#!/usr/bin/env python3
# pyre-strict
import math
from typing import Any, Callable, cast, Generator, List, Optional, Tuple, Union
import torch
from captum._utils.common import (
_expand_additional_forward_args,
_expand_target,
_format_additional_forward_args,
_format_feature_mask,
_format_output,
_is_tuple,
_run_forward,
)
from captum._utils.exceptions import FeatureAblationFutureError
from captum._utils.progress import progress
from captum._utils.typing import BaselineType, TargetType, TensorOrTupleOfTensorsGeneric
from captum.attr._utils.attribution import PerturbationAttribution
from captum.attr._utils.common import _format_input_baseline
from captum.log import log_usage
from torch import dtype, Tensor
from torch.futures import collect_all, Future
[docs]
class FeatureAblation(PerturbationAttribution):
r"""
A perturbation based approach to computing attribution, involving
replacing each input feature with a given baseline / reference, and
computing the difference in output. By default, each scalar value within
each input tensor is taken as a feature and replaced independently. Passing
a feature mask, allows grouping features to be ablated together. This can
be used in cases such as images, where an entire segment or region
can be ablated, measuring the importance of the segment (feature group).
Each input scalar in the group will be given the same attribution value
equal to the change in target as a result of ablating the entire feature
group.
The forward function can either return a scalar per example or a tensor
of a fixed sized tensor (or scalar value) for the full batch, i.e. the
output does not grow as the batch size increase. If the output is fixed
we consider this model to be an "aggregation" of the inputs. In the fixed
sized output mode we require `perturbations_per_eval == 1` and the
`feature_mask` to be either `None` or for all of them to have 1 as their
first dimension (i.e. a feature mask requires to be applied to all inputs).
"""
def __init__(
self, forward_func: Callable[..., Union[int, float, Tensor, Future[Tensor]]]
) -> None:
r"""
Args:
forward_func (Callable): The forward function of the model or
any modification of it.
"""
PerturbationAttribution.__init__(self, forward_func)
self.use_weights = False
# only used when perturbations_per_eval > 1, where the 1st dim of forward_func's
# output must grow as the input batch size. If forward's output is aggregated,
# we cannot expand the input to include more perturbations in one call.
# If it's False, we will force the validation by comparing the outpus of
# the original input and the modified input whose batch size expanded based on
# perturbations_per_eval. Set the flag to True if the output of the modified
# input grow as expected. Once it turns to True, we will assume the model's
# behavior stays consistent and no longer check again
self._is_output_shape_valid = False
[docs]
@log_usage()
def attribute(
self,
inputs: TensorOrTupleOfTensorsGeneric,
baselines: BaselineType = None,
target: TargetType = None,
additional_forward_args: Optional[object] = None,
feature_mask: Union[None, Tensor, Tuple[Tensor, ...]] = None,
perturbations_per_eval: int = 1,
show_progress: bool = False,
**kwargs: Any,
) -> TensorOrTupleOfTensorsGeneric:
r"""
Args:
inputs (Tensor or tuple[Tensor, ...]): Input for which ablation
attributions 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 (aka batch size), and if
multiple input tensors are provided, the examples must
be aligned appropriately.
baselines (scalar, Tensor, tuple of scalar, or Tensor, optional):
Baselines define reference value which replaces each
feature when ablated.
Baselines can be provided as:
- a single tensor, if inputs is a single tensor, with
exactly the same dimensions as inputs or
broadcastable to match the dimensions of inputs
- a single scalar, if inputs is a single tensor, which will
be broadcasted for each input value in input tensor.
- a tuple of tensors or scalars, the baseline corresponding
to each tensor in the inputs' tuple can be:
- either a tensor with matching dimensions to
corresponding tensor in the inputs' tuple
or the first dimension is one and the remaining
dimensions match with the corresponding
input tensor.
- or a scalar, corresponding to a tensor in the
inputs' tuple. This scalar value is broadcasted
for corresponding input tensor.
In the cases when `baselines` is not provided, we internally
use zero scalar corresponding to each input tensor.
Default: None
target (int, tuple, Tensor, or list, optional): 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 (Any, optional): 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.
For a tensor, the first dimension of the tensor must
correspond to the number of examples. For all other types,
the given argument is used for all forward evaluations.
Note that attributions are not computed with respect
to these arguments.
Default: None
feature_mask (Tensor or tuple[Tensor, ...], optional):
feature_mask defines a mask for the input, grouping
features which should be ablated together. feature_mask
should contain the same number of tensors as inputs.
Each tensor should
be the same size as the corresponding input or
broadcastable to match the input tensor. Each tensor
should contain integers in the range 0 to num_features
- 1, and indices corresponding to the same feature should
have the same value.
Note that features within each input tensor are ablated
independently (not across tensors).
If the forward function returns a single scalar per batch,
we enforce that the first dimension of each mask must be 1,
since attributions are returned batch-wise rather than per
example, so the attributions must correspond to the
same features (indices) in each input example.
If None, then a feature mask is constructed which assigns
each scalar within a tensor as a separate feature, which
is ablated independently.
Default: None
perturbations_per_eval (int, optional): Allows ablation of multiple
features to be processed simultaneously in one call to
forward_fn.
Each forward pass will contain a maximum of
perturbations_per_eval * #examples samples.
For DataParallel models, each batch is split among the
available devices, so evaluations on each available
device contain at most
(perturbations_per_eval * #examples) / num_devices
samples.
If the forward function's number of outputs does not
change as the batch size grows (e.g. if it outputs a
scalar value), you must set perturbations_per_eval to 1
and use a single feature mask to describe the features
for all examples in the batch.
Default: 1
show_progress (bool, optional): Displays the progress of computation.
It will try to use tqdm if available for advanced features
(e.g. time estimation). Otherwise, it will fallback to
a simple output of progress.
Default: False
**kwargs (Any, optional): Any additional arguments used by child
classes of FeatureAblation (such as Occlusion) to construct
ablations. These arguments are ignored when using
FeatureAblation directly.
Default: None
Returns:
*Tensor* or *tuple[Tensor, ...]* of **attributions**:
- **attributions** (*Tensor* or *tuple[Tensor, ...]*):
The attributions with respect to each input feature.
If the forward function returns
a scalar value per example, attributions will be
the same size as the provided inputs, with each value
providing the attribution of the corresponding input index.
If the forward function returns a scalar per batch, then
attribution tensor(s) will have first dimension 1 and
the remaining dimensions will match the input.
If a single tensor is provided as inputs, a single tensor is
returned. If a tuple of tensors is provided for inputs, a
tuple of corresponding sized tensors is returned.
Examples::
>>> # SimpleClassifier takes a single input tensor of size Nx4x4,
>>> # and returns an Nx3 tensor of class probabilities.
>>> net = SimpleClassifier()
>>> # Generating random input with size 2 x 4 x 4
>>> input = torch.randn(2, 4, 4)
>>> # Defining FeatureAblation interpreter
>>> ablator = FeatureAblation(net)
>>> # Computes ablation attribution, ablating each of the 16
>>> # scalar input independently.
>>> attr = ablator.attribute(input, target=1)
>>> # Alternatively, we may want to ablate features in groups, e.g.
>>> # grouping each 2x2 square of the inputs and ablating them together.
>>> # This can be done by creating a feature mask as follows, which
>>> # defines the feature groups, e.g.:
>>> # +---+---+---+---+
>>> # | 0 | 0 | 1 | 1 |
>>> # +---+---+---+---+
>>> # | 0 | 0 | 1 | 1 |
>>> # +---+---+---+---+
>>> # | 2 | 2 | 3 | 3 |
>>> # +---+---+---+---+
>>> # | 2 | 2 | 3 | 3 |
>>> # +---+---+---+---+
>>> # With this mask, all inputs with the same value are ablated
>>> # simultaneously, and the attribution for each input in the same
>>> # group (0, 1, 2, and 3) per example are the same.
>>> # The attributions can be calculated as follows:
>>> # feature mask has dimensions 1 x 4 x 4
>>> feature_mask = torch.tensor([[[0,0,1,1],[0,0,1,1],
>>> [2,2,3,3],[2,2,3,3]]])
>>> attr = ablator.attribute(input, target=1, feature_mask=feature_mask)
"""
# Keeps track whether original input is a tuple or not before
# converting it into a tuple.
is_inputs_tuple = _is_tuple(inputs)
formatted_inputs, baselines = _format_input_baseline(inputs, baselines)
formatted_additional_forward_args = _format_additional_forward_args(
additional_forward_args
)
num_examples = formatted_inputs[0].shape[0]
formatted_feature_mask = _format_feature_mask(feature_mask, formatted_inputs)
assert (
isinstance(perturbations_per_eval, int) and perturbations_per_eval >= 1
), "Perturbations per evaluation must be an integer and at least 1."
with torch.no_grad():
if show_progress:
attr_progress = self._attribute_progress_setup(
formatted_inputs,
formatted_feature_mask,
**kwargs,
perturbations_per_eval=perturbations_per_eval,
)
attr_progress.update(0)
# Computes initial evaluation with all features, which is compared
# to each ablated result.
initial_eval: Union[Tensor, Future[Tensor]] = _run_forward(
self.forward_func,
formatted_inputs,
target,
formatted_additional_forward_args,
)
if show_progress:
attr_progress.update()
total_attrib: List[Tensor] = []
weights: List[Tensor] = []
flattened_initial_eval: Tensor
n_outputs: int
attrib_type: dtype
if isinstance(initial_eval, torch.Future):
raise AssertionError(
"when using the attribute function, initial_eval should have "
f"non-Future type rather than {type(initial_eval)}"
)
(
total_attrib,
weights,
initial_eval,
flattened_initial_eval,
n_outputs,
attrib_type,
) = self._process_initial_eval(
initial_eval,
formatted_inputs,
)
# Iterate through each feature tensor for ablation
for i in range(len(formatted_inputs)):
# Skip any empty input tensors
if torch.numel(formatted_inputs[i]) == 0:
continue
for (
current_inputs,
current_add_args,
current_target,
current_mask,
) in self._ith_input_ablation_generator(
i,
formatted_inputs,
formatted_additional_forward_args,
target,
baselines,
formatted_feature_mask,
perturbations_per_eval,
**kwargs,
):
# modified_eval has (n_feature_perturbed * n_outputs) elements
# shape:
# agg mode: (*initial_eval.shape)
# non-agg mode:
# (feature_perturbed * batch_size, *initial_eval.shape[1:])
modified_eval: Union[Tensor, Future[Tensor]] = _run_forward(
self.forward_func,
current_inputs,
current_target,
current_add_args,
)
if show_progress:
attr_progress.update()
assert not isinstance(modified_eval, torch.Future), (
"when use_futures is True, modified_eval should have "
f"non-Future type rather than {type(modified_eval)}"
)
total_attrib, weights = self._process_ablated_out(
modified_eval,
current_inputs,
current_mask,
perturbations_per_eval,
num_examples,
initial_eval,
flattened_initial_eval,
formatted_inputs,
n_outputs,
total_attrib,
weights,
i,
attrib_type,
)
if show_progress:
attr_progress.close()
# pyre-fixme[7]: Expected `Variable[TensorOrTupleOfTensorsGeneric <:
# [Tensor, typing.Tuple[Tensor, ...]]]`
# but got `Union[Tensor, typing.Tuple[Tensor, ...]]`.
return self._generate_result(total_attrib, weights, is_inputs_tuple) # type: ignore # noqa: E501 line too long
def _initial_eval_to_processed_initial_eval_fut(
self, initial_eval: Future[Tensor], formatted_inputs: Tuple[Tensor, ...]
) -> Tuple[List[Tensor], List[Tensor], Tensor, Tensor, int, dtype]:
try:
initial_eval_processed = initial_eval.value()
if not isinstance(initial_eval_processed, Tensor):
raise AssertionError(
"initial_eval_to_processed_initial_eval_fut: "
"initial_eval should be a Tensor"
)
result = self._process_initial_eval(
initial_eval_processed, formatted_inputs
)
except FeatureAblationFutureError as e:
raise FeatureAblationFutureError(
"initial_eval_to_processed_initial_eval_fut func failed"
) from e
return result
[docs]
@log_usage()
def attribute_future(
self,
inputs: TensorOrTupleOfTensorsGeneric,
baselines: BaselineType = None,
target: TargetType = None,
additional_forward_args: Optional[object] = None,
feature_mask: Union[None, Tensor, Tuple[Tensor, ...]] = None,
perturbations_per_eval: int = 1,
show_progress: bool = False,
**kwargs: Any,
) -> Future[TensorOrTupleOfTensorsGeneric]:
r"""
Almost the same as the attribute function, except that it requires a
forward function that returns a Future, and it returns a Future.
"""
# Keeps track whether original input is a tuple or not before
# converting it into a tuple.
is_inputs_tuple = _is_tuple(inputs)
formatted_inputs, baselines = _format_input_baseline(inputs, baselines)
formatted_additional_forward_args = _format_additional_forward_args(
additional_forward_args
)
num_examples = formatted_inputs[0].shape[0]
formatted_feature_mask = _format_feature_mask(feature_mask, formatted_inputs)
assert (
isinstance(perturbations_per_eval, int) and perturbations_per_eval >= 1
), "Perturbations per evaluation must be an integer and at least 1."
with torch.no_grad():
if show_progress:
attr_progress = self._attribute_progress_setup(
formatted_inputs,
formatted_feature_mask,
**kwargs,
perturbations_per_eval=perturbations_per_eval,
)
attr_progress.update(0)
# Computes initial evaluation with all features, which is compared
# to each ablated result.
initial_eval: Union[Tensor, Future[Tensor]] = _run_forward(
self.forward_func,
formatted_inputs,
target,
formatted_additional_forward_args,
)
if show_progress:
attr_progress.update()
processed_initial_eval_fut: Optional[
Future[Tuple[List[Tensor], List[Tensor], Tensor, Tensor, int, dtype]]
] = None
if not isinstance(initial_eval, torch.Future):
raise AssertionError(
"when using attribute_future, initial_eval should have "
f"Future type rather than {type(initial_eval)}"
)
processed_initial_eval_fut = initial_eval.then(
lambda initial_eval: self._initial_eval_to_processed_initial_eval_fut(
initial_eval,
formatted_inputs,
)
)
# The will be the same amount futures as modified_eval down there,
# since we cannot add up the evaluation result adhoc under async mode.
all_modified_eval_futures: List[
List[Future[Tuple[List[Tensor], List[Tensor]]]]
] = [[] for _ in range(len(inputs))]
# Iterate through each feature tensor for ablation
for i in range(len(formatted_inputs)):
# Skip any empty input tensors
if torch.numel(formatted_inputs[i]) == 0:
continue
for (
current_inputs,
current_add_args,
current_target,
current_mask,
) in self._ith_input_ablation_generator(
i,
formatted_inputs,
formatted_additional_forward_args,
target,
baselines,
formatted_feature_mask,
perturbations_per_eval,
**kwargs,
):
# modified_eval has (n_feature_perturbed * n_outputs) elements
# shape:
# agg mode: (*initial_eval.shape)
# non-agg mode:
# (feature_perturbed * batch_size, *initial_eval.shape[1:])
modified_eval: Union[Tensor, Future[Tensor]] = _run_forward(
self.forward_func,
current_inputs,
current_target,
current_add_args,
)
if show_progress:
attr_progress.update()
if not isinstance(modified_eval, torch.Future):
raise AssertionError(
"when using attribute_future, modified_eval should have "
f"Future type rather than {type(modified_eval)}"
)
if processed_initial_eval_fut is None:
raise AssertionError(
"processed_initial_eval_fut should not be None"
)
# Need to collect both initial eval and modified_eval
eval_futs: Future[
List[
Future[
Union[
Tuple[
List[Tensor],
List[Tensor],
Tensor,
Tensor,
int,
dtype,
],
Tensor,
]
]
]
] = collect_all(
[
processed_initial_eval_fut,
modified_eval,
]
)
ablated_out_fut: Future[Tuple[List[Tensor], List[Tensor]]] = (
eval_futs.then(
lambda eval_futs, current_inputs=current_inputs, current_mask=current_mask, i=i: self._eval_fut_to_ablated_out_fut( # type: ignore # noqa: E501 line too long
eval_futs=eval_futs,
current_inputs=current_inputs,
current_mask=current_mask,
i=i,
perturbations_per_eval=perturbations_per_eval,
num_examples=num_examples,
formatted_inputs=formatted_inputs,
)
)
)
all_modified_eval_futures[i].append(ablated_out_fut)
if show_progress:
attr_progress.close()
return self._generate_async_result(all_modified_eval_futures, is_inputs_tuple) # type: ignore # noqa: E501 line too long
# pyre-fixme[3] return type must be annotated
def _attribute_progress_setup(
self,
formatted_inputs: Tuple[Tensor, ...],
feature_mask: Tuple[Tensor, ...],
perturbations_per_eval: int,
**kwargs: Any,
):
feature_counts = self._get_feature_counts(
formatted_inputs, feature_mask, **kwargs
)
total_forwards = (
sum(math.ceil(count / perturbations_per_eval) for count in feature_counts)
+ 1
) # add 1 for the initial eval
attr_progress = progress(
desc=f"{self.get_name()} attribution", total=total_forwards
)
return attr_progress
def _eval_fut_to_ablated_out_fut(
self,
# pyre-ignore Invalid type parameters [24]
eval_futs: Future[List[Future[List[object]]]],
current_inputs: Tuple[Tensor, ...],
current_mask: Tensor,
i: int,
perturbations_per_eval: int,
num_examples: int,
formatted_inputs: Tuple[Tensor, ...],
) -> Tuple[List[Tensor], List[Tensor]]:
try:
modified_eval = cast(Tensor, eval_futs.value()[1].value())
initial_eval_tuple = cast(
Tuple[
List[Tensor],
List[Tensor],
Tensor,
Tensor,
int,
dtype,
],
eval_futs.value()[0].value(),
)
if len(initial_eval_tuple) != 6:
raise AssertionError(
"eval_fut_to_ablated_out_fut: "
"initial_eval_tuple should have 6 elements: "
"total_attrib, weights, initial_eval, "
"flattened_initial_eval, n_outputs, attrib_type "
)
if not isinstance(modified_eval, Tensor):
raise AssertionError(
"eval_fut_to_ablated_out_fut: " "modified eval should be a Tensor"
)
(
total_attrib,
weights,
initial_eval,
flattened_initial_eval,
n_outputs,
attrib_type,
) = initial_eval_tuple
result = self._process_ablated_out( # type: ignore # noqa: E501 line too long
modified_eval=modified_eval,
current_inputs=current_inputs,
current_mask=current_mask,
perturbations_per_eval=perturbations_per_eval,
num_examples=num_examples,
initial_eval=initial_eval,
flattened_initial_eval=flattened_initial_eval,
inputs=formatted_inputs,
n_outputs=n_outputs,
total_attrib=total_attrib,
weights=weights,
i=i,
attrib_type=attrib_type,
)
except FeatureAblationFutureError as e:
raise FeatureAblationFutureError(
"eval_fut_to_ablated_out_fut func failed)"
) from e
return result
def _ith_input_ablation_generator(
self,
i: int,
inputs: TensorOrTupleOfTensorsGeneric,
additional_args: Optional[Tuple[object, ...]],
target: TargetType,
baselines: BaselineType,
input_mask: Union[None, Tensor, Tuple[Tensor, ...]],
perturbations_per_eval: int,
**kwargs: Any,
) -> Generator[
Tuple[
Tuple[Tensor, ...],
object,
TargetType,
Tensor,
],
None,
None,
]:
"""
This method returns a generator of ablation perturbations of the i-th input
Returns:
ablation_iter (Generator): yields each perturbation to be evaluated
as a tuple (inputs, additional_forward_args, targets, mask).
"""
extra_args = {}
for key, value in kwargs.items():
# For any tuple argument in kwargs, we choose index i of the tuple.
if isinstance(value, tuple):
extra_args[key] = value[i]
else:
extra_args[key] = value
cur_input_mask = input_mask[i] if input_mask is not None else None
min_feature, num_features, cur_input_mask = self._get_feature_range_and_mask(
inputs[i], cur_input_mask, **extra_args
)
num_examples = inputs[0].shape[0]
perturbations_per_eval = min(perturbations_per_eval, num_features)
baseline = baselines[i] if isinstance(baselines, tuple) else baselines
if isinstance(baseline, torch.Tensor):
baseline = baseline.reshape((1,) + tuple(baseline.shape))
additional_args_repeated: object
if perturbations_per_eval > 1:
# Repeat features and additional args for batch size.
all_features_repeated = [
torch.cat([inputs[j]] * perturbations_per_eval, dim=0)
for j in range(len(inputs))
]
additional_args_repeated = (
_expand_additional_forward_args(additional_args, perturbations_per_eval)
if additional_args is not None
else None
)
target_repeated = _expand_target(target, perturbations_per_eval)
else:
all_features_repeated = list(inputs)
additional_args_repeated = additional_args
target_repeated = target
num_features_processed = min_feature
current_additional_args: object
while num_features_processed < num_features:
current_num_ablated_features = min(
perturbations_per_eval, num_features - num_features_processed
)
# Store appropriate inputs and additional args based on batch size.
if current_num_ablated_features != perturbations_per_eval:
current_features = [
feature_repeated[0 : current_num_ablated_features * num_examples]
for feature_repeated in all_features_repeated
]
current_additional_args = (
_expand_additional_forward_args(
additional_args, current_num_ablated_features
)
if additional_args is not None
else None
)
current_target = _expand_target(target, current_num_ablated_features)
else:
current_features = all_features_repeated
current_additional_args = additional_args_repeated
current_target = target_repeated
# Store existing tensor before modifying
original_tensor = current_features[i]
# Construct ablated batch for features in range num_features_processed
# to num_features_processed + current_num_ablated_features and return
# mask with same size as ablated batch. ablated_features has dimension
# (current_num_ablated_features, num_examples, inputs[i].shape[1:])
# Note that in the case of sparse tensors, the second dimension
# may not necessarilly be num_examples and will match the first
# dimension of this tensor.
current_reshaped = current_features[i].reshape(
(current_num_ablated_features, -1)
+ tuple(current_features[i].shape[1:])
)
ablated_features, current_mask = self._construct_ablated_input(
current_reshaped,
cur_input_mask,
baseline,
num_features_processed,
num_features_processed + current_num_ablated_features,
**extra_args,
)
# current_features[i] has dimension
# (current_num_ablated_features * num_examples, inputs[i].shape[1:]),
# which can be provided to the model as input.
current_features[i] = ablated_features.reshape(
(-1,) + tuple(ablated_features.shape[2:])
)
yield tuple(
current_features
), current_additional_args, current_target, current_mask
# Replace existing tensor at index i.
current_features[i] = original_tensor
num_features_processed += current_num_ablated_features
def _construct_ablated_input(
self,
expanded_input: Tensor,
input_mask: Union[None, Tensor, Tuple[Tensor, ...]],
baseline: Union[None, float, Tensor],
start_feature: int,
end_feature: int,
**kwargs: Any,
) -> Tuple[Tensor, Tensor]:
r"""
Ablates given expanded_input tensor with given feature mask, feature range,
and baselines. expanded_input shape is (`num_features`, `num_examples`, ...)
with remaining dimensions corresponding to remaining original tensor
dimensions and `num_features` = `end_feature` - `start_feature`.
input_mask has same number of dimensions as original input tensor (one less
than `expanded_input`), and can have first dimension either 1, applying same
feature mask to all examples, or `num_examples`. baseline is expected to
be broadcastable to match `expanded_input`.
This method returns the ablated input tensor, which has the same
dimensionality as `expanded_input` as well as the corresponding mask with
either the same dimensionality as `expanded_input` or second dimension
being 1. This mask contains 1s in locations which have been ablated (and
thus counted towards ablations for that feature) and 0s otherwise.
"""
current_mask = torch.stack(
cast(List[Tensor], [input_mask == j for j in range(start_feature, end_feature)]), # type: ignore # noqa: E501 line too long
dim=0,
).long()
current_mask = current_mask.to(expanded_input.device)
assert baseline is not None, "baseline must be provided"
ablated_tensor = (
expanded_input
* (1 - current_mask).to(expanded_input.dtype)
# pyre-fixme[58]: `*` is not supported for operand types `Union[None, float,
# Tensor]` and `Tensor`.
) + (baseline * current_mask.to(expanded_input.dtype))
return ablated_tensor, current_mask
def _get_feature_range_and_mask(
self,
input: Tensor,
input_mask: Optional[Tensor],
**kwargs: Any,
) -> Tuple[int, int, Union[None, Tensor, Tuple[Tensor, ...]]]:
if input_mask is None:
# Obtain feature mask for selected input tensor, matches size of
# 1 input example, (1 x inputs[i].shape[1:])
input_mask = torch.reshape(
torch.arange(torch.numel(input[0]), device=input.device),
input[0:1].shape,
).long()
return (
int(torch.min(input_mask).item()),
int(torch.max(input_mask).item() + 1),
input_mask,
)
def _get_feature_counts(
self,
inputs: TensorOrTupleOfTensorsGeneric,
feature_mask: Tuple[Tensor, ...],
**kwargs: Any,
) -> Tuple[float, ...]:
"""return the numbers of input features"""
if not feature_mask:
return tuple(inp[0].numel() if inp.numel() else 0 for inp in inputs)
return tuple(
(
(mask.max() - mask.min()).item() + 1
if mask is not None
else (inp[0].numel() if inp.numel() else 0)
)
for inp, mask in zip(inputs, feature_mask)
)
def _parse_forward_out(self, forward_output: Tensor) -> Tensor:
"""
A temp wrapper for global _run_forward util to force forward output
type assertion & conversion.
Remove after the strict logic is supported by all attr classes
"""
if isinstance(forward_output, Tensor):
return forward_output
output_type = type(forward_output)
assert output_type is int or output_type is float, (
"the return of forward_func must be a tensor, int, or float,"
f" received: {forward_output}"
)
# using python built-in type as torch dtype
# int -> torch.int64, float -> torch.float64
# ref: https://github.com/pytorch/pytorch/pull/21215
return torch.tensor(forward_output, dtype=cast(dtype, output_type))
def _process_initial_eval(
self,
initial_eval: Tensor,
inputs: TensorOrTupleOfTensorsGeneric,
) -> Tuple[List[Tensor], List[Tensor], Tensor, Tensor, int, dtype]:
initial_eval = self._parse_forward_out(initial_eval)
# number of elements in the output of forward_func
n_outputs = initial_eval.numel() if isinstance(initial_eval, Tensor) else 1
# flatten eval outputs into 1D (n_outputs)
# add the leading dim for n_feature_perturbed
flattened_initial_eval = initial_eval.reshape(1, -1)
# Initialize attribution totals and counts
attrib_type = flattened_initial_eval.dtype
total_attrib = [
# attribute w.r.t each output element
torch.zeros(
(n_outputs,) + input.shape[1:],
dtype=attrib_type,
device=input.device,
)
for input in inputs
]
# Weights are used in cases where ablations may be overlapping.
weights = []
if self.use_weights:
weights = [
torch.zeros((n_outputs,) + input.shape[1:], device=input.device).float()
for input in inputs
]
return (
total_attrib,
weights,
initial_eval,
flattened_initial_eval,
n_outputs,
attrib_type,
)
def _process_ablated_out(
self,
modified_eval: Tensor,
current_inputs: Tuple[Tensor, ...],
current_mask: Tensor,
perturbations_per_eval: int,
num_examples: int,
initial_eval: Tensor,
flattened_initial_eval: Tensor,
inputs: TensorOrTupleOfTensorsGeneric,
n_outputs: int,
total_attrib: List[Tensor],
weights: List[Tensor],
i: int,
attrib_type: dtype,
) -> Tuple[List[Tensor], List[Tensor]]:
modified_eval = self._parse_forward_out(modified_eval)
# if perturbations_per_eval > 1, the output shape must grow with
# input and not be aggregated
if perturbations_per_eval > 1 and not self._is_output_shape_valid:
current_batch_size = current_inputs[0].shape[0]
# number of perturbation, which is not the same as
# perturbations_per_eval when not enough features to perturb
n_perturb = current_batch_size / num_examples
current_output_shape = modified_eval.shape
# use initial_eval as the forward of perturbations_per_eval = 1
initial_output_shape = initial_eval.shape
assert (
# check if the output is not a scalar
current_output_shape
and initial_output_shape
# check if the output grow in same ratio, i.e., not agg
and current_output_shape[0] == n_perturb * initial_output_shape[0]
), (
"When perturbations_per_eval > 1, forward_func's output "
"should be a tensor whose 1st dim grow with the input "
f"batch size: when input batch size is {num_examples}, "
f"the output shape is {initial_output_shape}; "
f"when input batch size is {current_batch_size}, "
f"the output shape is {current_output_shape}"
)
self._is_output_shape_valid = True
# reshape the leading dim for n_feature_perturbed
# flatten each feature's eval outputs into 1D of (n_outputs)
modified_eval = modified_eval.reshape(-1, n_outputs)
# eval_diff in shape (n_feature_perturbed, n_outputs)
eval_diff = flattened_initial_eval - modified_eval
# append the shape of one input example
# to make it broadcastable to mask
eval_diff = eval_diff.reshape(eval_diff.shape + (inputs[i].dim() - 1) * (1,))
eval_diff = eval_diff.to(total_attrib[i].device)
if self.use_weights:
weights[i] += current_mask.float().sum(dim=0)
total_attrib[i] += (eval_diff * current_mask.to(attrib_type)).sum(dim=0)
return total_attrib, weights
def _fut_tuple_to_accumulate_fut_list(
self,
total_attrib: List[Tensor],
weights: List[Tensor],
i: int,
fut_tuple: Future[Tuple[List[Tensor], List[Tensor]]],
) -> None:
try:
attrib, weight = fut_tuple.value()
self._accumulate_for_single_input(total_attrib, weights, i, attrib, weight)
except FeatureAblationFutureError as e:
raise FeatureAblationFutureError(
"fut_tuple_to_accumulate_fut_list failed"
) from e
def _generate_async_result(
self,
futs: List[List[Future[Tuple[List[Tensor], List[Tensor]]]]],
is_inputs_tuple: bool,
) -> Future[Union[Tensor, Tuple[Tensor, ...]]]:
# Each element of the 2d list contains evalutaion results for a feature
# Need to add up all the results for each input
accumulate_fut_list: List[Future[None]] = []
total_attrib: List[Tensor] = []
weights: List[Tensor] = []
for i, fut_tuples in enumerate(futs):
for fut_tuple in fut_tuples:
accumulate_fut_list.append(
fut_tuple.then(
lambda fut_tuple, i=i: self._fut_tuple_to_accumulate_fut_list( # type: ignore # noqa: E501 line too long
total_attrib, weights, i, fut_tuple
)
)
)
result_fut = collect_all(accumulate_fut_list).then(
lambda x: self._generate_result(total_attrib, weights, is_inputs_tuple)
)
return result_fut
def _accumulate_for_single_input(
self,
total_attrib: List[Tensor],
weights: List[Tensor],
idx: int,
attrib: List[Tensor],
weight: List[Tensor],
) -> None:
if total_attrib:
total_attrib[idx] = attrib[idx]
else:
total_attrib.extend(attrib)
if self.use_weights:
if weights:
weights[idx] = weight[idx]
else:
weights.extend(weight)
def _generate_result(
self,
total_attrib: List[Tensor],
weights: List[Tensor],
is_inputs_tuple: bool,
) -> Union[Tensor, Tuple[Tensor, ...]]:
# Divide total attributions by counts and return formatted attributions
if self.use_weights:
attrib = tuple(
single_attrib.float() / weight
for single_attrib, weight in zip(total_attrib, weights)
)
else:
attrib = tuple(total_attrib)
return _format_output(is_inputs_tuple, attrib)
