Enforce an epoch scheduler interval when using SWA (#6588)

Co-authored-by: Carlos Mocholi <[email protected]>

Author: 2 people

pytorch_lightning/callbacks/swa.py

@@ -189,14 +189,15 @@ def on_train_epoch_start(self, trainer: 'pl.Trainer', pl_module: 'pl.LightningMo
                 anneal_strategy=self._annealing_strategy,
                 last_epoch=trainer.max_epochs if self._annealing_strategy == "cos" else -1
             )
+            _scheduler_config = _get_default_scheduler_config()
+            assert _scheduler_config["interval"] == "epoch" and _scheduler_config["frequency"] == 1
+            _scheduler_config["scheduler"] = self._swa_scheduler
 
             if trainer.lr_schedulers:
                 lr_scheduler = trainer.lr_schedulers[0]["scheduler"]
                 rank_zero_warn(f"Swapping lr_scheduler {lr_scheduler} for {self._swa_scheduler}")
-                trainer.lr_schedulers[0]["scheduler"] = self._swa_scheduler
+                trainer.lr_schedulers[0] = _scheduler_config
             else:
-                _scheduler_config = _get_default_scheduler_config()
-                _scheduler_config["scheduler"] = self._swa_scheduler
                 trainer.lr_schedulers.append(_scheduler_config)
 
             self.n_averaged = torch.tensor(0, dtype=torch.long, device=pl_module.device)

tests/callbacks/test_swa.py

@@ -24,19 +24,22 @@
 from pytorch_lightning.utilities import _TORCH_GREATER_EQUAL_1_6
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
 from tests.helpers import BoringModel, RandomDataset
+from tests.helpers.runif import RunIf
 
 if _TORCH_GREATER_EQUAL_1_6:
     from pytorch_lightning.callbacks import StochasticWeightAveraging
+    from torch.optim.swa_utils import SWALR
 
     class SwaTestModel(BoringModel):
 
-        def __init__(self, batchnorm: bool = True):
+        def __init__(self, batchnorm: bool = True, interval: str = "epoch"):
             super().__init__()
             layers = [nn.Linear(32, 32)]
             if batchnorm:
                 layers.append(nn.BatchNorm1d(32))
             layers += [nn.ReLU(), nn.Linear(32, 2)]
             self.layer = nn.Sequential(*layers)
+            self.interval = interval
 
         def training_step(self, batch, batch_idx):
             output = self.forward(batch)
@@ -46,6 +49,14 @@ def training_step(self, batch, batch_idx):
         def train_dataloader(self):
             return DataLoader(RandomDataset(32, 64), batch_size=2)
 
+        def configure_optimizers(self):
+            optimizer = torch.optim.SGD(self.layer.parameters(), lr=0.1)
+            return {
+                "optimizer": optimizer,
+                "scheduler": torch.optim.lr_scheduler.StepLR(optimizer, step_size=1),
+                "interval": self.interval,
+            }
+
     class SwaTestCallback(StochasticWeightAveraging):
         update_parameters_calls: int = 0
         transfer_weights_calls: int = 0
@@ -61,6 +72,10 @@ def transfer_weights(self, *args, **kwargs):
         def on_train_epoch_start(self, trainer, *args):
             super().on_train_epoch_start(trainer, *args)
             assert trainer.train_loop._skip_backward == (trainer.current_epoch > self.swa_end)
+            if self.swa_start <= trainer.current_epoch:
+                assert isinstance(trainer.lr_schedulers[0]["scheduler"], SWALR)
+                assert trainer.lr_schedulers[0]["interval"] == "epoch"
+                assert trainer.lr_schedulers[0]["frequency"] == 1
 
         def on_train_epoch_end(self, trainer, *args):
             super().on_train_epoch_end(trainer, *args)
@@ -89,8 +104,8 @@ def on_train_end(self, trainer, pl_module):
 
 
 @mock.patch.dict(os.environ, {"PL_DEV_DEBUG": "1"})
-def train_with_swa(tmpdir, batchnorm=True, accelerator=None, gpus=None, num_processes=1):
-    model = SwaTestModel(batchnorm=batchnorm)
+def train_with_swa(tmpdir, batchnorm=True, accelerator=None, gpus=None, num_processes=1, interval="epoch"):
+    model = SwaTestModel(batchnorm=batchnorm, interval=interval)
     swa_start = 2
     max_epochs = 5
     swa_callback = SwaTestCallback(swa_epoch_start=swa_start, swa_lrs=0.1)
@@ -147,7 +162,13 @@ def test_swa_callback(tmpdir, batchnorm):
     train_with_swa(tmpdir, batchnorm=batchnorm)
 
 
-@pytest.mark.skipif(not _TORCH_GREATER_EQUAL_1_6, reason="SWA available from PyTorch 1.6.0")
+@RunIf(min_torch="1.6.0")
+@pytest.mark.parametrize("interval", ("epoch", "step"))
+def test_swa_callback_scheduler_step(tmpdir, interval: bool):
+    train_with_swa(tmpdir, interval=interval)
+
+
+@RunIf(min_torch="1.6.0")
 def test_swa_raises():
     with pytest.raises(MisconfigurationException, match=">0 integer or a float between 0 and 1"):
         StochasticWeightAveraging(swa_epoch_start=0, swa_lrs=0.1)