Log information of infinity in gradients and update 0.0

Signed-off-by: Beat Buesser <[email protected]>

Author: Beat Buesser

art/attacks/evasion/fast_gradient.py

@@ -344,14 +344,17 @@ def _compute_perturbation(
         # Check for NaN before normalisation an replace with 0
         if np.isnan(grad).any():
             logger.warning("Elements of the loss gradient are NaN and have been replaced with 0.0.")
-            grad = np.where(np.isnan(grad), np.zeros_like(grad), grad)
+            grad = np.where(np.isnan(grad), 0.0, grad)
 
         # Apply mask
         if mask is not None:
             grad = np.where(mask == 0.0, 0.0, grad)
 
         # Apply norm bound
         def _apply_norm(grad, object_type=False):
+            if np.isinf(grad).any():
+                logger.info("The loss gradient array contains at least one positive or negative infinity.")
+
             if self.norm in [np.inf, "inf"]:
                 grad = np.sign(grad)
             elif self.norm == 1: