#!/usr/bin/env python3
# pyre-strict
from enum import Enum
from typing import Any, Callable, cast, Dict, List, Optional, Sequence, Tuple, Union
import torch
from captum._utils.common import (
_expand_and_update_additional_forward_args,
_expand_and_update_baselines,
_expand_and_update_feature_mask,
_expand_and_update_target,
_format_output,
_format_tensor_into_tuples,
_is_tuple,
)
from captum._utils.typing import TensorOrTupleOfTensorsGeneric
from captum.attr._utils.attribution import Attribution, GradientAttribution
from captum.attr._utils.common import _validate_noise_tunnel_type
from captum.log import log_usage
from torch import Tensor
class NoiseTunnelType(Enum):
smoothgrad = 1
smoothgrad_sq = 2
vargrad = 3
SUPPORTED_NOISE_TUNNEL_TYPES: List[str] = list(NoiseTunnelType.__members__.keys())
[docs]
class NoiseTunnel(Attribution):
r"""
Adds gaussian noise to each input in the batch `nt_samples` times
and applies the given attribution algorithm to each of the samples.
The attributions of the samples are combined based on the given noise
tunnel type (nt_type):
If nt_type is `smoothgrad`, the mean of the sampled attributions is
returned. This approximates smoothing the given attribution method
with a Gaussian Kernel.
If nt_type is `smoothgrad_sq`, the mean of the squared sample attributions
is returned.
If nt_type is `vargrad`, the variance of the sample attributions is
returned.
More details about adding noise can be found in the following papers:
* https://arxiv.org/abs/1810.03292
* https://arxiv.org/abs/1810.03307
* https://arxiv.org/abs/1706.03825
* https://arxiv.org/abs/1806.10758
This method currently also supports batches of multiple examples input,
however it can be computationally expensive depending on the model,
the dimensionality of the data and execution environment.
It is assumed that the batch size is the first dimension of input tensors.
"""
is_delta_supported: bool
_multiply_by_inputs: bool
is_gradient_method: bool
def __init__(self, attribution_method: Attribution) -> None:
r"""
Args:
attribution_method (Attribution): An instance of any attribution algorithm
of type `Attribution`. E.g. Integrated Gradients,
Conductance or Saliency.
"""
self.attribution_method = attribution_method
self.is_delta_supported = self.attribution_method.has_convergence_delta()
self._multiply_by_inputs = self.attribution_method.multiplies_by_inputs
self.is_gradient_method = isinstance(
self.attribution_method, GradientAttribution
)
Attribution.__init__(self, self.attribution_method.forward_func)
@property
def multiplies_by_inputs(self) -> bool:
return self._multiply_by_inputs
[docs]
@log_usage()
def attribute(
self,
inputs: Union[Tensor, Tuple[Tensor, ...]],
nt_type: str = "smoothgrad",
nt_samples: int = 5,
nt_samples_batch_size: Optional[int] = None,
stdevs: Union[float, Tuple[float, ...]] = 1.0,
draw_baseline_from_distrib: bool = False,
**kwargs: Any,
) -> Union[
Union[
Tensor,
Tuple[Tensor, Tensor],
Tuple[Tensor, ...],
Tuple[Tuple[Tensor, ...], Tensor],
]
]:
r"""
Args:
inputs (Tensor or tuple[Tensor, ...]): Input for which integrated
gradients are computed. If forward_func takes a single
tensor as input, a single input tensor should be provided.
If forward_func takes multiple tensors as input, a tuple
of the input tensors should be provided. It is assumed
that for all given input tensors, dimension 0 corresponds
to the number of examples, and if multiple input tensors
are provided, the examples must be aligned appropriately.
nt_type (str, optional): Smoothing type of the attributions.
`smoothgrad`, `smoothgrad_sq` or `vargrad`
Default: `smoothgrad` if `type` is not provided.
nt_samples (int, optional): The number of randomly generated examples
per sample in the input batch. Random examples are
generated by adding gaussian random noise to each sample.
Default: `5` if `nt_samples` is not provided.
nt_samples_batch_size (int, optional): The number of the `nt_samples`
that will be processed together. With the help
of this parameter we can avoid out of memory situation and
reduce the number of randomly generated examples per sample
in each batch.
Default: None if `nt_samples_batch_size` is not provided. In
this case all `nt_samples` will be processed together.
stdevs (float or tuple of float, optional): The standard deviation
of gaussian noise with zero mean that is added to each
input in the batch. If `stdevs` is a single float value
then that same value is used for all inputs. If it is
a tuple, then it must have the same length as the inputs
tuple. In this case, each stdev value in the stdevs tuple
corresponds to the input with the same index in the inputs
tuple.
Default: `1.0` if `stdevs` is not provided.
draw_baseline_from_distrib (bool, optional): Indicates whether to
randomly draw baseline samples from the `baselines`
distribution provided as an input tensor.
Default: False
**kwargs (Any, optional): Contains a list of arguments that are passed
to `attribution_method` attribution algorithm.
Any additional arguments that should be used for the
chosen attribution method should be included here.
For instance, such arguments include
`additional_forward_args` and `baselines`.
Returns:
**attributions** or 2-element tuple of **attributions**, **delta**:
- **attributions** (*Tensor* or *tuple[Tensor, ...]*):
Attribution with
respect to each input feature. attributions will always be
the same size as the provided inputs, with each value
providing the attribution of the corresponding input index.
If a single tensor is provided as inputs, a single tensor is
returned. If a tuple is provided for inputs, a tuple of
corresponding sized tensors is returned.
- **delta** (*float*, returned if return_convergence_delta=True):
Approximation error computed by the
attribution algorithm. Not all attribution algorithms
return delta value. It is computed only for some
algorithms, e.g. integrated gradients.
Delta is computed for each input in the batch
and represents the arithmetic mean
across all `nt_samples` perturbed tensors for that input.
Examples::
>>> # ImageClassifier takes a single input tensor of images Nx3x32x32,
>>> # and returns an Nx10 tensor of class probabilities.
>>> net = ImageClassifier()
>>> ig = IntegratedGradients(net)
>>> input = torch.randn(2, 3, 32, 32, requires_grad=True)
>>> # Creates noise tunnel
>>> nt = NoiseTunnel(ig)
>>> # Generates 10 perturbed input tensors per image.
>>> # Computes integrated gradients for class 3 for each generated
>>> # input and averages attributions across all 10
>>> # perturbed inputs per image
>>> attribution = nt.attribute(input, nt_type='smoothgrad',
>>> nt_samples=10, target=3)
"""
return_convergence_delta: bool = (
"return_convergence_delta" in kwargs and kwargs["return_convergence_delta"]
)
with torch.no_grad():
nt_samples_batch_size = (
nt_samples
if nt_samples_batch_size is None
else min(nt_samples, nt_samples_batch_size)
)
nt_samples_partition = nt_samples // nt_samples_batch_size
# Keeps track whether original input is a tuple or not before
# converting it into a tuple.
is_inputs_tuple = isinstance(inputs, tuple)
inputs = _format_tensor_into_tuples(inputs) # type: ignore
_validate_noise_tunnel_type(nt_type, SUPPORTED_NOISE_TUNNEL_TYPES)
kwargs_copy = kwargs.copy()
self._expand_partial(
nt_samples_batch_size, kwargs_copy, inputs, draw_baseline_from_distrib
)
sum_attributions: Sequence[Union[None, Tensor]] = []
sum_attributions_sq: Sequence[Union[None, Tensor]] = []
delta_partial_list: List[Tensor] = []
is_attrib_tuple = is_inputs_tuple
for _ in range(nt_samples_partition):
inputs_with_noise = self._add_noise_to_inputs(
nt_samples_batch_size, inputs, stdevs
)
(
attributions_partial,
is_attrib_tuple,
delta_partial,
) = self._compute_partial_attribution(
inputs_with_noise,
kwargs_copy,
is_inputs_tuple,
return_convergence_delta,
)
if len(sum_attributions) == 0:
sum_attributions = [None] * len(attributions_partial)
sum_attributions_sq = [None] * len(attributions_partial)
self._update_partial_attribution_and_delta(
attributions_partial,
cast(Tensor, delta_partial),
cast(List[Tensor], sum_attributions),
cast(List[Tensor], sum_attributions_sq),
delta_partial_list,
nt_samples_batch_size,
return_convergence_delta,
)
nt_samples_remaining = (
nt_samples - nt_samples_partition * nt_samples_batch_size
)
if nt_samples_remaining > 0:
inputs_with_noise = self._add_noise_to_inputs(
nt_samples_remaining, inputs, stdevs
)
self._expand_partial(
nt_samples_remaining, kwargs, inputs, draw_baseline_from_distrib
)
(
attributions_partial,
is_attrib_tuple,
delta_partial,
) = self._compute_partial_attribution(
inputs_with_noise,
kwargs,
is_inputs_tuple,
return_convergence_delta,
)
self._update_partial_attribution_and_delta(
attributions_partial,
cast(Tensor, delta_partial),
cast(List[Tensor], sum_attributions),
cast(List[Tensor], sum_attributions_sq),
delta_partial_list,
nt_samples_remaining,
return_convergence_delta,
)
expected_attributions = tuple(
[
cast(Tensor, sum_attribution) * 1 / nt_samples
for sum_attribution in sum_attributions
]
)
expected_attributions_sq = tuple(
[
cast(Tensor, sum_attribution_sq) * 1 / nt_samples
for sum_attribution_sq in sum_attributions_sq
]
)
attributions = self._compute_smoothing(
expected_attributions,
expected_attributions_sq,
nt_type,
)
delta = None
if self.is_delta_supported and return_convergence_delta:
delta = torch.cat(delta_partial_list, dim=0)
return self._apply_checks_and_return_attributions(
attributions,
is_attrib_tuple,
return_convergence_delta,
delta,
)
# pyre-fixme[24] Generic type `Callable` expects 2 type parameters.
[docs]
def attribute_future(self) -> Callable:
r"""
This method is not implemented for NoiseTunnel.
"""
raise NotImplementedError("attribute_future is not implemented for NoiseTunnel")
def _add_noise_to_inputs(
self,
nt_samples_partition: int,
inputs: Tuple[Tensor, ...],
stdevs: Union[float, Tuple[float, ...]],
) -> Tuple[Tensor, ...]:
if isinstance(stdevs, tuple):
assert len(stdevs) == len(inputs), (
"The number of input tensors "
"in {} must be equal to the number of stdevs values {}".format(
len(inputs), len(stdevs)
)
)
stdevs_ = stdevs
else:
assert isinstance(
stdevs, float
), "stdevs must be type float. " "Given: {}".format(type(stdevs))
stdevs_ = (stdevs,) * len(inputs)
return tuple(
(
self._add_noise_to_input(
input, stdev, nt_samples_partition
).requires_grad_()
if self.is_gradient_method
else self._add_noise_to_input(input, stdev, nt_samples_partition)
)
for (input, stdev) in zip(inputs, stdevs_)
)
@staticmethod
def _add_noise_to_input(
input: Tensor, stdev: float, nt_samples_partition: int
) -> Tensor:
# batch size
bsz = input.shape[0]
# expand input size by the number of drawn samples
input_expanded_size = (bsz * nt_samples_partition,) + tuple(input.shape[1:])
# expand stdev for the shape of the input and number of drawn samples
stdev_expanded = torch.tensor(stdev, device=input.device).repeat(
input_expanded_size
)
# draws `np.prod(input_expanded_size)` samples from normal distribution
# with given input parametrization
# FIXME it look like it is very difficult to make torch.normal
# deterministic this needs an investigation
noise = torch.normal(0, stdev_expanded)
return input.repeat_interleave(nt_samples_partition, dim=0) + noise
@staticmethod
def _update_sum_attribution_and_sq(
sum_attribution: List[Tensor],
sum_attribution_sq: List[Tensor],
attribution: Tensor,
i: int,
nt_samples_batch_size_inter: int,
) -> None:
bsz = attribution.shape[0] // nt_samples_batch_size_inter
attribution_shape = cast(Tuple[int, ...], (bsz, nt_samples_batch_size_inter))
if len(attribution.shape) > 1:
attribution_shape += tuple(attribution.shape[1:])
attribution = attribution.view(attribution_shape)
current_attribution_sum = attribution.sum(dim=1, keepdim=False)
current_attribution_sq = torch.sum(
torch.pow(attribution, 2), dim=1, keepdim=False
)
sum_attribution[i] = (
current_attribution_sum
if not isinstance(sum_attribution[i], torch.Tensor)
else sum_attribution[i] + current_attribution_sum
)
sum_attribution_sq[i] = (
current_attribution_sq
if not isinstance(sum_attribution_sq[i], torch.Tensor)
else sum_attribution_sq[i] + current_attribution_sq
)
def _compute_partial_attribution(
self,
inputs_with_noise_partition: Tuple[Tensor, ...],
kwargs_partition: object,
is_inputs_tuple: bool,
return_convergence_delta: bool,
) -> Tuple[Tuple[Tensor, ...], bool, Union[None, Tensor]]:
attr_func = self.attribution_method.attribute
# smoothgrad_Attr(x) = 1 / n * sum(Attr(x + N(0, sigma^2))
# NOTE: using __wrapped__ such that it does not log the inner logs
attributions = attr_func.__wrapped__( # type: ignore
self.attribution_method, # self
(
inputs_with_noise_partition
if is_inputs_tuple
else inputs_with_noise_partition[0]
),
**kwargs_partition,
)
delta = None
if self.is_delta_supported and return_convergence_delta:
attributions, delta = attributions
is_attrib_tuple = _is_tuple(attributions)
attributions = _format_tensor_into_tuples(attributions)
return (
cast(Tuple[Tensor, ...], attributions),
cast(bool, is_attrib_tuple),
delta,
)
@staticmethod
def _expand_partial(
nt_samples_partition: int,
kwargs_partial: Dict[str, Any],
inputs: Tuple[Tensor, ...],
draw_baseline_from_distrib: bool,
) -> None:
# if the algorithm supports targets, baselines and/or
# additional_forward_args they will be expanded based
# on the nt_samples_partition and corresponding kwargs
# variables will be updated accordingly
_expand_and_update_additional_forward_args(nt_samples_partition, kwargs_partial)
_expand_and_update_target(nt_samples_partition, kwargs_partial)
_expand_and_update_baselines(
inputs,
nt_samples_partition,
kwargs_partial,
draw_baseline_from_distrib=draw_baseline_from_distrib,
)
_expand_and_update_feature_mask(nt_samples_partition, kwargs_partial)
@staticmethod
def _compute_smoothing(
expected_attributions: Tuple[Union[Tensor], ...],
expected_attributions_sq: Tuple[Union[Tensor], ...],
nt_type: str,
) -> Tuple[Tensor, ...]:
if NoiseTunnelType[nt_type] == NoiseTunnelType.smoothgrad:
return expected_attributions
if NoiseTunnelType[nt_type] == NoiseTunnelType.smoothgrad_sq:
return expected_attributions_sq
vargrad = tuple(
expected_attribution_sq - expected_attribution * expected_attribution
for expected_attribution, expected_attribution_sq in zip(
expected_attributions, expected_attributions_sq
)
)
return vargrad
def _update_partial_attribution_and_delta(
self,
attributions_partial: Tuple[Tensor, ...],
delta_partial: Tensor,
sum_attributions: List[Tensor],
sum_attributions_sq: List[Tensor],
delta_partial_list: List[Tensor],
nt_samples_partial: int,
return_convergence_delta: bool,
) -> None:
for i, attribution_partial in enumerate(attributions_partial):
self._update_sum_attribution_and_sq(
sum_attributions,
sum_attributions_sq,
attribution_partial,
i,
nt_samples_partial,
)
if self.is_delta_supported and return_convergence_delta:
delta_partial_list.append(delta_partial)
def _apply_checks_and_return_attributions(
self,
attributions: Tuple[Tensor, ...],
is_attrib_tuple: bool,
return_convergence_delta: bool,
delta: Union[None, Tensor],
# pyre-fixme[34]: `Variable[TensorOrTupleOfTensorsGeneric <:
# [torch._tensor.Tensor, typing.Tuple[torch._tensor.Tensor, ...]]]`
# isn't present in the function's parameters.
) -> Union[
TensorOrTupleOfTensorsGeneric, Tuple[TensorOrTupleOfTensorsGeneric, Tensor]
]:
attributions_tuple = _format_output(is_attrib_tuple, attributions)
ret = (
(attributions_tuple, cast(Tensor, delta))
if self.is_delta_supported and return_convergence_delta
else attributions_tuple
)
ret = cast(
# pyre-fixme[34]: `Variable[TensorOrTupleOfTensorsGeneric <:
# [torch._tensor.Tensor, typing.Tuple[torch._tensor.Tensor, ...]]]`
# isn't present in the function's parameters.
Union[
TensorOrTupleOfTensorsGeneric,
Tuple[TensorOrTupleOfTensorsGeneric, Tensor],
],
ret,
)
return ret
[docs]
def has_convergence_delta(self) -> bool:
return self.is_delta_supported
