"""File with LRP algorithm explainer classes.
Based on https://github.com/pytorch/captum/blob/master/captum/attr/_core/lrp.py
and https://github.com/pytorch/captum/blob/master/captum/attr/_core/layer/layer_lrp.py.
"""
from abc import abstractmethod
from typing import Any, Optional, Union
import torch
from captum._utils.typing import TargetType
from captum.attr import LRP, LayerLRP
from captum.attr._utils.lrp_rules import EpsilonRule, GammaRule
from foxai.array_utils import validate_result
from foxai.explainer.base_explainer import Explainer
from foxai.explainer.computer_vision.model_utils import (
get_last_conv_model_layer,
modify_modules,
)
[docs]class BaseLRPCVExplainer(Explainer):
"""Base LRP algorithm explainer."""
[docs] @abstractmethod
def create_explainer(
self,
model: torch.nn.Module,
**kwargs,
) -> Union[LRP, LayerLRP]:
"""Create explainer object.
Args:
model: The forward function of the model or any
modification of it.
Returns:
Explainer object.
"""
[docs] def calculate_features(
self,
model: torch.nn.Module,
input_data: torch.Tensor,
pred_label_idx: TargetType = None,
additional_forward_args: Any = None,
attribute_to_layer_input: bool = False,
verbose: bool = False,
**kwargs,
) -> torch.Tensor:
"""Generate model's attributes with LRP algorithm explainer.
Args:
model: The forward function of the model or any
modification of it.
input_data: Input for which relevance is
propagated. If forward_func takes a single
tensor as input, a single input tensor should be provided.
pred_label_idx: Output indices for
which gradients are computed (for classification cases,
this is usually the target class).
If the network returns a scalar value per example,
no target index is necessary.
For general 2D outputs, targets can be either:
- a single integer or a tensor containing a single
integer, which is applied to all input examples
- a list of integers or a 1D tensor, with length matching
the number of examples in inputs (dim 0). Each integer
is applied as the target for the corresponding example.
For outputs with > 2 dimensions, targets can be either:
- A single tuple, which contains #output_dims - 1
elements. This target index is applied to all examples.
- A list of tuples with length equal to the number of
examples in inputs (dim 0), and each tuple containing
#output_dims - 1 elements. Each tuple is applied as the
target for the corresponding example.
Default: None
additional_forward_args: If the forward function
requires additional arguments other than the inputs for
which attributions should not be computed, this argument
can be provided. It must be either a single additional
argument of a Tensor or arbitrary (non-tuple) type or a tuple
containing multiple additional arguments including tensors
or any arbitrary python types. These arguments are provided to
forward_func in order, following the arguments in inputs.
Note that attributions are not computed with respect
to these arguments.
Default: None
attribute_to_layer_input: Indicates whether to
compute the attribution with respect to the layer input
or output. If `attribute_to_layer_input` is set to True
then the attributions will be computed with respect to
layer input, otherwise it will be computed with respect
to layer output.
verbose: Indicates whether information on application
of rules is printed during propagation.
Default: False
Returns:
Features matrix.
Raises:
RuntimeError: if attribution has shape (0).
"""
layer: Optional[torch.nn.Module] = kwargs.get("layer", None)
lrp = self.create_explainer(model=model, layer=layer)
if isinstance(lrp, LRP):
attributions = lrp.attribute(
input_data,
target=pred_label_idx,
additional_forward_args=additional_forward_args,
return_convergence_delta=False,
verbose=verbose,
)
else:
attributions = lrp.attribute(
input_data,
target=pred_label_idx,
additional_forward_args=additional_forward_args,
return_convergence_delta=False,
attribute_to_layer_input=attribute_to_layer_input,
verbose=verbose,
)
validate_result(attributions=attributions)
return attributions
[docs] def add_rules(self, model: torch.nn.Module) -> torch.nn.Module:
"""Add rules for the LRP explainer,
according to https://arxiv.org/pdf/1910.09840.pdf.
Args:
model: The forward function of the model or any
modification of it.
Returns:
Modified DNN object.
"""
layers_number: int = len(list(model.modules()))
for idx_layer, module in enumerate(model.modules()):
if idx_layer <= layers_number // 2:
setattr(module, "rule", GammaRule())
elif idx_layer != (layers_number - 1):
setattr(module, "rule", EpsilonRule())
else:
setattr(module, "rule", EpsilonRule(epsilon=0)) # LRP-0
return model
[docs]class LRPCVExplainer(BaseLRPCVExplainer):
"""LRP algorithm explainer."""
[docs] def create_explainer(
self,
model: torch.nn.Module,
**kwargs,
) -> Union[LRP, LayerLRP]:
"""Create explainer object.
Args:
model: The forward function of the model or any
modification of it.
Returns:
Explainer object.
"""
model = self.add_rules(modify_modules(model))
return LRP(model=model)
[docs]class LayerLRPCVExplainer(BaseLRPCVExplainer):
"""Layer LRP algorithm explainer."""
[docs] def create_explainer(
self,
model: torch.nn.Module,
layer: Optional[torch.nn.Module] = None,
**kwargs,
) -> Union[LRP, LayerLRP]:
"""Create explainer object.
Uses parameter `layer` from `kwargs`. If not provided function will call
`get_last_conv_model_layer` function to obtain last `torch.nn.Conv2d` layer
from provided model.
Args:
model: The forward function of the model or any
modification of it.
layer: Layer for which attributions are computed.
Output size of attribute matches this layer's input or
output dimensions, depending on whether we attribute to
the inputs or outputs of the layer, corresponding to
attribution of each neuron in the input or output of
this layer.
Default: None
Returns:
Explainer object.
Raises:
ValueError: if model does not contain conv layers.
"""
if layer is None:
layer = get_last_conv_model_layer(model=model)
model = self.add_rules(modify_modules(model))
return LayerLRP(model=model, layer=layer)