twodlearn.bayesnet.bayesnet module¶

Definition of several bayesian neural-networks

class twodlearn.bayesnet.bayesnet.AffineBernoulliLayer(units, *args, **kargs)[source]¶

Bases: twodlearn.feedforward.AffineLayer

Implements the layer y=dropout(x) W + b

class Output(model, inputs, options=None, name=None)[source]¶

Bases: twodlearn.feedforward.Output

affine[source]¶

property keep_prob[source]¶

value[source]¶

keep_prob[source]¶: Keep prob for dropout.

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.bayesnet.bayesnet.AffineNormalLayer(units, *args, **kargs)[source]¶

Bases: twodlearn.bayesnet.bayesnet.LinearNormalLayer

class Output(**kargs)[source]¶: Bases: twodlearn.bayesnet.bayesnet.Output

bias[source]¶: Autoinit with arguments [‘initializer’, ‘trainable’]

class twodlearn.bayesnet.bayesnet.BayesianMlp(n_inputs, n_outputs, n_hidden, afunction=<function selu01>, options=None, name='BayesianMlp')[source]¶

Bases: twodlearn.feedforward.StackedModel

Mlp composed of layers whose weights are sampled from a variational posterior distribution

class BayesMlpOutput(**kargs)[source]¶

Bases: twodlearn.core.common.OutputModel

property kernels[source]¶

property shape[source]¶

evaluate(*args, **kargs)[source]¶

property kernels[source]¶: list the weights distributions from the layers

layers[source]¶

property n_inputs[source]¶: size of the input vectors

property n_outputs[source]¶: size of the output vectors

property parameters[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.bayesnet.bayesnet.BernoulliBayesianMlp(n_inputs, n_outputs, n_hidden, keep_prob=0.8, options=None, afunction=<function relu>, name='BernoulliBayesianMlp')[source]¶: Bases: twodlearn.bayesnet.bayesnet.BayesianMlp

class twodlearn.bayesnet.bayesnet.BoundedBayesianMlp(n_inputs, n_outputs, n_hidden, lower=1e-07, upper=None, afunction=<function selu01>, options=None, name='BayesianMlp')[source]¶

Bases: twodlearn.bayesnet.bayesnet.BayesianMlp

Multi-layer bayesian neural network, with bounded output

layers[source]¶

class twodlearn.bayesnet.bayesnet.BoundedBernoulliBayesianMlp(n_inputs, n_outputs, n_hidden, keep_prob, lower=1e-07, upper=None, afunction=<function selu01>, options=None, name='BayesianMlp')[source]¶

Bases: twodlearn.bayesnet.bayesnet.BernoulliBayesianMlp

Multi-layer bayesian neural network, with bounded output

layers[source]¶

class twodlearn.bayesnet.bayesnet.ConditionalNormal(loc=None, scale=None, shape=None, name='ConditionalNormal')[source]¶

Bases: twodlearn.bayesnet.bayesnet.Normal

class NormalOutput(model, inputs, name=None)[source]¶

Bases: twodlearn.bayesnet.bayesnet.McNormal

inputs[source]¶

loc[source]¶

scale[source]¶

loc[source]¶

class twodlearn.bayesnet.bayesnet.DenseBernoulliLayer(activation=<function relu>, name=None, **kargs)[source]¶

Bases: twodlearn.bayesnet.bayesnet.AffineBernoulliLayer

class Output(model, inputs, options=None, name=None)[source]¶

Bases: twodlearn.bayesnet.bayesnet.Output

property keep_prob[source]¶

value[source]¶

activation[source]¶

class twodlearn.bayesnet.bayesnet.DenseNormalLayer(activation=<function relu>, name=None, **kargs)[source]¶

Bases: twodlearn.bayesnet.bayesnet.AffineNormalLayer

class Output(**kargs)[source]¶

Bases: twodlearn.bayesnet.bayesnet.Output

property value[source]¶: Sample from the output distribution.

activation[source]¶

class twodlearn.bayesnet.bayesnet.Entropy(prob, name=None)[source]¶

Bases: twodlearn.core.common.TdlModel

property prob[source]¶

property value[source]¶

class twodlearn.bayesnet.bayesnet.GaussianKL(p, q, name='GaussianKL')[source]¶

Bases: twodlearn.losses.Loss

Evaluate KL(p||q) for p and q Normal

evaluate(p, q)[source]¶: Evaluate KL(p||q)

fromlist(p_list)[source]¶

classmethod fromstats(p_loc, p_scale, q_loc, q_scale)[source]¶

get_n_vars(p, q)[source]¶

property p[source]¶

property q[source]¶

class twodlearn.bayesnet.bayesnet.GaussianNegLogLikelihood(y, labels=None, name='NegLogLikelihood')[source]¶

Bases: twodlearn.losses.EmpiricalLoss

define_fit_loss(y, labels)[source]¶

property labels[source]¶: Labels for computing the loss, if not provided, they are created automatically

property n_outputs[source]¶

property y[source]¶

class twodlearn.bayesnet.bayesnet.HeteroscedasticNormalMlp(loc_args, scale_args, LocClass=<class 'twodlearn.bayesnet.bayesnet.BayesianMlp'>, ScaleClass=<class 'twodlearn.bayesnet.bayesnet.BoundedBayesianMlp'>, options=None, name='HeteroscedasticGaussianMlp')[source]¶

Bases: twodlearn.bayesnet.bayesnet.NormalMlp

Defines a conditional gaussian N(loc=BayesianMlp(x), scale=BoundedOutputBayesianMlp(x))

class twodlearn.bayesnet.bayesnet.LinearNormalLayer(units, *args, **kargs)[source]¶

Bases: twodlearn.feedforward.LinearLayer

Tdl autoinitialization with arguments:

kernel[source]¶: (ParameterInit) Autoinit with arguments [‘initializer’, ‘trainable’, ‘max_scale’]

input_shape[source]¶: (InputArgument)

regularizer[source]¶: (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.

units[source]¶: (InputArgument) Number of output units (int).

tolerance[source]¶: (InputArgument)

class Output(**kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

affine[source]¶: Normal distribution for the outputs.

inputs[source]¶

model[source]¶

property shape[source]¶

value[source]¶: Sample from the output distribution.

call(inputs, *args, **kargs)[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.

kernel[source]: Autoinit with arguments [‘initializer’, ‘trainable’, ‘max_scale’]

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.

tolerance[source]

class twodlearn.bayesnet.bayesnet.McEstimate(value, name='mc_estimate')[source]¶

Bases: twodlearn.core.common.TdlModel

mean[source]¶

stddev[source]¶

class twodlearn.bayesnet.bayesnet.McNormal(loc, scale, samples=None, name='McNormal', **kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

loc[source]¶

resample(*args, **kargs)[source]¶

property samples[source]¶

scale[source]¶

property value[source]¶

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

Bases: twodlearn.core.layers.Layer

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.

input_shape[source]¶: Input tensor shape.

sample_dim[source]¶: Sample dimension across which mean and stddev are computed.

class twodlearn.bayesnet.bayesnet.McSample(**kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

distribution[source]¶

sample_axis[source]¶

class twodlearn.bayesnet.bayesnet.Normal(loc=None, scale=None, shape=None, name='Normal', **kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

class NormalOutput(model, inputs, name=None)[source]¶

Bases: twodlearn.bayesnet.bayesnet.McNormal

inputs[source]¶

loc[source]¶

scale[source]¶

evaluate(inputs=None, name=None)[source]¶

loc[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.

scale[source]¶

class twodlearn.bayesnet.bayesnet.NormalMlp(loc_args, scale_args={}, LocClass=<class 'twodlearn.bayesnet.bayesnet.BayesianMlp'>, ScaleClass=None, options=None, name='GaussianMlp')[source]¶

Bases: twodlearn.bayesnet.bayesnet.ConditionalNormal

class McNormalOutput(model, n_particles, inputs, name=None)[source]¶

Bases: twodlearn.bayesnet.bayesnet.NormalOutput

inputs[source]¶

property n_particles[source]¶

class NormalOutput(model, inputs, name=None)[source]¶

Bases: twodlearn.bayesnet.bayesnet.NormalOutput

inputs[source]¶

property n_inputs[source]¶

property n_outputs[source]¶

loc[source]¶

mc_evaluate(n_particles, x=None, name=None)[source]¶

property n_inputs[source]¶

property n_outputs[source]¶

scale[source]¶

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

Bases: twodlearn.core.layers.Layer

Tdl autoinitialization with arguments:

scale[source]¶: (SubmodelInit) Autoinit with arguments [‘initializer’, ‘trainable’, ‘tolerance’]

input_shape[source]¶: (InputArgument) Input tensor shape.

batch_shape[source]¶: (InputArgument)

loc[source]¶: (SubmodelInit) Autoinit with arguments [‘initializer’, ‘trainable’]

batch_shape[source]

build(input_shape=None)[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, *args, **kargs)[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.

input_shape[source]: Input tensor shape.

loc[source]: Autoinit with arguments [‘initializer’, ‘trainable’]

scale[source]: Autoinit with arguments [‘initializer’, ‘trainable’, ‘tolerance’]

class twodlearn.bayesnet.bayesnet.Particles(n_particles, base=None, shape=None, name='Particles', **kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

base[source]¶

property base_shape[source]¶

property n_particles[source]¶

property value[source]¶

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

Bases: twodlearn.core.layers.Layer

Tdl autoinitialization with arguments:

particles[source]¶: (InputArgument) Number of particles.

input_shape[source]¶: (InputArgument) Input tensor shape.

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.

compute_output_shape(input_shape=None)[source]¶

Computes the output shape of the layer.

Assumes that the layer will be built to match that input shape provided.

Parameters: input_shape – Shape tuple (tuple of integers) or list of shape tuples (one per output tensor of the layer). Shape tuples can include None for free dimensions, instead of an integer.
Returns: An input shape tuple.