networks.NestMap(
    nested_layers: tf_agents.typing.types.NestedLayer,
    input_spec: typing.Optional[types.NestedTensorSpec] = None,
    name: typing.Optional[typing.Text] = None
)

It is a TF-Agents network that can be used to process nested inputs.

Stateful Keras layers (e.g. LSTMCell, RNN, LSTM, TF-Agents DynamicUnroll) are all supported. The state_spec of NestMap has a structure matching that of nested_layers.

NestMap can be used in conjunction with NestFlatten and a combiner (e.g. tf.keras.layers.Add or tf.keras.layers.Concatenate) to process and aggregate in a preprocessing step.

Usage:

net = NestMap({"inp1": layer1, "inp2": layer2})
outputs, next_state = net({"inp1": inp1, "inp2": inp2}, state)

nested_layers A nest of layers and/or networks. These will be used to process the inputs (input nest structure will have to match this structure). Any layers that are subclasses of tf.keras.layers.{RNN,LSTM,GRU,...} are wrapped in tf_agents.keras_layers.RNNWrapper.
input_spec (Optional.) A nest of tf.TypeSpec representing the input observations. The structure of input_spec must match that of nested_layers.
name (Optional.) Network name.

TypeError If any of the layers are not instances of keras Layer.
ValueError If input_spec is provided but its nest structure does not match that of nested_layers.
RuntimeError If not tf.executing_eagerly(); as this is required to be able to create deep copies of layers in layers.

input_tensor_spec Returns the spec of the input to the network of type InputSpec.
layers Get the list of all (nested) sub-layers used in this Network.
nested_layers

state_spec

Methods

copy

View source

copy(
    **kwargs
) -> 'NestMap'

Make a copy of a NestMap instance.

Args
**kwargs Args to override when recreating this network. Commonly overridden args include 'name'.

Returns
A deep copy of this network.

create_variables

View source

create_variables(
    input_tensor_spec=None, **kwargs
)

Force creation of the network's variables.

Return output specs.

Args
input_tensor_spec (Optional). Override or provide an input tensor spec when creating variables.
**kwargs Other arguments to network.call(), e.g. training=True.

Returns
Output specs - a nested spec calculated from the outputs (excluding any batch dimensions). If any of the output elements is a tfp Distribution, the associated spec entry returned is a DistributionSpec.

Raises
ValueError If no input_tensor_spec is provided, and the network did not provide one during construction.

get_initial_state

View source

get_initial_state(
    batch_size=None
)

Returns an initial state usable by the network.

Args
batch_size Tensor or constant: size of the batch dimension. Can be None in which case not dimensions gets added.

Returns
A nested object of type self.state_spec containing properly initialized Tensors.

get_layer

View source

get_layer(
    name=None, index=None
)

Retrieves a layer based on either its name (unique) or index.

If name and index are both provided, index will take precedence. Indices are based on order of horizontal graph traversal (bottom-up).

Args
name String, name of layer.
index Integer, index of layer.

Returns
A layer instance.

Raises
ValueError In case of invalid layer name or index.

summary

View source

summary(
    line_length=None, positions=None, print_fn=None
)

Prints a string summary of the network.

Args
line_length Total length of printed lines (e.g. set this to adapt the display to different terminal window sizes).
positions Relative or absolute positions of log elements in each line. If not provided, defaults to [.33, .55, .67, 1.].
print_fn Print function to use. Defaults to print. It will be called on each line of the summary. You can set it to a custom function in order to capture the string summary.

Raises
ValueError if summary() is called before the model is built.