new section + legacy 1.1.6

CHANGELOG.md

@@ -5,6 +5,24 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 
+## [unreleased.Bugfixes] - YYYY-MM-DD
+
+### Added
+
+
+### Changed
+
+
+### Deprecated
+
+
+### Removed
+
+
+### Fixed
+
+
+
 ## [1.1.6] - 2021-01-26
 
 ### Changed

tests/checkpointing/test_legacy_checkpoints.py

@@ -47,6 +47,7 @@
     "1.1.3",
     "1.1.4",
     "1.1.5",
+    "1.1.6",
 ])
 def test_resume_legacy_checkpoints(tmpdir, pl_version):
     path_dir = os.path.join(LEGACY_CHECKPOINTS_PATH, pl_version)

tests/models/test_hooks.py

@@ -94,11 +94,10 @@ def training_epoch_end(self, outputs):
 
 def test_training_epoch_end_metrics_collection_on_override(tmpdir):
     """ Test that batch end metrics are collected when training_epoch_end is overridden at the end of an epoch. """
-    num_epochs = 1
 
     class LoggingCallback(pl.Callback):
 
-        def on_train_epoch_end(self, trainer, pl_module):
+        def on_train_epoch_start(self, trainer, pl_module):
             self.len_outputs = 0
 
         def on_train_epoch_end(self, trainer, pl_module, outputs):
@@ -110,7 +109,6 @@ def on_train_epoch_start(self):
             self.num_train_batches = 0
 
         def training_epoch_end(self, outputs):  # Overridden
-            pass
             return
 
         def on_train_batch_end(self, outputs, batch, batch_idx, dataloader_idx):
@@ -129,19 +127,20 @@ def on_train_batch_end(self, outputs, batch, batch_idx, dataloader_idx):
 
     callback = LoggingCallback()
     trainer = Trainer(
-        max_epochs=num_epochs,
+        max_epochs=1,
         default_root_dir=tmpdir,
         overfit_batches=2,
         callbacks=[callback],
     )
 
-    result = trainer.fit(overridden_model)
+    trainer.fit(overridden_model)
+    # outputs from on_train_batch_end should be accessible in on_train_epoch_end hook
+    # if training_epoch_end is overridden
     assert callback.len_outputs == overridden_model.num_train_batches
-    # outputs from on_train_batch_end should be accessible in on_train_epoch_end hook if training_epoch_end is overridden
 
-    result = trainer.fit(not_overridden_model)
-    assert callback.len_outputs == 0
+    trainer.fit(not_overridden_model)
     # outputs from on_train_batch_end should be empty
+    assert callback.len_outputs == 0
 
 
 @pytest.mark.skipif(not torch.cuda.is_available(), reason="test requires GPU machine")