twodlearn.bayesnet.recurrent module

This module defines recurrent bayesian neural-networks

class twodlearn.bayesnet.recurrent.BayesNarx(cell, window_size, name=None, options=None, **kargs)

Bases: twodlearn.recurrent.SimpleRnn

class McRnnOutput(n_particles, x0=None, inputs=None, n_unrollings=None, options=None, name=None, **kargs)

Bases: twodlearn.bayesnet.recurrent.RnnOutput

inputs

property n_particles

x0

class RnnOutput(model, x0=None, inputs=None, n_unrollings=None, options=None, name=None)

Bases: twodlearn.recurrent.RnnOutput

property window_size

x0

list of observations of the initial state. :param value: value for x0 :type value: None, tf.Tensor, McNormal

Returns

x0

Return type

list(McNormal)

mc_evaluate(n_particles, x0=None, inputs=None, n_unrollings=None, options=None, name=None, **kargs)

property window_size

class twodlearn.bayesnet.recurrent.DropoutMlpCell(n_inputs, n_outputs, n_hidden, window_size, keep_prob, options=None, name='DropoutMlpCell')

Bases: twodlearn.bayesnet.recurrent.GaussianMlpCell

property keep_prob

class twodlearn.bayesnet.recurrent.GaussianMlpCell(n_inputs, n_outputs, n_hidden, window_size, options=None, name='GaussianMlpCell')

Bases: twodlearn.core.common.TdlModel

ModelOutput

alias of GaussianMlpCell.Output

class Output(model, state, inputs, t, options=None, name=None)

Bases: twodlearn.core.common.TdlModel

property dz

Output delta increment before applying observation noise

evaluate(xt, ut, t)

model

property u_inputs

Exogenous inputs to the cell

property x

State of the cell, in the case of Narx, is the list of window_size previous observations y

property x_inputs

State inputs to the cell

property y

Observable output of the cell

evaluate(state, inputs, t, options=None, name=None)

mlp

property n_hidden

property n_inputs

property n_outputs

property normalizer

regularizer

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.

setup(*args, **kargs)

property window_size

class twodlearn.bayesnet.recurrent.NormalNarxCell(fz, fy=None, options=None, name='GaussianMlpCell', **kargs)

Bases: twodlearn.core.common.TdlModel

(state, input) -> normalizer -> finput -> fz -> fy

class NarxConcat(axis, name='Concat')

Bases: twodlearn.feedforward.Concat

class NormalNarxOutput(**kargs)

Bases: twodlearn.core.common.OutputModel

loss

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

evaluate(*args, **kargs)

finput

fy

fz

normalizer

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

regularizer

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.

twodlearn.bayesnet.recurrent.normal_residual_wrapper(model)