add some tests

tests/base/model_train_steps.py

@@ -21,12 +21,21 @@ def training_step(self, batch, batch_idx, optimizer_idx=None):
 
         # calculate loss
         loss_val = self.loss(y, y_hat)
+        loss_scalar = loss_val.item()
 
         # alternate possible outputs to test
+        if batch_idx % 2 == 0:
+            output = OrderedDict({
+                'loss': loss_val,
+                'progress_bar': {'some_val': loss_val * loss_val},
+                'log': {'train_some_val': loss_val * loss_val},
+            })
+
+        # return scalars for "log" and "progress_bar"
         output = OrderedDict({
             'loss': loss_val,
-            'progress_bar': {'some_val': loss_val * loss_val},
-            'log': {'train_some_val': loss_val * loss_val},
+            'progress_bar': {'some_val': loss_scalar * loss_scalar},
+            'log': {'train_some_val': loss_scalar * loss_scalar},
         })
         return output
 

tests/base/model_valid_epoch_ends.py

@@ -7,6 +7,7 @@ class ValidationEpochEndVariations(ABC):
     """
     Houses all variations of validation_epoch_end steps
     """
+
     def validation_epoch_end(self, outputs):
         """
         Called at the end of validation to aggregate outputs
@@ -23,7 +24,11 @@ def _mean(res, key):
         val_loss_mean = _mean(outputs, 'val_loss')
         val_acc_mean = _mean(outputs, 'val_acc')
 
-        metrics_dict = {'val_loss': val_loss_mean.item(), 'val_acc': val_acc_mean.item()}
+        # alternate between tensor and scalar
+        if self.current_epoch % 2:
+            metrics_dict = {'val_loss': val_loss_mean.item(), 'val_acc': val_acc_mean.item()}
+        else:
+            metrics_dict = {'val_loss': val_loss_mean, 'val_acc': val_acc_mean}
         results = {'progress_bar': metrics_dict, 'log': metrics_dict}
         return results