Base class for recurrent layers


Base class for recurrent layers


  return_sequences = FALSE, 
  return_state = FALSE, 
  go_backwards = FALSE, 
  stateful = FALSE, 
  unroll = FALSE, 
  time_major = FALSE, 
  zero_output_for_mask = FALSE 


Arguments Description
object What to compose the new Layer instance with. Typically a Sequential model or a Tensor (e.g., as returned by layer_input()). The return value depends on object. If object is:
- missing or NULL, the Layer instance is returned.
- a Sequential model, the model with an additional layer is returned.
- a Tensor, the output tensor from layer_instance(object) is returned.
cell A RNN cell instance or a list of RNN cell instances. A RNN cell is a class that has:
- A call(input_at_t, states_at_t) method, returning (output_at_t, states_at_t_plus_1). The call method of the cell can also take the optional argument constants, see section “Note on passing external constants” below.
- A state_size attribute. This can be a single integer (single state) in which case it is the size of the recurrent state. This can also be a list of integers (one size per state). The state_size can also be TensorShape or list of TensorShape, to represent high dimension state.
- A output_size attribute. This can be a single integer or a TensorShape, which represent the shape of the output. For backward compatible reason, if this attribute is not available for the cell, the value will be inferred by the first element of the state_size.
- A get_initial_state(inputs=NULL, batch_size=NULL, dtype=NULL) method that creates a tensor meant to be fed to call() as the initial state, if the user didn’t specify any initial state via other means. The returned initial state should have a shape of [batch_size, cell$state_size]. The cell might choose to create a tensor full of zeros, or full of other values based on the cell’s implementation. inputs is the input tensor to the RNN layer, which should contain the batch size as first dimension (inputs$shape[1]), and also dtype (inputs$dtype). Note that the shape[1] might be NULL during the graph construction. Either the inputs or the pair of batch_size and dtype are provided. batch_size is a scalar tensor that represents the batch size of the inputs. dtype is tf.DType that represents the dtype of the inputs. For backward compatibility, if this method is not implemented by the cell, the RNN layer will create a zero filled tensor with the size of [batch_size, cell$state_size]. In the case that cell is a list of RNN cell instances, the cells will be stacked on top of each other in the RNN, resulting in an efficient stacked RNN.
return_sequences Boolean (default FALSE). Whether to return the last output in the output sequence, or the full sequence.
return_state Boolean (default FALSE). Whether to return the last state in addition to the output.
go_backwards Boolean (default FALSE). If TRUE, process the input sequence backwards and return the reversed sequence.
stateful Boolean (default FALSE). If TRUE, the last state for each sample at index i in a batch will be used as initial state for the sample of index i in the following batch.
unroll Boolean (default FALSE). If TRUE, the network will be unrolled, else a symbolic loop will be used. Unrolling can speed-up a RNN, although it tends to be more memory-intensive. Unrolling is only suitable for short sequences.
time_major The shape format of the inputs and outputs tensors. If TRUE, the inputs and outputs will be in shape (timesteps, batch, ...), whereas in the FALSE case, it will be (batch, timesteps, ...). Using time_major = TRUE is a bit more efficient because it avoids transposes at the beginning and end of the RNN calculation. However, most TensorFlow data is batch-major, so by default this function accepts input and emits output in batch-major form.
standard layer arguments.
zero_output_for_mask Boolean (default FALSE). Whether the output should use zeros for the masked timesteps. Note that this field is only used when return_sequences is TRUE and mask is provided. It can useful if you want to reuse the raw output sequence of the RNN without interference from the masked timesteps, eg, merging bidirectional RNNs.


See the Keras RNN API guide for details about the usage of RNN API.


Call arguments

  • inputs: Input tensor.

  • mask: Binary tensor of shape [batch_size, timesteps] indicating whether a given timestep should be masked. An individual TRUE entry indicates that the corresponding timestep should be utilized, while a FALSE entry indicates that the corresponding timestep should be ignored.

  • training: R or Python Boolean indicating whether the layer should behave in training mode or in inference mode. This argument is passed to the cell when calling it. This is for use with cells that use dropout.

  • initial_state: List of initial state tensors to be passed to the first call of the cell.

  • constants: List of constant tensors to be passed to the cell at each timestep.

Input shapes

N-D tensor with shape (batch_size, timesteps, ...), or (timesteps, batch_size, ...) when time_major = TRUE.

Output shape

  • if return_state: a list of tensors. The first tensor is the output. The remaining tensors are the last states, each with shape (batch_size, state_size), where state_size could be a high dimension tensor shape.

  • if return_sequences: N-D tensor with shape [batch_size, timesteps, output_size], where output_size could be a high dimension tensor shape, or [timesteps, batch_size, output_size] when time_major is TRUE

  • else, N-D tensor with shape [batch_size, output_size], where output_size could be a high dimension tensor shape.


This layer supports masking for input data with a variable number of timesteps. To introduce masks to your data, use layer_embedding() with the mask_zero parameter set to TRUE.

Statefulness in RNNs

You can set RNN layers to be ‘stateful’, which means that the states computed for the samples in one batch will be reused as initial states for the samples in the next batch. This assumes a one-to-one mapping between samples in different successive batches. For intuition behind statefulness, there is a helpful blog post here: https://philipperemy.github.io/keras-stateful-lstm/ To enable statefulness:

  • Specify stateful = TRUE in the layer constructor.

  • Specify a fixed batch size for your model. For sequential models, pass batch_input_shape = list(...) to the first layer in your model. For functional models with 1 or more Input layers, pass batch_shape = list(...) to all the first layers in your model. This is the expected shape of your inputs including the batch size. It should be a list of integers, e.g. list(32, 10, 100). For dimensions which can vary (are not known ahead of time), use NULL in place of an integer, e.g. list(32, NULL, NULL).

  • Specify shuffle = FALSE when calling fit().

    To reset the states of your model, call layer$reset_states() on either a specific layer, or on your entire model.

Initial State of RNNs

You can specify the initial state of RNN layers symbolically by calling them with the keyword argument initial_state. The value of initial_state should be a tensor or list of tensors representing the initial state of the RNN layer. You can specify the initial state of RNN layers numerically by calling reset_states with the named argument states. The value of states should be an array or list of arrays representing the initial state of the RNN layer.

Passing external constants to RNNs

You can pass “external” constants to the cell using the constants named argument of RNN$__call__ (as well as RNN$call) method. This requires that the cell$call method accepts the same keyword argument constants. Such constants can be used to condition the cell transformation on additional static inputs (not changing over time), a.k.a. an attention mechanism.

See Also