Fix deeplift mypy error (pytorch#1459)

Summary:
Pull Request resolved: pytorch#1459

Currently, Captum OSS tests are failing due to mypy failures (likely from new version) in DeepLift test cases. Adds fix for type failure caused by different signature between DeepLift and DeepLiftShap.

Reviewed By: cyrjano

Differential Revision: D67538043

fbshipit-source-id: cc0236fa819c666c08e31d78a51f89e0807d9791

tests/attr/test_deeplift_classification.py

@@ -2,7 +2,7 @@
 
 # pyre-unsafe
 
-from typing import Union
+from typing import TypeVar, Union
 
 import torch
 from captum._utils.typing import TargetType
@@ -21,6 +21,8 @@
 from torch import Tensor
 from torch.nn import Module
 
+DeepLiftAttrMethod = TypeVar("DeepLiftAttrMethod", DeepLift, DeepLiftShap)
+
 
 class Test(BaseTest):
     def test_sigmoid_classification(self) -> None:
@@ -155,7 +157,7 @@ def test_convnet_with_maxpool1d_large_baselines(self) -> None:
     def softmax_classification(
         self,
         model: Module,
-        attr_method: Union[DeepLift, DeepLiftShap],
+        attr_method: DeepLiftAttrMethod,
         input: Tensor,
         baselines: Union[float, int, Tensor],
         target: TargetType,