twodlearn.constrained module

Constrained variables

This module defines a set of constrained variables. These variables are trainable, and can only take values inside a given region. For example, PositiveVariable is only allowed to take positive values.

class twodlearn.constrained.BoundedVariable(min, max, initializer, initial_value=None, trainable=True, collections=None, validate_shape=True, caching_device=None, name='variable', variable_def=None, dtype=None, expected_shape=None, import_scope=None, constraint=None, options=None)

Bases: twodlearn.constrained.TransformedVariable

Creates a variable that can only take values inside a range

inverse(value)

property max

property min

transform(value)

class twodlearn.constrained.ConstrainedVariable(initial_value, min=0.0, max=inf, name=None)

Bases: twodlearn.core.common.TdlModel

initial_value

property initializer

max

min

projection(x)

property shape

property value

variable

class twodlearn.constrained.PositiveVariable(initial_value, trainable=True, collections=None, validate_shape=True, caching_device=None, name=None, variable_def=None, dtype=None, expected_shape=None, import_scope=None, constraint=None, **kargs)

Bases: twodlearn.constrained.TransformedVariable

Creates a variable that can only take positive values.

This fuction uses tf.nn.softplus to ensure positive values.

Tdl autoinitialization with arguments:

tolerance

(InputArgument) tolerance for positive variables.

raw

(SimpleParameter) raw value before making a transformation

value

(OutputValue)

inverse(value)

tolerance

tolerance for positive variables.

transform(value)

class twodlearn.constrained.PositiveVariable2(initial_value, trainable=True, collections=None, validate_shape=True, caching_device=None, name=None, variable_def=None, dtype=None, expected_shape=None, import_scope=None, constraint=None, **kargs)

Bases: twodlearn.constrained.TransformedVariable

Creates a variable that can only take positive values. This function uses pow(x,2) as a reparameterization of the variable

inverse(value)

tolerance

tolerance for positive variables.

transform(value)

class twodlearn.constrained.PositiveVariableExp(initial_value, max=None, trainable=True, collections=None, validate_shape=True, caching_device=None, name=None, tolerance=None)

Bases: twodlearn.constrained.TransformedVariable

Creates a variable that can only take positive values.

This function uses exp() as a reparameterization of the variable

Tdl autoinitialization with arguments:

raw

(ParameterInit) raw value before making a transformation

tolerance

(InputArgument) tolerance for positive variables.

max

(InputArgument)

value

(OutputValue)

inverse(value)

max

raw

raw value before making a transformation

tolerance

tolerance for positive variables.

transform(value)

class twodlearn.constrained.TransformedVariable(initial_value, trainable=True, collections=None, validate_shape=True, caching_device=None, name=None, variable_def=None, dtype=None, expected_shape=None, import_scope=None, constraint=None, **kargs)

Bases: twodlearn.core.common.TdlModel

Variable that uses a transformation to ensure the output is constrained.

Initialization is performed using the inverse method raw = tf.Variable(inverse(initial_value))

The value is defined using the transform value = transform(raw)

Tdl autoinitialization with arguments:

raw

(SimpleParameter) raw value before making a transformation

value

(OutputValue)

classmethod init_wrapper(initializer, trainable, **kargs)

wraps the object to be used as an initializer.

property initializer

inverse(value)

raw

raw value before making a transformation

property shape

transform(value)

value