twodlearn.recurrent module¶

class twodlearn.recurrent.BaseCell(trainable=True, name=None, *args, **kwargs)[source]¶

Bases: twodlearn.core.layers.Layer

build(input_shape)[source]¶

Creates the variables of the layer (optional, for subclass implementers).

This is a method that implementers of subclasses of Layer or Model can override if they need a state-creation step in-between layer instantiation and layer call.

This is typically used to create the weights of Layer subclasses.

Parameters: input_shape – Instance of TensorShape, or list of instances of TensorShape if the layer expects a list of inputs (one instance per input).

call(inputs)[source]¶

This is where the layer’s logic lives.

Parameters

inputs – Input tensor, or list/tuple of input tensors.
**kwargs – Additional keyword arguments.

Returns

A tensor or list/tuple of tensors.

call_cell(inputs, state)[source]¶

get_initial_state(inputs=None, batch_size=None, initializer=None)[source]¶

input_shape[source]¶

state_shape[source]¶

class twodlearn.recurrent.DenseCell(trainable=True, name=None, *args, **kwargs)[source]¶

Bases: twodlearn.recurrent.BaseCell

Base RNN cell for which the inputs and states are represented using a dense tensor.

get_initial_state(batch_size=None, initializer=None)[source]¶

get_inputs(batch_size=None, initializer=None)[source]¶

class twodlearn.recurrent.Lstm(n_inputs, n_outputs, n_hidden, name=None, **kargs)[source]¶

Bases: twodlearn.recurrent.Rnn

class LstmSetup(**kargs)[source]¶

Bases: twodlearn.recurrent.RnnSetup

class LstmStateAndOutput(hidden, y)[source]¶

Bases: twodlearn.core.common.TdlModel

hidden[source]¶

y[source]¶

property labels[source]¶

ModelOutput[source]¶: alias of Lstm.LstmSetup

define_cell(n_inputs, n_outputs, n_hidden)[source]¶

n_hidden[source]¶

property parameters[source]¶

property weights[source]¶

class twodlearn.recurrent.Lstm2Lstm(n_inputs, n_outputs, n_hidden, afunction=<function tanh>, encoder_afunction=<function tanh>, name=None)[source]¶

Bases: twodlearn.core.common.TdlModel

Uses an Lstm to convert a fixed length sequence into the initial state for an LSTM sequential model

ModelOutput[source]¶: alias of Lstm2Lstm.Output

class Output(model, n_unrollings=1, encoder_n_unrollings=1, batch_size=None, inputs=None, encoder_inputs=None, compute_loss=True, options=None, name='lstm2lstm')[source]¶

Bases: twodlearn.core.common.TdlModel

property encoder[source]¶

property fit_loss[source]¶

property inputs[source]¶

property labels[source]¶

property loss[source]¶

property lstm[source]¶

property x0[source]¶: Inputs to the encoder

property y[source]¶

encoder[source]¶: Autoinit with arguments [‘n_inputs’, ‘n_outputs’, ‘n_hidden’, ‘afunction’, ‘name’]

lstm[source]¶: Autoinit with arguments [‘n_inputs’, ‘n_outputs’, ‘n_hidden’, ‘afunction’, ‘name’]

property parameters[source]¶

property weights[source]¶

class twodlearn.recurrent.LstmCellOptimized(n_inputs, n_units, afunction=<function tanh>, name='LstmCell')[source]¶

Bases: twodlearn.core.common.TdlModel

Single lstm cell defined as in: “Generating Sequences with: Recurrent Neural Networks”, Alex Graves, 2014

n_inputs[source]¶: number of inputs

n_nodes[source]¶: number of nodes

afunction[source]¶: activation function

name[source]¶: name used in all TensorFlow variables’ names

class LstmCellSetup(**kargs)[source]¶

Bases: twodlearn.core.common.OutputModel

input_state[source]¶

inputs[source]¶

value[source]¶

afunction[source]: activation function for the cell

evaluate(*args, **kargs)[source]¶

n_inputs[source]

property n_outputs[source]¶

n_units[source]¶

parameters[source]¶

property weights[source]¶

class twodlearn.recurrent.LstmState(h, x)[source]¶

Bases: twodlearn.core.common.TdlModel

h[source]¶

x[source]¶

class twodlearn.recurrent.Mlp2Lstm(n_inputs, n_outputs, window_size=1, name=None, **kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

Uses an MLP to convert a fixed length sequence into the initial state for an LSTM sequential model

class Mlp2LstmOutput(**kargs)[source]¶

Bases: twodlearn.core.common.OutputModel

batch_size[source]¶

inputs[source]¶: Autoinit with arguments [‘n_unrollings’, ‘Type’, ‘batch_size’]

lstm[source]¶

lstm_x0[source]¶

property n_unrollings[source]¶

property window_size[source]¶

x0[source]¶: Autoinit with arguments [‘Type’, ‘batch_size’]

evaluate(x0=None, inputs=None, **kargs)[source]¶

lstm[source]¶: Autoinit with arguments [‘n_hidden’]

mlp[source]¶: Autoinit with arguments [‘n_hidden’]

n_inputs[source]¶

n_outputs[source]¶

property parameters[source]¶

property weights[source]¶

window_size[source]¶

class twodlearn.recurrent.MlpNarx(n_inputs, n_outputs, window_size, n_hidden, afunction=<function relu>, name='mlp_narx', **kargs)[source]¶

Bases: twodlearn.recurrent.Narx

Narx that uses an Mlp as cell

CellModel[source]¶: alias of twodlearn.feedforward.MlpNet

ModelOutput[source]¶: alias of MlpNarx.Output

class Output(model, x0=None, n_unrollings=1, batch_size=None, inputs=None, compute_loss=True, options=None, name=None)[source]¶

Bases: twodlearn.recurrent.NarxSetup

property labels[source]¶

property afunction[source]¶: activation function for the MLP

define_cell(n_inputs, n_outputs, window_size)[source]¶

property n_hidden[source]¶: Number of hidden layers

setup(*args, **kargs)[source]¶

class twodlearn.recurrent.MultilayerLstmCell(n_inputs, n_hidden, n_outputs=None, output_layer=None, name=None, **kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

class MultilayerLstmCellSetup(**kargs)[source]¶

Bases: twodlearn.core.common.OutputModel

hidden[source]¶

input_h[source]¶

input_x[source]¶

output[source]¶

property state[source]¶

evaluate(*args, **kargs)[source]¶

hidden_layers[source]¶

n_hidden[source]¶

n_inputs[source]¶

property n_outputs[source]¶

output_layer[source]¶

property parameters[source]¶

property weights[source]¶

class twodlearn.recurrent.Narx(n_inputs, n_outputs, window_size=1, name='narx', **kargs)[source]¶

Bases: twodlearn.recurrent.Rnn

class NarxSetup(**kargs)[source]¶

Bases: twodlearn.recurrent.RnnSetup

property window_size[source]¶

cell[source]¶

define_cell(n_inputs, n_outputs, window_size)[source]¶

property window_size[source]¶

class twodlearn.recurrent.Rnn(n_inputs, n_outputs, n_states=None, name='rnn', **kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

class RnnSetup(**kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

inputs[source]¶: exogenous inputs

property loss[source]¶

model[source]¶

property n_inputs[source]¶: number of exogenous inputs

property n_outputs[source]¶: number of outputs from the model

n_unrollings[source]¶

property outputs[source]¶

reset_inputs[source]¶

property states[source]¶: State of the network

property unrolled[source]¶: list of unrolled networks

x0[source]¶: initial state

cell[source]¶

define_cell(n_inputs, n_outputs, n_states)[source]¶

evaluate(x0=None, inputs=None, n_unrollings=None, options=None, name=None, **kargs)[source]¶

n_inputs[source]¶

n_outputs[source]¶

class twodlearn.recurrent.SimpleRnn(cell, name='SimpleRnn', options=None, **kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

class RnnOutput(model, x0=None, inputs=None, n_unrollings=None, options=None, name=None)[source]¶

Bases: twodlearn.core.common.TdlModel

inputs[source]¶: Setup either tf variables or placeholders for the external (control) inputs

loss[source]¶: Decorator used to specify an optional property inside a model. The decorator works similar to @property, but the specified method correponds to the initialization of the property

property n_unrollings[source]¶

reset_inputs[source]¶

property unrolled[source]¶: list of unrolled networks

property x[source]¶

x0[source]¶

property y[source]¶

cell[source]¶

evaluate(x0=None, inputs=None, n_unrollings=None, options=None, name=None, **kargs)[source]¶

regularizer[source]¶: Decorator used to specify a regularizer for a model. The decorator works similar to @property, but the specified method correponds to the initialization of the regularizer.

class twodlearn.recurrent.StateSpaceCell(trainable=True, name=None, *args, **kwargs)[source]¶

Bases: twodlearn.recurrent.BaseCell

call_cell(inputs, state)[source]¶

input_shape[source]¶

output_model[source]¶

outputs = call(state)

Type: Callable model with signature

state_model[source]¶

next_state = call(inputs, state)

Type: Callable model with signature

state_shape[source]¶

class twodlearn.recurrent.StateSpaceDense(trainable=True, name=None, *args, **kwargs)[source]¶: Bases: twodlearn.recurrent.StateSpaceCell, twodlearn.recurrent.DenseCell

twodlearn.recurrent.explicit_call_wrapper(model)[source]¶: Re-factors the call function of a model/layer to have the signature: call(inputs, state)

twodlearn.recurrent.same_state_output_wrapper(cell)[source]¶: Re-factors the call function of a model/layer to have the signature: outputs, outputs = call(inputs, state)