Set the logger explicitly in tests (#15815)

Author: carmocca

requirements/pytorch/test.txt

@@ -12,8 +12,8 @@ scikit-learn>0.22.1, <1.1.3
 onnxruntime<1.14.0
 psutil<5.9.4 # for `DeviceStatsMonitor`
 pandas>1.0, <1.5.2  # needed in benchmarks
-fastapi<0.87.0
-uvicorn<0.19.1
+fastapi<0.87.0  # for `ServableModuleValidator`
+uvicorn<0.19.1  # for `ServableModuleValidator`
 
-tensorboard>=2.9.1, <2.12.0
+tensorboard>=2.9.1, <2.12.0  # for `TensorBoardLogger`
 protobuf<=3.20.1 # strict  # an extra is updating protobuf, this pin prevents TensorBoard failure

tests/tests_pytorch/callbacks/progress/test_rich_progress_bar.py

@@ -23,6 +23,7 @@
 from pytorch_lightning.callbacks import ProgressBarBase, RichProgressBar
 from pytorch_lightning.callbacks.progress.rich_progress import RichProgressBarTheme
 from pytorch_lightning.demos.boring_classes import BoringModel, RandomDataset, RandomIterableDataset
+from pytorch_lightning.loggers import CSVLogger
 from tests_pytorch.helpers.runif import RunIf
 
 
@@ -330,7 +331,7 @@ def training_step(self, *args, **kwargs):
 
     progress_bar = RichProgressBar()
     model = CustomModel()
-    trainer = Trainer(default_root_dir=tmpdir, callbacks=progress_bar, fast_dev_run=True)
+    trainer = Trainer(default_root_dir=tmpdir, callbacks=progress_bar, fast_dev_run=True, logger=CSVLogger(tmpdir))
 
     trainer.fit(model)
     main_progress_bar_id = progress_bar.main_progress_bar_id
@@ -384,6 +385,7 @@ def test_step(self, batch, batch_idx):
         enable_checkpointing=False,
         log_every_n_steps=1,
         callbacks=pbar,
+        logger=CSVLogger(tmpdir),
     )
 
     trainer.fit(model)

tests/tests_pytorch/callbacks/progress/test_tqdm_progress_bar.py

@@ -29,6 +29,7 @@
 from pytorch_lightning.callbacks.progress.tqdm_progress import Tqdm
 from pytorch_lightning.core.module import LightningModule
 from pytorch_lightning.demos.boring_classes import BoringModel, RandomDataset
+from pytorch_lightning.loggers import CSVLogger
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
 from tests_pytorch.helpers.runif import RunIf
 
@@ -706,6 +707,7 @@ def test_step(self, batch, batch_idx):
         enable_checkpointing=False,
         log_every_n_steps=1,
         callbacks=pbar,
+        logger=CSVLogger(tmpdir),
     )
 
     trainer.fit(model)

tests/tests_pytorch/callbacks/test_device_stats_monitor.py

@@ -155,13 +155,13 @@ def test_prefix_metric_keys():
     assert converted_metrics == {"foo.1": 1.0, "foo.2": 2.0, "foo.3": 3.0}
 
 
-def test_device_stats_monitor_warning_when_psutil_not_available(monkeypatch):
+def test_device_stats_monitor_warning_when_psutil_not_available(monkeypatch, tmp_path):
     """Test that warning is raised when psutil is not available."""
     import pytorch_lightning.callbacks.device_stats_monitor as imports
 
     monkeypatch.setattr(imports, "_PSUTIL_AVAILABLE", False)
     monitor = DeviceStatsMonitor()
-    trainer = Trainer()
+    trainer = Trainer(logger=CSVLogger(tmp_path))
     assert trainer.strategy.root_device == torch.device("cpu")
     # TODO: raise an exception from v1.9
     with pytest.warns(UserWarning, match="psutil` is not installed"):

tests/tests_pytorch/callbacks/test_lr_monitor.py

@@ -20,6 +20,7 @@
 from pytorch_lightning.callbacks.callback import Callback
 from pytorch_lightning.callbacks.finetuning import BackboneFinetuning
 from pytorch_lightning.demos.boring_classes import BoringModel
+from pytorch_lightning.loggers import CSVLogger
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
 from tests_pytorch.helpers.datamodules import ClassifDataModule
 from tests_pytorch.helpers.runif import RunIf
@@ -32,7 +33,12 @@ def test_lr_monitor_single_lr(tmpdir):
 
     lr_monitor = LearningRateMonitor()
     trainer = Trainer(
-        default_root_dir=tmpdir, max_epochs=2, limit_val_batches=0.1, limit_train_batches=0.5, callbacks=[lr_monitor]
+        default_root_dir=tmpdir,
+        max_epochs=2,
+        limit_val_batches=0.1,
+        limit_train_batches=0.5,
+        callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
     )
     trainer.fit(model)
 
@@ -70,6 +76,7 @@ def configure_optimizers(self):
         limit_train_batches=5,
         log_every_n_steps=1,
         callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
     )
     trainer.fit(model)
 
@@ -96,6 +103,7 @@ def configure_optimizers(self):
         limit_train_batches=5,
         log_every_n_steps=1,
         callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
     )
     with pytest.warns(RuntimeWarning, match="optimizers do not have momentum."):
         trainer.fit(model)
@@ -117,7 +125,12 @@ def configure_optimizers(self):
 
     lr_monitor = LearningRateMonitor()
     trainer = Trainer(
-        default_root_dir=tmpdir, max_epochs=2, limit_val_batches=0.1, limit_train_batches=0.5, callbacks=[lr_monitor]
+        default_root_dir=tmpdir,
+        max_epochs=2,
+        limit_val_batches=0.1,
+        limit_train_batches=0.5,
+        callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
     )
 
     trainer.fit(model)
@@ -154,6 +167,7 @@ def configure_optimizers(self):
         limit_train_batches=5,
         log_every_n_steps=1,
         callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
     )
     trainer.fit(model)
 
@@ -179,6 +193,7 @@ def configure_optimizers(self):
         limit_train_batches=5,
         log_every_n_steps=1,
         callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
     )
     with pytest.warns(RuntimeWarning, match="optimizers do not have momentum."):
         trainer.fit(model)
@@ -226,6 +241,7 @@ def configure_optimizers(self):
         limit_train_batches=7,
         limit_val_batches=0.1,
         callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
     )
     trainer.fit(model)
 
@@ -269,6 +285,7 @@ def configure_optimizers(self):
         limit_train_batches=7,
         limit_val_batches=0.1,
         callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
     )
     trainer.fit(model)
 
@@ -305,7 +322,12 @@ def configure_optimizers(self):
 
     lr_monitor = LearningRateMonitor()
     trainer = Trainer(
-        default_root_dir=tmpdir, max_epochs=2, limit_val_batches=0.1, limit_train_batches=0.5, callbacks=[lr_monitor]
+        default_root_dir=tmpdir,
+        max_epochs=2,
+        limit_val_batches=0.1,
+        limit_train_batches=0.5,
+        callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
     )
     trainer.fit(model, datamodule=dm)
 
@@ -330,6 +352,7 @@ def configure_optimizers(self):
         callbacks=[lr_monitor],
         enable_progress_bar=False,
         enable_model_summary=False,
+        logger=CSVLogger(tmpdir),
     )
     trainer.fit(TestModel())
     assert list(lr_monitor.lrs) == ["my_logging_name"]
@@ -349,6 +372,7 @@ def configure_optimizers(self):
         limit_val_batches=2,
         limit_train_batches=2,
         callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
         enable_progress_bar=False,
         enable_model_summary=False,
     )
@@ -384,6 +408,7 @@ def configure_optimizers(self):
         limit_val_batches=2,
         limit_train_batches=2,
         callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
         enable_progress_bar=False,
         enable_model_summary=False,
     )
@@ -475,6 +500,7 @@ def finetune_function(self, pl_module, epoch: int, optimizer, opt_idx: int):
         limit_val_batches=0,
         limit_train_batches=2,
         callbacks=[TestFinetuning(), lr_monitor, Check()],
+        logger=CSVLogger(tmpdir),
         enable_progress_bar=False,
         enable_model_summary=False,
         enable_checkpointing=False,
@@ -533,6 +559,7 @@ def configure_optimizers(self):
         limit_val_batches=2,
         limit_train_batches=2,
         callbacks=[lr_monitor],
+        logger=CSVLogger(tmpdir),
         enable_progress_bar=False,
         enable_model_summary=False,
     )

tests/tests_pytorch/callbacks/test_stochastic_weight_avg.py

@@ -303,13 +303,14 @@ def _swa_resume_training_from_checkpoint(tmpdir, model, resume_model, ddp=False)
         "limit_val_batches": 0,
         "accumulate_grad_batches": 2,
         "enable_progress_bar": False,
+        "logger": False,
     }
     trainer = Trainer(callbacks=SwaTestCallback(swa_epoch_start=swa_start, swa_lrs=0.1), **trainer_kwargs)
 
     with _backward_patch(trainer), pytest.raises(Exception, match="SWA crash test"):
         trainer.fit(model)
 
-    checkpoint_dir = Path(tmpdir) / "lightning_logs" / "version_0" / "checkpoints"
+    checkpoint_dir = Path(tmpdir) / "checkpoints"
     checkpoint_files = os.listdir(checkpoint_dir)
     assert len(checkpoint_files) == 1
     ckpt_path = str(checkpoint_dir / checkpoint_files[0])

tests/tests_pytorch/checkpointing/test_model_checkpoint.py

@@ -35,7 +35,7 @@
 from pytorch_lightning import seed_everything, Trainer
 from pytorch_lightning.callbacks import ModelCheckpoint
 from pytorch_lightning.demos.boring_classes import BoringModel
-from pytorch_lightning.loggers import TensorBoardLogger
+from pytorch_lightning.loggers import CSVLogger, TensorBoardLogger
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
 from pytorch_lightning.utilities.imports import _OMEGACONF_AVAILABLE
 from tests_pytorch.helpers.runif import RunIf
@@ -301,9 +301,11 @@ def test_model_checkpoint_with_non_string_input(tmpdir, save_top_k: int):
 
     checkpoint = ModelCheckpoint(monitor="early_stop_on", dirpath=None, filename="{epoch}", save_top_k=save_top_k)
     max_epochs = 2
-    trainer = Trainer(default_root_dir=tmpdir, callbacks=[checkpoint], overfit_batches=0.20, max_epochs=max_epochs)
+    trainer = Trainer(
+        default_root_dir=tmpdir, callbacks=[checkpoint], overfit_batches=0.20, max_epochs=max_epochs, logger=False
+    )
     trainer.fit(model)
-    assert checkpoint.dirpath == tmpdir / trainer.logger.name / "version_0" / "checkpoints"
+    assert checkpoint.dirpath == tmpdir / "checkpoints"
 
     if save_top_k == -1:
         ckpt_files = os.listdir(checkpoint.dirpath)
@@ -753,15 +755,20 @@ def test_default_checkpoint_behavior(tmpdir):
 
     model = LogInTwoMethods()
     trainer = Trainer(
-        default_root_dir=tmpdir, max_epochs=3, enable_progress_bar=False, limit_train_batches=5, limit_val_batches=5
+        default_root_dir=tmpdir,
+        max_epochs=3,
+        enable_progress_bar=False,
+        limit_train_batches=5,
+        limit_val_batches=5,
+        logger=False,
     )
 
     with patch.object(trainer, "save_checkpoint", wraps=trainer.save_checkpoint) as save_mock:
         trainer.fit(model)
         results = trainer.test()
 
     assert len(results) == 1
-    save_dir = tmpdir / "lightning_logs" / "version_0" / "checkpoints"
+    save_dir = tmpdir / "checkpoints"
     save_weights_only = trainer.checkpoint_callback.save_weights_only
     save_mock.assert_has_calls(
         [
@@ -867,6 +874,7 @@ def validation_step(self, batch, batch_idx):
         "enable_model_summary": False,
         "log_every_n_steps": 1,
         "default_root_dir": tmpdir,
+        "logger": CSVLogger(tmpdir),
     }
     trainer = Trainer(**trainer_kwargs, callbacks=[checkpoint_callback])
     trainer.fit(model)
@@ -931,6 +939,7 @@ def assert_checkpoint_log_dir(idx):
         limit_val_batches=3,
         limit_test_batches=4,
         callbacks=[checkpoint_cb],
+        logger=TensorBoardLogger(tmpdir),
     )
     trainer = Trainer(**trainer_config)
     assert_trainer_init(trainer)
@@ -953,6 +962,7 @@ def assert_checkpoint_log_dir(idx):
         assert_checkpoint_content(ckpt_dir)
 
         # load from checkpoint
+        trainer_config["logger"] = TensorBoardLogger(tmpdir)
         trainer = pl.Trainer(**trainer_config)
         assert_trainer_init(trainer)
 

tests/tests_pytorch/loggers/test_logger.py

@@ -239,7 +239,12 @@ def __init__(self, param_one, param_two):
 
     model = TestModel("pytorch", "lightning")
     trainer = Trainer(
-        default_root_dir=tmpdir, max_epochs=1, limit_train_batches=0.1, limit_val_batches=0.1, num_sanity_val_steps=0
+        default_root_dir=tmpdir,
+        max_epochs=1,
+        limit_train_batches=0.1,
+        limit_val_batches=0.1,
+        num_sanity_val_steps=0,
+        logger=TensorBoardLogger(tmpdir),
     )
     trainer.fit(model)
 
@@ -270,6 +275,7 @@ class _Test:
 
     trainer = Trainer(
         default_root_dir=tmpdir,
+        logger=TensorBoardLogger(tmpdir),
         max_epochs=1,
         limit_train_batches=0.1,
         limit_val_batches=0.1,
@@ -294,6 +300,7 @@ class _Test:
     dm = TestDataModule(diff_params)
     trainer = Trainer(
         default_root_dir=tmpdir,
+        logger=TensorBoardLogger(tmpdir),
         max_epochs=1,
         limit_train_batches=0.1,
         limit_val_batches=0.1,
@@ -311,6 +318,7 @@ class _Test:
     dm = TestDataModule(tensor_params)
     trainer = Trainer(
         default_root_dir=tmpdir,
+        logger=TensorBoardLogger(tmpdir),
         max_epochs=1,
         limit_train_batches=0.1,
         limit_val_batches=0.1,

tests/tests_pytorch/loggers/test_tensorboard.py

@@ -39,7 +39,7 @@ def __init__(self, b1=0.5, b2=0.999):
             super().__init__()
             self.save_hyperparameters()
 
-    trainer = Trainer(max_steps=1, default_root_dir=tmpdir)
+    trainer = Trainer(max_steps=1, default_root_dir=tmpdir, logger=TensorBoardLogger(tmpdir))
     model = CustomModel()
     assert trainer.log_dir == trainer.logger.log_dir
     trainer.fit(model)

tests/tests_pytorch/models/test_grad_norm.py

@@ -18,6 +18,7 @@
 
 from pytorch_lightning import Trainer
 from pytorch_lightning.demos.boring_classes import BoringModel
+from pytorch_lightning.loggers import CSVLogger
 
 
 class ModelWithManualGradTracker(BoringModel):
@@ -86,7 +87,13 @@ def on_train_batch_end(self, *_) -> None:
 @pytest.mark.parametrize("log_every_n_steps", [1, 2, 3])
 def test_grad_tracking_interval(tmpdir, log_every_n_steps):
     """Test that gradient norms get tracked in the right interval and that everytime the same keys get logged."""
-    trainer = Trainer(default_root_dir=tmpdir, track_grad_norm=2, log_every_n_steps=log_every_n_steps, max_steps=10)
+    trainer = Trainer(
+        default_root_dir=tmpdir,
+        track_grad_norm=2,
+        log_every_n_steps=log_every_n_steps,
+        max_steps=10,
+        logger=CSVLogger(tmpdir),
+    )
 
     with patch.object(trainer.logger, "log_metrics") as mocked:
         model = BoringModel()