analogvnn.parameter.PseudoParameter#

Module Contents#

Classes#

PseudoParameter

A parameterized parameter which acts like a normal parameter during gradient updates.
class analogvnn.parameter.PseudoParameter.PseudoParameter(data=None, requires_grad=True, transformation=None)[source]#

Bases: torch.nn.Module

A parameterized parameter which acts like a normal parameter during gradient updates.

PyTorch’s ParameterizedParameters vs AnalogVNN’s PseudoParameters:

  • Similarity (Forward or Parameterizing the data):

    > Data -> ParameterizingModel -> Parameterized Data

  • Difference (Backward or Gradient Calculations): - ParameterizedParameters

    > Parameterized Data -> ParameterizingModel -> Data

    • PseudoParameters > Parameterized Data -> Data

Variables:

  • _transformation (Callable) – the transformation.

  • _transformed (nn.Parameter) – the transformed parameter.

Properties:

grad (Tensor): the gradient of the parameter. module (PseudoParameterModule): the module that wraps the parameter and the transformation. transformation (Callable): the transformation.

property transformation[source]#

Returns the transformation.

Returns:

the transformation.

Return type:

Callable

_transformation: Callable[source]#
_transformed: torch.nn.Parameter[source]#
forward[source]#

Alias for __call__

_call_impl[source]#

Alias for __call__

right_inverse[source]#

Alias for set_original_data.

static identity(x: Any) → Any[source]#

The identity function.

Parameters:

x (Any) – the input tensor.

Returns:

the input tensor.

Return type:

Any

__call__(*args, **kwargs)[source]#

Transforms the parameter.

Parameters:

  • *args – additional arguments.

  • **kwargs – additional keyword arguments.

Returns:

the transformed parameter.

Return type:

nn.Parameter

Raises:

RuntimeError – if the transformation callable fails.

set_original_data(data: torch.Tensor) → PseudoParameter[source]#

Set data to the original parameter.

Parameters:

data (Tensor) – the data to set.

Returns:

self.

Return type:

PseudoParameter

__repr__()[source]#

Returns a string representation of the parameter.

Returns:

the string representation.

Return type:

str

set_transformation(transformation) → PseudoParameter[source]#

Sets the transformation.

Parameters:

transformation (Callable) – the transformation.

Returns:

self.

Return type:

PseudoParameter

static substitute_member(tensor_from: Any, tensor_to: Any, property_name: str, setter: bool = True)[source]#

Substitutes a member of a tensor as property of another tensor.

Parameters:

  • tensor_from (Any) – the tensor property to substitute.

  • tensor_to (Any) – the tensor property to substitute to.

  • property_name (str) – the name of the property.

  • setter (bool) – whether to substitute the setter.

classmethod parameterize(module: torch.nn.Module, param_name: str, transformation: Callable) → PseudoParameter[source]#

Parameterizes a parameter.

Parameters:

  • module (nn.Module) – the module.

  • param_name (str) – the name of the parameter.

  • transformation (Callable) – the transformation to apply.

Returns:

the parameterized parameter.

Return type:

PseudoParameter

classmethod parametrize_module(module: torch.nn.Module, transformation: Callable, requires_grad: bool = True)[source]#

Parametrize all parameters of a module.

Parameters:

  • module (nn.Module) – the module parameters to parametrize.

  • transformation (Callable) – the transformation.

  • requires_grad (bool) – if True, only parametrized parameters that require gradients.