Merge updates from master branch

Author: ProGamerGov

captum/attr/_core/lime.py

@@ -632,7 +632,7 @@ def default_exp_kernel(original_inp, perturbed_inp, __, **kwargs):
             distance = torch.norm(flattened_original_inp - flattened_perturbed_inp)
         else:
             raise ValueError("distance_mode must be either cosine or euclidean.")
-        return math.exp(-1 * (distance ** 2) / (2 * (kernel_width ** 2)))
+        return math.exp(-1 * (distance**2) / (2 * (kernel_width**2)))
 
     return default_exp_kernel
 

captum/attr/_core/noise_tunnel.py

@@ -229,7 +229,7 @@ def update_sum_attribution_and_sq(
 
             attribution = attribution.view(attribution_shape)
             current_attribution_sum = attribution.sum(dim=1, keepdim=False)
-            current_attribution_sq = torch.sum(attribution ** 2, dim=1, keepdim=False)
+            current_attribution_sq = torch.sum(attribution**2, dim=1, keepdim=False)
 
             sum_attribution[i] = (
                 current_attribution_sum

captum/attr/_utils/stat.py

@@ -223,7 +223,7 @@ def update(self, x: Tensor):
 
     def get(self) -> Optional[Tensor]:
         var = self.var.get()
-        return var ** 0.5 if var is not None else None
+        return var**0.5 if var is not None else None
 
 
 class GeneralAccumFn(Stat):

captum/attr/_utils/visualization.py

@@ -12,7 +12,7 @@
 from numpy import ndarray
 
 try:
-    from IPython.core.display import display, HTML
+    from IPython.display import display, HTML
 
     HAS_IPYTHON = True
 except ImportError:

captum/influence/_core/tracincp_fast_rand_proj.py

@@ -494,8 +494,8 @@ def get_checkpoint_contribution(checkpoint):
             )
 
             return (
-                torch.sum(batch_jacobian ** 2, dim=1)
-                * torch.sum(batch_layer_input ** 2, dim=1)
+                torch.sum(batch_jacobian**2, dim=1)
+                * torch.sum(batch_layer_input**2, dim=1)
                 * learning_rate
             )
 
@@ -1063,17 +1063,17 @@ def _set_projections_tracincp_fast_rand_proj(
             # `projection_dim` corresponds to the variable d in the top of page 15 of
             # the TracIn paper: https://arxiv.org/pdf/2002.08484.pdf.
             if jacobian_dim * layer_input_dim > projection_dim:
-                jacobian_projection_dim = min(int(projection_dim ** 0.5), jacobian_dim)
+                jacobian_projection_dim = min(int(projection_dim**0.5), jacobian_dim)
                 layer_input_projection_dim = min(
-                    int(projection_dim ** 0.5), layer_input_dim
+                    int(projection_dim**0.5), layer_input_dim
                 )
                 jacobian_projection = torch.normal(
                     torch.zeros(jacobian_dim, jacobian_projection_dim),
-                    1.0 / jacobian_projection_dim ** 0.5,
+                    1.0 / jacobian_projection_dim**0.5,
                 )
                 layer_input_projection = torch.normal(
                     torch.zeros(layer_input_dim, layer_input_projection_dim),
-                    1.0 / layer_input_projection_dim ** 0.5,
+                    1.0 / layer_input_projection_dim**0.5,
                 )
 
                 projection_quantities = jacobian_projection, layer_input_projection
@@ -1157,7 +1157,7 @@ def _get_intermediate_quantities_tracincp_fast_rand_proj(
             ), "None returned from `checkpoints`, cannot load."
 
             learning_rate = self.checkpoints_load_func(self.model, checkpoint)
-            learning_rate_root = learning_rate ** 0.5
+            learning_rate_root = learning_rate**0.5
 
             for batch in dataloader:
 

captum/metrics/_core/infidelity.py

@@ -570,7 +570,7 @@ def _sum_infidelity_tensors(agg_tensors, tensors):
         beta = safe_div(beta_num, beta_denorm)
 
         infidelity_values = (
-            beta ** 2 * agg_tensors[0] - 2 * beta * agg_tensors[1] + agg_tensors[2]
+            beta**2 * agg_tensors[0] - 2 * beta * agg_tensors[1] + agg_tensors[2]
         )
     else:
         infidelity_values = agg_tensors[0]

captum/optim/_core/loss.py

@@ -267,7 +267,7 @@ class DeepDream(BaseLoss):
     def __call__(self, targets_to_values: ModuleOutputMapping) -> torch.Tensor:
         activations = targets_to_values[self.target]
         activations = activations[self.batch_index[0] : self.batch_index[1]]
-        return activations ** 2
+        return activations**2
 
 
 @loss_wrapper
@@ -587,7 +587,7 @@ def __call__(self, targets_to_values: ModuleOutputMapping) -> torch.Tensor:
             return activations * vec
 
         dot = torch.mean(activations * vec)
-        cossims = dot / (self.eps + torch.sqrt(torch.sum(activations ** 2)))
+        cossims = dot / (self.eps + torch.sqrt(torch.sum(activations**2)))
         return dot * torch.clamp(cossims, min=0.1) ** self.cossim_pow
 
 

captum/optim/_param/image/images.py

@@ -385,7 +385,7 @@ def _setup_input(
                 x = F.interpolate(init.clone(), size=(h, w), mode="bilinear")
                 x = x / 6  # Prevents output from being all white
             upsample = torch.nn.Upsample(scale_factor=scale, mode="nearest")
-            x = x * (scale ** power) / (32 ** power)
+            x = x * (scale**power) / (32**power)
             x = torch.nn.Parameter(x)
             tensor_params.append(x)
             scaler.append(upsample)

captum/optim/_param/image/transforms.py

@@ -1217,9 +1217,9 @@ def __init__(
         padding_transform: Optional[nn.Module] = nn.ConstantPad2d(2, value=0.5),
         translate: Optional[Union[int, List[int]]] = [4] * 10,
         scale: Optional[NumSeqOrTensorOrProbDistType] = [
-            0.995 ** n for n in range(-5, 80)
+            0.995**n for n in range(-5, 80)
         ]
-        + [0.998 ** n for n in 2 * list(range(20, 40))],
+        + [0.998**n for n in 2 * list(range(20, 40))],
         degrees: Optional[NumSeqOrTensorOrProbDistType] = list(range(-20, 20))
         + list(range(-10, 10))
         + list(range(-5, 5))

captum/robust/_core/fgsm.py

@@ -64,7 +64,7 @@ def __init__(
         self.forward_func = forward_func
         self.loss_func = loss_func
         self.bound = lambda x: torch.clamp(x, min=lower_bound, max=upper_bound)
-        self.zero_thresh = 10 ** -6
+        self.zero_thresh = 10**-6
 
     def perturb(
         self,