Do not warn about the scheduler's interval key during manual optim

src/pytorch_lightning/CHANGELOG.md

@@ -160,6 +160,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed a type error when dividing the chunk size in the ColossalAI strategy ([#16212](https://github.com/Lightning-AI/lightning/pull/16212))
 
 
+- Fixed bug where the ``interval`` key of the scheduler would be ignored during manual optimization, making the LearningRateMonitor callback fail to log the learning rate ([#16308](https://github.com/Lightning-AI/lightning/pull/16308))
+
+
 ## [1.8.6] - 2022-12-21
 
 - minor cleaning

src/pytorch_lightning/core/optimizer.py

@@ -322,7 +322,9 @@ def _configure_schedulers_manual_opt(schedulers: list) -> List[LRSchedulerConfig
     lr_scheduler_configs = []
     for scheduler in schedulers:
         if isinstance(scheduler, dict):
-            invalid_keys = {"interval", "frequency", "reduce_on_plateau", "monitor", "strict"}
+            # interval is not in this list even though the user needs to manually call the scheduler because
+            # the `LearningRateMonitor` callback needs to check its value to know when to log the learning rate
+            invalid_keys = {"frequency", "reduce_on_plateau", "monitor", "strict"}
             keys_to_warn = [k for k in scheduler.keys() if k in invalid_keys]
 
             if keys_to_warn:

tests/tests_pytorch/trainer/optimization/test_manual_optimization.py

@@ -933,6 +933,7 @@ def configure_optimizers(self):
                 scheduler = {
                     "scheduler": torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer),
                     "monitor": "train_loss",
+                    "interval": "step",  # not warned
                 }
             else:
                 scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer)
@@ -946,8 +947,9 @@ def configure_optimizers(self):
     )
 
     if scheduler_as_dict:
-        with pytest.warns(RuntimeWarning, match="but the keys will be ignored"):
+        with pytest.warns(RuntimeWarning, match=r"\['monitor'\], but the keys will be ignored"):
             trainer.fit(model)
+        assert trainer.lr_scheduler_configs[0].interval == "step"
     else:
         trainer.fit(model)
 

tests/tests_pytorch/trainer/optimization/test_optimizers.py

@@ -557,25 +557,6 @@ def configure_optimizers(self):
         trainer.fit(model)
 
 
-def test_warn_invalid_scheduler_key_in_manual_optimization(tmpdir):
-    """Test warning when invalid scheduler keys are provided in manual optimization."""
-
-    class TestModel(BoringModel):
-        def __init__(self):
-            super().__init__()
-            self.automatic_optimization = False
-
-        def configure_optimizers(self):
-            opt = optim.SGD(self.layer.parameters(), lr=0.1)
-            sch = optim.lr_scheduler.StepLR(opt, step_size=1)
-            return [opt], [{"scheduler": sch, "interval": "epoch"}]
-
-    model = TestModel()
-    trainer = Trainer(default_root_dir=tmpdir, fast_dev_run=True)
-    with pytest.warns(RuntimeWarning, match="the keys will be ignored"):
-        trainer.fit(model)
-
-
 @RunIf(min_cuda_gpus=2, standalone=True)
 def test_optimizer_state_on_device(tmpdir):
     """Test that optimizers that create state initially at instantiation still end up with the state on the GPU."""