twodlearn.core.common module¶

This module defines common structures used in twodlearn

class twodlearn.core.common.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)[source]¶

Bases: twodlearn.core.common.TransformedVariable

Creates a variable that can only take values inside a range

inverse(value)[source]¶

property max[source]¶

property min[source]¶

transform(value)[source]¶

class twodlearn.core.common.ConstrainedVariable(initial_value, min=0.0, max=inf, name=None)[source]¶

Bases: twodlearn.core.common.TdlModel

initial_value[source]¶

property initializer[source]¶

max[source]¶

min[source]¶

projection(x)[source]¶

property value[source]¶

variable[source]¶

class twodlearn.core.common.EagerMethod(finit=None, feval=None)[source]¶

Bases: object

Decorator used to specify methods that perform operations using tensorflow, but are evaluated immediatly. These methods consist of an initialization method (speficied with the function given to the decorator) and an execute

eval(feval)[source]¶

class twodlearn.core.common.EncapsulatedMethod(finit=None, feval=None)[source]¶

Bases: object

Decorator used to specify methods that have a set of local variables.

These methods consist of an initialization method (speficied with the function given to the decorator) and an execute method.

class MethodData(func)[source]¶: Bases: object

eval(feval)[source]¶

init(obj, value)[source]¶

init_local(obj, value)[source]¶

class twodlearn.core.common.Identity(**kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

evaluate(x)[source]¶

class twodlearn.core.common.InferenceInput(finit=None, doc=None, name=None)[source]¶

Bases: twodlearn.core.common.InputArgument

Decorator used to specify a model input required to perform inference. The decorator works similar to @property, but the specified method correponds to the initialization of the argument, ussually checking for types and setting default values

class twodlearn.core.common.InputArgument(finit=None, doc=None, name=None)[source]¶

Bases: twodlearn.core.common.TdlDescriptor

Decorator used to specify the input arguments for a model. The decorator works similar to @property, but the specified method correponds to the initialization of the argument, ussually checking for types and setting default values.

class twodlearn.core.common.InputModel(finit=None, doc=None, name=None)[source]¶

Bases: twodlearn.core.common.InputArgument

Decorator used to specify the input models for a model. The decorator works similar to @property, but the specified method correponds to the initialization of the argument, ussually checking for types and setting default values

class twodlearn.core.common.InputModelInit(finit=None, inference_input=None, lazzy=False, doc=None)[source]¶: Bases: twodlearn.core.common.SubmodelInit

class twodlearn.core.common.InputParameter(finit=None, doc=None, name=None)[source]¶

Bases: twodlearn.core.common.InputArgument

Decorator used to specify the input arguments for a model. These inputs will serve as parameters. The decorator works similar to @property, but the specified method correponds to the initialization of the argument, ussually checking for types and setting default values

class twodlearn.core.common.LazzyProperty(finit=None)[source]¶

Bases: object

autoinit(obj, force=True)[source]¶

init_property(obj)[source]¶

class twodlearn.core.common.MethodInit(finit=None, feval=None, doc=None)[source]¶

Bases: object

Descriptor that encapsulates a set of local variables for the method. The following is an example of how this descriptor can be used:

Examples

The descriptor can be used to define both, an initialization method, and an evaluation method

class Example(tdl.core.TdlModel):
    @tdl.core.MethodInit
    def evaluate(self, local, units=20):
        local.model = tdf.core.dense.DenseLayer(units=units)

    @evaluate.eval
    def evaluate(self, local, inputs, n_samples=100):
        return local.model(inputs)/n_samples

class MethodData(obj, attr_name, finit, feval)[source]¶

Bases: object

init(*args, **kargs)[source]¶

autoinit(obj, force=False)[source]¶

eval(feval)[source]¶

init(obj, val=None)[source]¶

class twodlearn.core.common.ModelMethod(output_vars, input_vars, OutputClass=None)[source]¶

Bases: object

Decorator used to specify an operation inside a model.

The decorator works similar to @property, but the specified method correponds to the definition of the operation.

Examples

Model methods can be used as follows

class MyModel(tdl.TdlModel):
    _submodels = ['evaluate']
    @tdl.ModelMethod(['y'],     # list of outputs
                     ['x' ,'y'] # list of inputs
                    )
    def evaluate(self, x, u):
        return x+u

class twodlearn.core.common.OptionalProperty(finit=None, fset=None)[source]¶

Bases: object

Decorator used to specify an optional property inside a model. The decorator works similar to @property, but the specified method correponds to the initialization of the property

init(obj, val)[source]¶

setter(fset)[source]¶

class twodlearn.core.common.OptionalPropertyWrapper(obj, finit, fset, value)[source]¶

Bases: twodlearn.core.common.TdlOp

call(*args, **kargs)[source]¶

init(*args, **kargs)[source]¶

property is_set[source]¶

set_value(value)[source]¶

property value[source]¶

class twodlearn.core.common.Options(*argv, **kargs)[source]¶: Bases: dict

class twodlearn.core.common.OutputModel(**kargs)[source]¶

Bases: twodlearn.core.common.TdlModel

eval(input_vars)[source]¶

get_save_data()[source]¶

model[source]¶

class twodlearn.core.common.OutputValue(finit=None, doc=None, name=None)[source]¶

Bases: twodlearn.core.common.TdlDescriptor

Decorator used to specify the output value of a model. The decorator works similar to @property, but the specified method correponds to the definition of the output value. These methods are called at the end of the auto initialization procedure.

class twodlearn.core.common.Parameter(fget=None, finit=None)[source]¶

Bases: object

init(finit)[source]¶

class twodlearn.core.common.ParameterInit(finit=None, inference_input=None, lazzy=False, doc=None)[source]¶: Bases: twodlearn.core.common.SubmodelInit

class twodlearn.core.common.PositiveVariable(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)[source]¶

Bases: twodlearn.core.common.TransformedVariable

Creates a variable that can only take positive values

inverse(value)[source]¶

transform(value)[source]¶

class twodlearn.core.common.PositiveVariable2(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)[source]¶

Bases: twodlearn.core.common.TransformedVariable

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

inverse(value)[source]¶

transform(value)[source]¶

class twodlearn.core.common.PositiveVariableExp(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)[source]¶

Bases: twodlearn.core.common.TransformedVariable

Creates a variable that can only take positive values. This function uses exp() as a reparameterization of the variable

inverse(value)[source]¶

transform(value)[source]¶

class twodlearn.core.common.Regularizer(finit=None, fset=None)[source]¶

Bases: twodlearn.core.common.OptionalProperty

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.core.common.ScopedMethod(feval=None)[source]¶

Bases: object

adds a tf.name_scope before executing the method

class ScopedWrapper(instance, feval, name)[source]¶: Bases: object

class twodlearn.core.common.SimpleNamespace[source]¶

Bases: types.SimpleNamespace

SimpleNamespace that works with tf.convert_to_tensor and is searchable

class twodlearn.core.common.SimpleParameter(finit=None, doc=None, name=None)[source]¶

Bases: twodlearn.core.common.TdlDescriptor

Decorator used to specify a parameter inside a model. The decorator works similar to @property, but the specified method correponds to the initialization of the parameter

class twodlearn.core.common.Submodel(finit=None, doc=None, name=None)[source]¶

Bases: twodlearn.core.common.TdlDescriptor

Decorator used to specify a submodel inside a model. The decorator works similar to @property, but the specified method correponds to the initialization of the submodel.

class twodlearn.core.common.SubmodelInit(finit=None, inference_input=None, lazzy=False, doc=None)[source]¶

Bases: object

Indicate the initialization function for a property

Examples

Some examples of how can be used

class TestObject0(tdl.common.TdlObject):
    @tdl.core.SubmodelInit
    def submodel(self, x, y):
        return tdl.core.SimpleNamespace(x=x, y=y)

class TestObject1(tdl.common.TdlObject):
    submodel = tdl.core.SubmodelInit(inference_input=True)

    @submodel.initializer
    def submodel(self, x, y):
        return tdl.core.SimpleNamespace(x=x, y=y)

class TestObject2(tdl.common.TdlObject):
    @tdl.core.SubmodelInit(inference_input=True)
    def submodel(self, x, y):
        return tdl.core.SimpleNamespace(x=x, y=y)

class Initializer(obj, attr_name, finit)[source]¶

Bases: object

init(*args, **kargs)[source]¶

set_args(given_args)[source]¶

autoinit(obj, force=False)[source]¶

init(obj, val)[source]¶

initialization method called when TdlModel is initialized

Parameters

obj (TdlModel) – object to which the property belongs to.
val (type) – value used for initialization. If value is a dictionary, the initialization method will be called using the dictionary values, otherwise, the property will be set with the provided value.

initializer(finit)[source]¶

class twodlearn.core.common.SubmodelWithArgs(finit=None, doc=None, name=None)[source]¶

Bases: twodlearn.core.common.Submodel

Decorator used to specify a submodel inside a model. The decorator works similar to @property, but the specified method correponds to the initialization of the submodel.

init(obj, val)[source]¶

class twodlearn.core.common.TdlDescriptor(finit=None, doc=None, name=None)[source]¶

Bases: object

Decorator used to specify a parameter inside a model. The decorator works similar to @property, but the specified method correponds to the initialization of the parameter

autoinit(obj, force=False)[source]¶: initialized the attribute with None

init(obj, val)[source]¶

classmethod optional(name, doc=None, default=None)[source]¶

classmethod required(name, doc=None)[source]¶

class twodlearn.core.common.TdlModel(**kargs)[source]¶

Bases: twodlearn.core.common.TdlOp

add_weight(name, initializer, shape, trainable, **kargs)[source]¶

class twodlearn.core.common.TdlModelCallable(**kargs)[source]¶: Bases: twodlearn.core.common.TdlModel

class twodlearn.core.common.TdlObject(**kargs)[source]¶: Bases: object

class twodlearn.core.common.TdlOp(name, options=None)[source]¶

Bases: object

Base class for defining operations The operation is encapsulated inside a scope

property name[source]¶: name for the model

property options[source]¶

property scope[source]¶: scope for the model, used to define all operations

class twodlearn.core.common.TdlProgram(options=None, name=None, **kargs)[source]¶

Bases: object

Defines a program that executes operations over TdlModel instances

property name[source]¶: name for the model

property options[source]¶

property scope[source]¶: scope for the model, used to define all operations

class twodlearn.core.common.TransformedVariable(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)[source]¶

Bases: twodlearn.core.common.TdlModel

property initializer[source]¶

inverse(value)[source]¶

raw[source]¶: raw value before making a transformation

property shape[source]¶

transform(value)[source]¶

property value[source]¶

class twodlearn.core.common.UnsetProperty(obj, attr_name, finit)[source]¶

Bases: object

default_init()[source]¶

init(*args, **kargs)[source]¶

is_set = False[source]¶

twodlearn.core.common.add_init_descriptor(descriptor)[source]¶: Adds the descriptor to the list of initialization descriptors This list is checked by the class initializers to verify if the provided arguments have an initialization method.

twodlearn.core.common.add_tdl_descriptor(descriptor)[source]¶: Adds the descriptor to the list of existing descriptors This list is checked by the class initializers to verify if the provided arguments have an initialization method

twodlearn.core.common.any_initialized(obj, props: List[str]) → bool[source]¶

checks if any property is initialized

Parameters

obj – object that owns the properties.
props (typing.List[str]) – list of properties to be checked.

Returns

True if any property is already initialized.

Return type

bool

twodlearn.core.common.assert_any_available(object, property=None, reqs=None)[source]¶

Check if requirements are available.

This function checks if any of the requirements is already set. If no requirement is set, we rise an exeption and left _init_tdl_attrs function to handle it.

The requirements will be initialized only if they were provided by the user, no auto-initialization is performed.

Parameters

object (TdlModel) – object being initialized/defined.
reqs (str) – requirements.

twodlearn.core.common.assert_initialized(object, prop, reqs)[source]¶

Check if the requirements have been initialized.

Parameters

object – object being initialized
prop – property being initialized (string)
reqs – list of properties that are required to initialize prop (list of strings)

Raises

InitPreconditionsFailed – the exception is raised if any of the requirements is not initialized. During initialization, the exeption is handled by _init_tdl_attrs

twodlearn.core.common.assert_initialized_if_available(object, property=None, reqs=None)[source]¶

check that requirements are initialized if they were provided by the user, no auto-initialization is performed.

Parameters

object (TdlModel) – object being initialized/defined.
reqs (str) – requirements.

twodlearn.core.common.build(model, recursive=False)[source]¶

initialize parameters and submodels of a given model.

Parameters: recursive – True for building the parameters and submodels.

twodlearn.core.common.check_defaults(options, default)[source]¶

Adds the values in default to options.

Parameters

options (dict) – base options.
default (dict) – default options.

Returns

options with the missing values that are in default but not in options.

Return type

dict

twodlearn.core.common.convert_output_to_tensor(value, dtype=None, name=None, as_ref=False)[source]¶

twodlearn.core.common.convert_variable_to_tensor(value, dtype=None, name=None, as_ref=False)[source]¶

twodlearn.core.common.create_init_docstring(cls)[source]¶: Add a docstring to the initialization method of a tdl class.

class twodlearn.core.common.encapsulate_op(input_vars, output_vars=None)[source]¶: Bases: object

twodlearn.core.common.hasattr(object, name)[source]¶: Checks if the object has the attribute name bypassing the autoinitialization procedure.

twodlearn.core.common.init_attrs(model, attrs=None, AttrTypes='default')[source]¶

Run auto-initialization of the model attributes.

Parameters

model (TdlModel) – model to initialize attributes.
attrs (list) – list with the names of the attributes to initialize.
AttrTypes (str) – tdl decorator or list of tdl decorators. Use ‘default’ to initialize (InputArgument, InputParameter, InputModel, SimpleParameter, Submodel).

twodlearn.core.common.is_property_initialized(obj, prop)[source]¶

Checks if a property has been initialized. This function checks initialization even for properties that are set but not initialized, like MethodInit.

Parameters

obj (TdlModel) – model object.
prop (str) – property name.

Returns

True if property is set, False otherwise.

Return type

bool

twodlearn.core.common.is_property_set(obj, prop)[source]¶

Checks if a property has already been set. This function checks only if the property has been set. For attributes like MethodInit that are set using an initializer, this function returns True, even when they have not been initialized.

Parameters

obj (TdlModel) – model object.
prop (str) – property name.

Returns

True if property is set, False otherwise.

Return type

bool

twodlearn.core.common.merge_dicts(a, b)[source]¶

class twodlearn.core.common.reuse_scope[source]¶: Bases: object

twodlearn.core.common.tensor_rank(tensor)[source]¶