Update version to 0.8.0

README.md

@@ -25,8 +25,6 @@ of integrated gradients, saliency maps, smoothgrad, vargrad and others for
 PyTorch models. It has quick integration for models built with domain-specific
 libraries such as torchvision, torchtext, and others.
 
-*Captum is currently in beta and under active development!*
-
 
 #### About Captum
 
@@ -92,6 +90,7 @@ pip install -e .
 To customize the installation, you can also run the following variants of the
 above:
 * `pip install -e .[insights]`: Also installs all packages necessary for running Captum Insights.
+**NOTE**: Captum Insights is being deprecated. See further details [below](#captum-insights).
 * `pip install -e .[dev]`: Also installs all tools necessary for development
   (testing, linting, docs building; see [Contributing](#contributing) below).
 * `pip install -e .[tutorials]`: Also installs all packages necessary for running the tutorial notebooks.
@@ -388,6 +387,10 @@ Captum on different types of models can be found in our tutorials.
 
 ## Captum Insights
 
+**NOTE**: *Support for Captum Insights is being deprecated in an upcoming release.
+While the code will still be available, there will no longer be active
+development or support for it.*
+
 Captum provides a web interface called Insights for easy visualization and
 access to a number of our interpretability algorithms.
 

captum/__init__.py

@@ -9,6 +9,6 @@
 import captum.robust as robust
 
 
-__version__ = "0.7.0"
+__version__ = "0.8.0"
 
 __all__ = ["attr", "concept", "influence", "log", "metrics", "robust"]

captum/_utils/common.py

@@ -232,12 +232,12 @@ def _format_tensor_into_tuples(inputs: None) -> None: ...
 
 @overload
 def _format_tensor_into_tuples(
-    inputs: Union[Tensor, Tuple[Tensor, ...]]
+    inputs: Union[Tensor, Tuple[Tensor, ...]],
 ) -> Tuple[Tensor, ...]: ...
 
 
 def _format_tensor_into_tuples(
-    inputs: Union[None, Tensor, Tuple[Tensor, ...]]
+    inputs: Union[None, Tensor, Tuple[Tensor, ...]],
 ) -> Union[None, Tuple[Tensor, ...]]:
     if inputs is None:
         return None
@@ -261,7 +261,7 @@ def _format_inputs(inputs: Any, unpack_inputs: bool = True) -> Any:
 
 
 def _format_float_or_tensor_into_tuples(
-    inputs: Union[float, Tensor, Tuple[Union[float, Tensor], ...]]
+    inputs: Union[float, Tensor, Tuple[Union[float, Tensor], ...]],
 ) -> Tuple[Union[float, Tensor], ...]:
     if not isinstance(inputs, tuple):
         assert isinstance(
@@ -276,7 +276,7 @@ def _format_float_or_tensor_into_tuples(
 @overload
 def _format_additional_forward_args(
     # pyre-fixme[24]: Generic type `tuple` expects at least 1 type parameter.
-    additional_forward_args: Union[Tensor, Tuple]
+    additional_forward_args: Union[Tensor, Tuple],
     # pyre-fixme[24]: Generic type `tuple` expects at least 1 type parameter.
 ) -> Tuple: ...
 

captum/attr/_core/dataloader_attr.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python3
 
 # pyre-strict
+
 from collections import defaultdict
 from copy import copy
 from typing import Callable, cast, Dict, Iterable, List, Optional, Tuple, Union
@@ -30,7 +31,6 @@ class InputRole:
 
 
 # default reducer wehn reduce is None. Simply concat the outputs by the batch dimension
-# pyre-fixme[2]: Parameter must be annotated.
 def _concat_tensors(accum: Optional[Tensor], cur_output: Tensor, _) -> Tensor:
     return cur_output if accum is None else torch.cat([accum, cur_output])
 
@@ -87,9 +87,7 @@ def _perturb_inputs(
         else:
             baseline = baselines[attr_inp_count]
 
-            # pyre-fixme[58]: `*` is not supported for operand types `object` and
-            #  `Tensor`.
-            perturbed_inp = inp * pert_mask + baseline * (1 - pert_mask)
+            perturbed_inp = cast(Tensor, inp) * pert_mask + baseline * (1 - pert_mask)
             perturbed_inputs.append(perturbed_inp)
 
         attr_inp_count += 1

captum/attr/_core/layer/layer_lrp.py

@@ -307,7 +307,7 @@ def _get_output_relevance(
 
     @staticmethod
     def _convert_list_to_tuple(
-        relevances: Union[List[T], Tuple[T, ...]]
+        relevances: Union[List[T], Tuple[T, ...]],
     ) -> Tuple[T, ...]:
         if isinstance(relevances, list):
             return tuple(relevances)

captum/attr/_core/llm_attr.py

@@ -559,12 +559,19 @@ def _forward_func(
                         outputs.past_key_values = DynamicCache.from_legacy_cache(
                             outputs.past_key_values
                         )
+                    # nn.Module typing suggests non-base attributes are modules or
+                    # tensors
+                    _update_model_kwargs_for_generation = (
+                        self.model._update_model_kwargs_for_generation
+                    )
                     # pyre-fixme[29]: `Union[Module, Tensor]` is not a function.
-                    model_kwargs = self.model._update_model_kwargs_for_generation(
+                    model_kwargs = _update_model_kwargs_for_generation(  # type: ignore
                         outputs, model_kwargs
                     )
+                # nn.Module typing suggests non-base attributes are modules or tensors
+                prep_inputs_for_generation = self.model.prepare_inputs_for_generation
                 # pyre-fixme[29]: `Union[Module, Tensor]` is not a function.
-                model_inputs = self.model.prepare_inputs_for_generation(
+                model_inputs = prep_inputs_for_generation(  # type: ignore
                     model_inp, **model_kwargs
                 )
                 outputs = self.model.forward(**model_inputs)

captum/attr/_utils/common.py

@@ -373,7 +373,7 @@ def _find_output_mode_and_verify(
 
 
 def _construct_default_feature_mask(
-    inputs: Tuple[Tensor, ...]
+    inputs: Tuple[Tensor, ...],
 ) -> Tuple[Tuple[Tensor, ...], int]:
     feature_mask = []
     current_num_features = 0

captum/attr/_utils/stat.py

@@ -1,7 +1,8 @@
 #!/usr/bin/env python3
 
 # pyre-strict
-from typing import Any, Callable, List, Optional, TYPE_CHECKING
+
+from typing import Any, Callable, cast, List, Optional, TYPE_CHECKING
 
 import torch
 from torch import Tensor
@@ -117,20 +118,18 @@ def get(self) -> Optional[Tensor]:
         return self.rolling_mean
 
     def init(self) -> None:
-        # pyre-fixme[8]: Attribute has type `Optional[Count]`; used as `Optional[Stat]`.
-        self.n = self._get_stat(Count())  # type: ignore
+        self.n = cast(Count, self._get_stat(Count()))
 
     def update(self, x: Tensor) -> None:
-        # pyre-fixme[16]: `Optional` has no attribute `get`.
-        n = self.n.get()  # type: ignore
+        n = cast(Count, self.n).get()
 
         if self.rolling_mean is None:
             # Ensures rolling_mean is a float tensor
             self.rolling_mean = x.clone() if x.is_floating_point() else x.double()
         else:
             delta = x - self.rolling_mean
-            # pyre-fixme[16]: `Optional` has no attribute `__iadd__`.
-            self.rolling_mean += delta / n
+            # pyre-ignore[16]: `Optional` has no attribute `__iadd__` (false positive)
+            self.rolling_mean += delta / cast(int, n)
 
 
 class MSE(Stat):

captum/influence/_utils/common.py

@@ -338,7 +338,7 @@ def __len__(self) -> int:
 
 def _format_inputs_dataset(
     # pyre-fixme[2]: Parameter annotation cannot contain `Any`.
-    inputs_dataset: Union[Tuple[Any, ...], DataLoader]
+    inputs_dataset: Union[Tuple[Any, ...], DataLoader],
 ) -> DataLoader:
     # if `inputs_dataset` is not a `DataLoader`, turn it into one.
     # `_DatasetFromList` turns a list into a `Dataset` where `__getitem__`
@@ -604,7 +604,7 @@ def _flatten_params(_params: Tuple[Tensor, ...]) -> Tensor:
 
 # pyre-fixme[3]: Return type must be annotated.
 def _unflatten_params_factory(
-    param_shapes: Union[List[Tuple[int, ...]], Tuple[Tensor, ...]]
+    param_shapes: Union[List[Tuple[int, ...]], Tuple[Tensor, ...]],
 ):
     """
     returns a function which is the inverse of `_flatten_params`

captum/insights/attr_vis/_utils/transforms.py

@@ -7,7 +7,7 @@
 
 def format_transforms(
     # pyre-fixme[24]: Generic type `Callable` expects 2 type parameters.
-    transforms: Optional[Union[Callable, List[Callable]]]
+    transforms: Optional[Union[Callable, List[Callable]]],
     # pyre-fixme[24]: Generic type `Callable` expects 2 type parameters.
 ) -> List[Callable]:
     if transforms is None:

captum/insights/attr_vis/frontend/package.json

@@ -1,6 +1,6 @@
 {
   "name": "frontend",
-  "version": "0.7.0",
+  "version": "0.8.0",
   "private": true,
   "homepage": ".",
   "dependencies": {

captum/metrics/_core/infidelity.py

@@ -59,7 +59,7 @@ def infidelity_perturb_func_decorator(
     """
 
     def sub_infidelity_perturb_func_decorator(
-        perturb_func: Callable[..., TensorOrTupleOfTensorsGeneric]
+        perturb_func: Callable[..., TensorOrTupleOfTensorsGeneric],
     ) -> Callable[
         [TensorOrTupleOfTensorsGeneric, BaselineType],
         Tuple[Tuple[Tensor, ...], Tuple[Tensor, ...]],
@@ -611,6 +611,11 @@ def _next_infidelity_tensors(
             targets_expanded,
             additional_forward_args_expanded,
         )
+        if isinstance(inputs_perturbed_fwd, torch.futures.Future):
+            raise NotImplementedError(
+                f"Outputs from forward_func of type {type(inputs_perturbed_fwd)} are "
+                "not yet supported."
+            )
         inputs_fwd = _run_forward(forward_func, inputs, target, additional_forward_args)
         # _run_forward may return future of Tensor,
         # but we don't support it here now
@@ -619,8 +624,6 @@ def _next_infidelity_tensors(
         inputs_fwd = torch.repeat_interleave(
             inputs_fwd, current_n_perturb_samples, dim=0
         )
-        # pyre-fixme[58]: `-` is not supported for operand types `Tensor` and
-        #  `Union[Future[Tensor], Tensor]`.
         perturbed_fwd_diffs = inputs_fwd - inputs_perturbed_fwd
         attributions_expanded = tuple(
             torch.repeat_interleave(attribution, current_n_perturb_samples, dim=0)

captum/testing/helpers/influence/common.py

@@ -33,7 +33,7 @@ def _isSorted(x, key=lambda x: x, descending=True) -> bool:
 
 # pyre-fixme[2]: Parameter must be annotated.
 def _wrap_model_in_dataparallel(net) -> Module:
-    alt_device_ids = [0] + [x for x in range(torch.cuda.device_count() - 1, 0, -1)]
+    alt_device_ids = [0] + list(range(torch.cuda.device_count() - 1, 0, -1))
     net = net.cuda()
     return torch.nn.DataParallel(net, device_ids=alt_device_ids)
 
@@ -505,7 +505,7 @@ def get_random_model_and_data(
 
 # pyre-fixme[3]: Return type must be annotated.
 def generate_symmetric_matrix_given_eigenvalues(
-    eigenvalues: Union[Tensor, List[float]]
+    eigenvalues: Union[Tensor, List[float]],
 ):
     """
     following https://github.com/google-research/jax-influence/blob/74bd321156b5445bb35b9594568e4eaaec1a76a3/jax_influence/test_utils.py#L123  # noqa: E501
@@ -523,7 +523,7 @@ def generate_symmetric_matrix_given_eigenvalues(
 
 
 def generate_assymetric_matrix_given_eigenvalues(
-    eigenvalues: Union[Tensor, List[float]]
+    eigenvalues: Union[Tensor, List[float]],
 ) -> Tensor:
     """
     following https://github.com/google-research/jax-influence/blob/74bd321156b5445bb35b9594568e4eaaec1a76a3/jax_influence/test_utils.py#L105 # noqa: E501

pyproject.toml

@@ -2,4 +2,4 @@
 first_party_detection = false
 
 [tool.black]
-target-version = ['py36']
+target-version = ['py39']

setup.py

@@ -131,10 +131,14 @@ def get_package_files(root, subdirs):
             "conda": "https://anaconda.org/pytorch/captum",
         },
         keywords=[
+            "Model Interpretability",
+            "Model Understanding",
             "Model Interpretability",
             "Model Understanding",
             "Feature Importance",
             "Neuron Importance",
+            "Data Attribution",
+            "Explainable AI",
             "PyTorch",
         ],
         classifiers=[
@@ -148,7 +152,10 @@ def get_package_files(root, subdirs):
         ],
         long_description=long_description,
         long_description_content_type="text/markdown",
-        python_requires=">=3.9",
+        python_requires=">={required_major}.{required_minor}".format(
+            required_minor=REQUIRED_MINOR,
+            required_major=REQUIRED_MAJOR,
+        ),
         install_requires=[
             "matplotlib",
             "numpy<2.0",

tests/attr/layer/test_layer_integrated_gradients.py

@@ -137,7 +137,8 @@ def test_multiple_layers_multiple_inputs_shared_input(self) -> None:
             self,
             # last input for second layer is first input =>
             # add the attributions
-            (attribs_inputs[0] + attribs_inputs[1][-1],) + attribs_inputs[1][0:-1],
+            (attribs_inputs[0] + attribs_inputs[1][-1],)  # type: ignore
+            + attribs_inputs[1][0:-1],  # type: ignore
             attribs_inputs_regular_ig,
             delta=1e-5,
         )
@@ -183,7 +184,7 @@ def test_multiple_layers_multiple_input_outputs(self) -> None:
 
         assertTensorTuplesAlmostEqual(
             self,
-            (attribs_inputs[0],) + attribs_inputs[1],
+            (attribs_inputs[0],) + attribs_inputs[1],  # type: ignore
             attribs_inputs_regular_ig,
             delta=1e-7,
         )

tests/concept/test_tcav.py

@@ -792,7 +792,7 @@ def _compute_cavs_interpret(
         attribute_to_layer_input: bool = False,
     ) -> None:
         def wrap_in_list_if_not_already(
-            input: Union[str, float, List[float], List[str]]
+            input: Union[str, float, List[float], List[str]],
         ) -> Union[List[Union[float, str]], List[float], List[str]]:
             return (
                 input

tests/metrics/test_infidelity.py

@@ -41,7 +41,7 @@ def _local_perturb_func_default(
 # pyre-ignore[43]: The implementation of `_local_perturb_func` does not accept all
 #  possible arguments of overload defined on line `43`.
 def _local_perturb_func(
-    inputs: Tuple[Tensor, ...]
+    inputs: Tuple[Tensor, ...],
 ) -> Tuple[Tuple[Tensor, ...], Tuple[Tensor, ...]]: ...
 
 
@@ -83,7 +83,7 @@ def _global_perturb_func1_default(
 # pyre-fixme[43]: The implementation of `_global_perturb_func1` does not accept all
 #  possible arguments of overload defined on line `74`.
 def _global_perturb_func1(
-    inputs: Tuple[Tensor, ...]
+    inputs: Tuple[Tensor, ...],
 ) -> Tuple[Tuple[Tensor, ...], Tuple[Tensor, ...]]: ...