ref: (2/n) fix no log in epoch end (#3699)

pytorch_lightning/trainer/connectors/logger_connector.py

@@ -272,30 +272,31 @@ def log_train_epoch_end_metrics(self,
             self.callback_metrics.update(epoch_progress_bar_metrics)
 
     def training_epoch_end(self, model, epoch_output, num_optimizers):
+        if not is_overridden('training_epoch_end', model=model):
+            return Result()
+
         # run training_epoch_end
-        # a list with a result per optimizer index
-        if is_overridden('training_epoch_end', model=model):
-            # refresh the result for custom logging at the epoch level
-            model._current_fx_name = 'training_epoch_end'
-            model._results = Result()
+        # refresh the result for custom logging at the epoch level
+        model._current_fx_name = 'training_epoch_end'
+        model._results = Result()
 
-            epoch_output = self.__prepare_epoch_end_inputs(epoch_output)
+        epoch_output = self.__prepare_epoch_end_inputs(epoch_output)
 
-            if num_optimizers == 1:
-                epoch_output = epoch_output[0]
+        if num_optimizers == 1:
+            epoch_output = epoch_output[0]
 
-            # lightningmodule hook
-            epoch_output = model.training_epoch_end(epoch_output)
+        # lightningmodule hook
+        epoch_output = model.training_epoch_end(epoch_output)
 
-            model._current_fx_name = ''
+        model._current_fx_name = ''
 
-            if epoch_output is not None:
-                raise MisconfigurationException('training_epoch_end expects a return of None. '
-                                                'HINT: remove the return statement in training_epoch_end')
+        if epoch_output is not None:
+            raise MisconfigurationException('training_epoch_end expects a return of None. '
+                                            'HINT: remove the return statement in training_epoch_end')
 
-            # user can ALSO log at the end of an epoch
-            new_epoch_end_logs = model._results
-            return new_epoch_end_logs
+        # user can ALSO log at the end of an epoch
+        new_epoch_end_logs = model._results
+        return new_epoch_end_logs
 
     def __run_legacy_training_epoch_end(
             self,

tests/trainer/test_trainining_step_no_dict_result.py

@@ -7,6 +7,60 @@
 import torch
 
 
+def test_training_step_scalar_no_epoch_end_log(tmpdir):
+    """
+    Tests that only training_step can be used
+    """
+    os.environ['PL_DEV_DEBUG'] = '1'
+
+    class TestModel(DeterministicModel):
+        def training_step(self, batch, batch_idx):
+            acc = self.step(batch, batch_idx)
+            acc = acc + batch_idx
+            self.log('step_acc', acc, on_step=True, on_epoch=False)
+            self.log('epoch_acc', acc, on_step=False, on_epoch=True)
+            self.log('no_prefix_step_epoch_acc', acc, on_step=True, on_epoch=True)
+            self.log('pbar_step_acc', acc, on_step=True, prog_bar=True, on_epoch=False, logger=False)
+            self.log('pbar_epoch_acc', acc, on_step=False, on_epoch=True, prog_bar=True, logger=False)
+            self.log('pbar_step_epoch_acc', acc, on_step=True, on_epoch=True, prog_bar=True, logger=False)
+
+            self.training_step_called = True
+            return acc
+
+        def backward(self, trainer, loss, optimizer, optimizer_idx):
+            loss.backward()
+
+    model = TestModel()
+    model.val_dataloader = None
+
+    trainer = Trainer(
+        default_root_dir=tmpdir,
+        limit_train_batches=2,
+        limit_val_batches=2,
+        max_epochs=2,
+        row_log_interval=1,
+        weights_summary=None,
+    )
+    trainer.fit(model)
+
+    # make sure correct steps were called
+    assert model.training_step_called
+    assert not model.training_step_end_called
+
+    # make sure all the metrics are available for callbacks
+    metrics = [
+        'step_acc',
+        'epoch_acc',
+        'no_prefix_step_epoch_acc', 'step_no_prefix_step_epoch_acc', 'epoch_no_prefix_step_epoch_acc',
+        'pbar_step_acc',
+        'pbar_epoch_acc',
+        'pbar_step_epoch_acc', 'step_pbar_step_epoch_acc', 'epoch_pbar_step_epoch_acc',
+    ]
+    expected_metrics = set(metrics + ['debug_epoch'])
+    callback_metrics = set(trainer.callback_metrics.keys())
+    assert expected_metrics == callback_metrics
+
+
 def test_training_step_scalar(tmpdir):
     """
     Tests that only training_step can be used