Fix grid

Fixes an error that no CUDA devices are available.

In fact, CUDA might be available on the scheduler machine, and because this part of the code was run by the scheduler, torch.cuda.is_available() returned True, which caused the job to crash. This problem is solved by applying the modification in this merge request.

bob/pad/face/deep_pix_bis.py

@@ -178,9 +178,7 @@ class DeepPixBisClassifier(BaseEstimator, ClassifierMixin):
                     self.scoring_method
                 )
             )
-        self.device = torch.device(
-            device or "cuda" if torch.cuda.is_available() else "cpu"
-        )
+        self.device = device
         self.threshold = threshold
 
         logger.debug(
@@ -275,5 +273,9 @@ class DeepPixBisClassifier(BaseEstimator, ClassifierMixin):
         return {"requires_fit": False}
 
     def place_model_on_device(self):
+        if self.device is None:
+            self.device = torch.device(
+                "cuda" if torch.cuda.is_available() else "cpu"
+            )
         if self.model is not None:
             self.model.to(self.device)