Skip to content

IntegratedGradients

pnpxai.explainers.integrated_gradients

IntegratedGradients

Bases: Explainer

IntegratedGradients explainer.

Supported Modules: Linear, Convolution, Attention

Parameters:

Name Type Description Default
model Module

The PyTorch model for which attribution is to be computed.

required
baseline_fn Union[BaselineMethodOrFunction, Tuple[BaselineMethodOrFunction]]

The baseline function, accepting the attribution input, and returning the baseline accordingly.

'zeros'
n_steps int

The Number of steps the algorithm makes

20
layer Optional[Union[Union[str, Module], Sequence[Union[str, Module]]]]

The target module to be explained

None
n_classes Optional[int]

Number of classes

required
forward_arg_extractor Optional[Callable[[Tuple[Tensor]], Union[Tensor, Tuple[Tensor]]]]

A function that extracts forward arguments from the input batch(s) where the attribution scores are assigned.

None
additional_forward_arg_extractor Optional[Callable[[Tuple[Tensor]], Union[Tensor, Tuple[Tensor]]]]

A secondary function that extract additional forward arguments from the input batch(s).

None
**kwargs

Keyword arguments that are forwarded to the base implementation of the Explainer

required
Reference

Mukund Sundararajan, Ankur Taly, Qiqi Yan. Axiomatic Attribution for Deep Networks.

SUPPORTED_MODULES = [Linear, Convolution, Attention] class-attribute instance-attribute
layer = layer instance-attribute
n_steps = n_steps instance-attribute
baseline_fn = baseline_fn instance-attribute
explainer: Union[CaptumIntegratedGradients, CaptumLayerIntegratedGradients] property
EXPLANATION_TYPE: ExplanationType = 'attribution' class-attribute instance-attribute
TUNABLES = {} class-attribute instance-attribute
model = model.eval() instance-attribute
forward_arg_extractor = forward_arg_extractor instance-attribute
additional_forward_arg_extractor = additional_forward_arg_extractor instance-attribute
device property
__init__(model: Module, n_steps: int = 20, baseline_fn: Union[BaselineMethodOrFunction, Tuple[BaselineMethodOrFunction]] = 'zeros', layer: Optional[Callable[[Tuple[Tensor]], Union[Tensor, Tuple[Tensor]]]] = None, forward_arg_extractor: Optional[Callable[[Tuple[Tensor]], Union[Tensor, Tuple[Tensor]]]] = None, additional_forward_arg_extractor: Optional[Callable[[Tuple[Tensor]], Union[Tensor, Tuple[Tensor]]]] = None) -> None
attribute(inputs: Union[Tensor, Tuple[Tensor]], targets: Tensor) -> Union[Tensor, Tuple[Tensor]]

Computes attributions for the given inputs and targets.

Parameters:

Name Type Description Default
inputs Tensor

The input data.

required
targets Tensor

The target labels for the inputs.

required

Returns:

Type Description
Union[Tensor, Tuple[Tensor]]

Union[torch.Tensor, Tuple[torch.Tensor]]: The result of the explanation.

get_tunables() -> Dict[str, Tuple[type, dict]]

Provides Tunable parameters for the optimizer

Tunable parameters

noise_level (float): Value can be selected in the range of range(10, 100, 10)

baseline_fn (callable): BaselineFunction selects suitable values in accordance with the modality

__repr__()
copy()
set_kwargs(**kwargs)