Parameterize score aggregation function

supirfactor_dynamical/models/_base_trainer.py

@@ -659,22 +659,38 @@ def score(
         self,
         dataloader,
         loss_function=torch.nn.MSELoss(),
+        reduction='sum',
         **kwargs
     ):
 
         if dataloader is None:
             return None
 
-        _score = 0
+        _score = []
 
         with torch.no_grad():
             for data in dataloader:
-                _score += loss_function(
-                    self._slice_data_and_forward(data),
-                    self.output_data(data),
-                    **kwargs
+                _score.append(
+                    loss_function(
+                        self._slice_data_and_forward(data),
+                        self.output_data(data),
+                        **kwargs
+                    )
                 )
 
+        _score = torch.Tensor(_score)
+
+        if reduction == 'mean':
+            _score = torch.mean(_score)
+        elif reduction == 'sum':
+            _score = torch.sum(_score)
+        elif reduction is None:
+            pass
+        else:
+            raise ValueError(
+                f'reduction must be `mean`, `sum` or None; {reduction} passed'
+            )
+
         return _score
 
 

supirfactor_dynamical/postprocessing/results.py

@@ -158,12 +158,14 @@ def add_classification_metrics_to_dataframe(
     result_df[column_prefix + "accuracy"] = model_object.score(
         training_dataloader,
         loss_function=torcheval.metrics.functional.multilabel_accuracy,
-        criteria='hamming'
+        criteria='hamming',
+        reduction='mean'
     ).item()
 
     result_df[column_prefix + "auprc"] = model_object.score(
         training_dataloader,
-        loss_function=torcheval.metrics.functional.multilabel_auprc
+        loss_function=torcheval.metrics.functional.multilabel_auprc,
+        reduction='mean'
     ).item()
 
     result_df[column_prefix + "cross_entropy"] = model_object.score(