twodlearn.losses module

class twodlearn.losses.AddLoss(loss1, loss2, name='AddLoss')

Bases: twodlearn.losses.Loss

Tdl autoinitialization with arguments:

Attributes:

property loss1

property loss2

class twodlearn.losses.AddNLosses(losses, name='AddNLosses')

Bases: twodlearn.losses.Loss

Tdl autoinitialization with arguments:

Attributes:

property losses

mean()

class twodlearn.losses.ClassificationLoss(logits, labels=None, name=None)

Bases: twodlearn.losses.EmpiricalLoss

Tdl autoinitialization with arguments:

correct_prediction

(LazzyProperty)

accuracy

(LazzyProperty)

value

(OutputValue)

labels

(InputArgument) Labels for computing the loss, if not provided, they are created automatically

logits

(InputArgument)

cross_entropy

(Submodel)

accuracy

correct_prediction

cross_entropy

labels

Labels for computing the loss, if not provided, they are created automatically

logits

property n_classes

property n_outputs

value

class twodlearn.losses.EmpiricalLoss(**kargs)

Bases: twodlearn.losses.Loss

Tdl autoinitialization with arguments:

Attributes:

property labels

Labels for computing the loss, if not provided, they are created automatically

class twodlearn.losses.EmpiricalLossWrapper(loss, labels, name='EmpiricalLoss')

Bases: twodlearn.losses.EmpiricalLoss

Tdl autoinitialization with arguments:

Attributes:

property loss

property value

class twodlearn.losses.EmpiricalWithRegularization(empirical, regularizer, alpha=None, name='ERLoss')

Bases: twodlearn.losses.Loss

Linear combination of a Empirical and a Reguralizer loss: loss = empirical + alpha * regularizer If alpha is None, the result is the sum of the empirical and regularizer losses: loss = empirical + regularizer

Tdl autoinitialization with arguments:

Attributes:

property alpha

property empirical

property labels

property regularizer

class twodlearn.losses.GreaterThan(x, reference, mask=None, func=<function softplus>, name='GreaterThanLoss')

Bases: twodlearn.losses.LessThan

Loss that punishes values being larger than a given value.

\[loss = func(x - reference)\]

Tdl autoinitialization with arguments:

reference

(InputArgument)

x

(InputArgument)

mask

(InputArgument)

loss_eval(x, reference, mask)

class twodlearn.losses.L1Regularizer(weights, scale=None, name='L2Regularizer')

Bases: twodlearn.losses.L2Regularizer

Tdl autoinitialization with arguments:

Attributes:

define_loss(weights, scale)

class twodlearn.losses.L2Loss(y, labels=None, name='L2Loss')

Bases: twodlearn.losses.EmpiricalLoss

Computes (1/M)sum( (y - labels)**2 )

Tdl autoinitialization with arguments:

Attributes:

define_fit_loss(y, labels)

property n_outputs

property y

class twodlearn.losses.L2Regularizer(weights, scale=None, name='L2Regularizer')

Bases: twodlearn.losses.Loss

Tdl autoinitialization with arguments:

Attributes:

define_loss(weights, scale)

property scale

property weights

class twodlearn.losses.LessThan(x, reference, mask=None, func=<function softplus>, name='LessThanLoss')

Bases: twodlearn.losses.Loss

Defines a loss that punishes the variable being smaller than a given value loss = func(reference - x)

Tdl autoinitialization with arguments:

reference

(InputArgument)

x

(InputArgument)

mask

(InputArgument)

loss_eval(x, reference, mask)

mask

mean(*args, **kargs)

reference

x

class twodlearn.losses.Loss(**kargs)

Bases: twodlearn.core.common.TdlModel

Tdl autoinitialization with arguments:

Attributes:

property value

class twodlearn.losses.LossMethod(output_vars, input_vars, OutputClass=None)

Bases: twodlearn.core.common.ModelMethod

Decorator used to specify an operation for a loss inside a model. The decorator works similar to @property, but the specified method correponds to the definition of the operation

Examples

Usage of the decorator:

class MyModel(tdl.TdlModel):
    _submodels = ['evaluate']
    @tdl.LossMethod(['y'], # list of outputs
                    ['x']  # list of inputs
                    )
    def mean_loss(self, x):
        return tf.reduce_mean(x)

class twodlearn.losses.MultipliedLosses(loss1, loss2, name='MultipliedLosses')

Bases: twodlearn.losses.Loss

property loss1

property loss2

class twodlearn.losses.QuadraticLoss(x, q=None, target=None, name='QuadraticLoss')

Bases: twodlearn.losses.Loss

Defines a cuadratic loss that takes the form:

\[loss = (X-target) q (X-target)^T\]

Tdl autoinitialization with arguments:

q

(SimpleParameter)

target

(InputArgument)

x

(InputArgument)

mean(*args, **kargs)

q

target

x

class twodlearn.losses.ScaledLoss(alpha, loss, name='ScaledLoss')

Bases: twodlearn.losses.Loss

property alpha

property pre_scaled

twodlearn.losses.convert_loss_to_tensor(value, dtype=None, name=None, as_ref=False)