modules – Structural Components: Modules

class pybrain.structure.modules.BiasUnit(name=None)

Bases: pybrain.structure.modules.neuronlayer.NeuronLayer

A simple bias unit with a single constant output.

class pybrain.structure.modules.GaussianLayer(dim, name=None)

Bases: pybrain.structure.modules.neuronlayer.NeuronLayer, pybrain.structure.parametercontainer.ParameterContainer

A layer implementing a gaussian interpretation of the input. The mean is the input, the sigmas are stored in the module parameters.

setSigma(sigma)

Wrapper method to set the sigmas (the parameters of the module) to a certain value.

class pybrain.structure.modules.LinearLayer(dim, name=None)

Bases: pybrain.structure.modules.neuronlayer.NeuronLayer

The simplest kind of module, not doing any transformation.

__init__(dim, name=None)

Create a layer with dim number of units.

class pybrain.structure.modules.LSTMLayer(dim, peepholes=False, name=None)

Bases: pybrain.structure.modules.neuronlayer.NeuronLayer, pybrain.structure.parametercontainer.ParameterContainer

Long short-term memory cell layer.

The input consists of 4 parts, in the following order: - input gate - forget gate - cell input - output gate

__init__(dim, peepholes=False, name=None)

Parameter:dim – number of cells Key peepholes:enable peephole connections (from state to gates)?

class pybrain.structure.modules.MDLSTMLayer(dim, dimensions=1, peepholes=False, name=None)

Bases: pybrain.structure.modules.neuronlayer.NeuronLayer, pybrain.structure.parametercontainer.ParameterContainer

Multi-dimensional long short-term memory cell layer.

The cell-states are explicitly passed on through a part of the input/output buffers (which should be connected correctly with IdentityConnections).

The input consists of 4 parts, in the following order: - input gate - forget gates (1 per dim) - cell input - output gate - previous states (1 per dim)

The output consists of two parts: - cell output - current statte

Attention: this module has to be used with care: it’s last <size> input and outputs are reserved for transmitting internal states on flattened recursive multi-dim networks, and so its connections have always to be sliced!

class pybrain.structure.modules.SigmoidLayer(dim, name=None)

Bases: pybrain.structure.modules.neuronlayer.NeuronLayer

Layer implementing the sigmoid squashing function.

__init__(dim, name=None)

Create a layer with dim number of units.

class pybrain.structure.modules.SoftmaxLayer(dim, name=None)

Bases: pybrain.structure.modules.neuronlayer.NeuronLayer

A layer implementing a softmax distribution over the input.

__init__(dim, name=None)

Create a layer with dim number of units.

class pybrain.structure.modules.StateDependentLayer(dim, module, name=None, onesigma=True)

Bases: pybrain.structure.modules.neuronlayer.NeuronLayer, pybrain.structure.parametercontainer.ParameterContainer

class pybrain.structure.modules.TanhLayer(dim, name=None)

Bases: pybrain.structure.modules.neuronlayer.NeuronLayer

A layer implementing the tanh squashing function.

__init__(dim, name=None)

Create a layer with dim number of units.