Package crino :: Module criterion :: Class NegativeLogLikelihood

Class NegativeLogLikelihood

The Negative Log Likelihood criterion is used in classification task.

It is meant to be connected to a Softmax Module as the last layer of the MLP.

The output size should be the same as the number of possible classes.

The target size should be the same as the number of possible classes, with value in between [0..1]. If not weighted, all 0 except 1 on the correct class.

The Negative Log Likelihood can be written as follows :

L_NLL = − log( < ŷ, y > )

Instance Methods

[hide private]

__init__(self, outputs, targets)
Constructs a new NegativeLogLikelihood criterion.

source code

prepare(self)
Computes the Negative Log Likelihood symbolic expression.

source code

Instance Variables

[hide private]

Inherited from Criterion: expression, outputs, targets

Method Details

[hide private]

init(self, outputs, targets)
(Constructor)

source code

Constructs a new NegativeLogLikelihood criterion.

Parameters:

outputs (TensorVariable) - The symbolic outputs vector of a network
targets (TensorVariable) - The symbolic targets vector

Overrides: Criterion.__init__

prepare(self)

source code

Computes the Negative Log Likelihood symbolic expression.

Overrides: Criterion.prepare

Class NegativeLogLikelihood

__init__(self, outputs, targets) (Constructor)

prepare(self)

init(self, outputs, targets)
(Constructor)