Remove deprecated LoggerCollection (#14283)

Co-authored-by: Rohit Gupta <[email protected]>
Co-authored-by: Jirka Borovec <[email protected]>

Author: 3 people

src/pytorch_lightning/CHANGELOG.md

@@ -128,6 +128,9 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Removed the experimental `pytorch_lightning.utiltiies.meta` functions in favor of built-in https://github.com/pytorch/torchdistx support ([#13868](https://github.com/Lightning-AI/lightning/pull/13868))
 
 
+- Removed the deprecated `LoggerCollection`; `Trainer.logger` and `LightningModule.logger` now returns the first logger when more than one gets passed to the Trainer ([#14283](https://github.com/Lightning-AI/lightning/pull/14283))
+
+
 - Removed the deprecated the `trainer.lr_schedulers` ([#14408](https://github.com/Lightning-AI/lightning/pull/14408))
 
 

src/pytorch_lightning/core/module.py

@@ -18,7 +18,6 @@
 import numbers
 import os
 import tempfile
-import warnings
 import weakref
 from contextlib import contextmanager
 from pathlib import Path
@@ -43,7 +42,7 @@
 from pytorch_lightning.core.mixins import HyperparametersMixin
 from pytorch_lightning.core.optimizer import LightningOptimizer
 from pytorch_lightning.core.saving import ModelIO
-from pytorch_lightning.loggers import Logger, LoggerCollection
+from pytorch_lightning.loggers import Logger
 from pytorch_lightning.trainer.connectors.logger_connector.fx_validator import _FxValidator
 from pytorch_lightning.utilities import _IS_WINDOWS, _TORCH_GREATER_EQUAL_1_10, GradClipAlgorithmType
 from pytorch_lightning.utilities.exceptions import MisconfigurationException
@@ -267,26 +266,7 @@ def truncated_bptt_steps(self, truncated_bptt_steps: int) -> None:
     @property
     def logger(self) -> Optional[Logger]:
         """Reference to the logger object in the Trainer."""
-        # this should match the implementation of `trainer.logger`
-        # we don't reuse it so we can properly set the deprecation stacklevel
-        if self._trainer is None:
-            return None
-        loggers = self.trainer.loggers
-        if len(loggers) == 0:
-            return None
-        if len(loggers) == 1:
-            return loggers[0]
-        else:
-            if not self._running_torchscript:
-                rank_zero_deprecation(
-                    "Using `lightning_module.logger` when multiple loggers are configured."
-                    " This behavior will change in v1.8 when `LoggerCollection` is removed, and"
-                    " `lightning_module.logger` will return the first logger available.",
-                    stacklevel=5,
-                )
-            with warnings.catch_warnings():
-                warnings.simplefilter("ignore")
-                return LoggerCollection(loggers)
+        return self._trainer.logger if self._trainer is not None else None
 
     @property
     def loggers(self) -> List[Logger]:

src/pytorch_lightning/loggers/__init__.py

@@ -17,13 +17,13 @@
 from pytorch_lightning.loggers.base import LightningLoggerBase
 from pytorch_lightning.loggers.comet import _COMET_AVAILABLE, CometLogger  # noqa: F401
 from pytorch_lightning.loggers.csv_logs import CSVLogger
-from pytorch_lightning.loggers.logger import Logger, LoggerCollection
+from pytorch_lightning.loggers.logger import Logger
 from pytorch_lightning.loggers.mlflow import _MLFLOW_AVAILABLE, MLFlowLogger  # noqa: F401
 from pytorch_lightning.loggers.neptune import NeptuneLogger  # noqa: F401
 from pytorch_lightning.loggers.tensorboard import TensorBoardLogger
 from pytorch_lightning.loggers.wandb import WandbLogger  # noqa: F401
 
-__all__ = ["CSVLogger", "LightningLoggerBase", "Logger", "LoggerCollection", "TensorBoardLogger"]
+__all__ = ["CSVLogger", "LightningLoggerBase", "Logger", "TensorBoardLogger"]
 
 if _COMET_AVAILABLE:
     __all__.append("CometLogger")

src/pytorch_lightning/loggers/base.py

@@ -57,11 +57,6 @@ def __init__(self, *args, **kwargs) -> None:  # type: ignore[no-untyped-def]
         super().__init__(*args, **kwargs)
 
 
-class LoggerCollection(logger.LoggerCollection):
-    def __init__(self, *args, **kwargs) -> None:  # type: ignore[no-untyped-def]
-        super().__init__(*args, **kwargs)
-
-
 class DummyExperiment(logger.DummyExperiment):
     def __init__(self, *args, **kwargs) -> None:  # type: ignore[no-untyped-def]
         rank_zero_deprecation(

src/pytorch_lightning/loggers/logger.py

@@ -20,7 +20,7 @@
 from argparse import Namespace
 from collections import defaultdict
 from functools import wraps
-from typing import Any, Callable, Dict, Generator, Iterable, List, Mapping, Optional, Sequence, Union
+from typing import Any, Callable, Dict, List, Mapping, Optional, Sequence, Union
 from weakref import ReferenceType
 
 import numpy as np
@@ -212,97 +212,6 @@ def version(self) -> Optional[Union[int, str]]:
         """Return the experiment version."""
 
 
-class LoggerCollection(Logger):
-    """The :class:`LoggerCollection` class is used to iterate all logging actions over the given `logger_iterable`.
-
-    .. deprecated:: v1.6
-        `LoggerCollection` is deprecated in v1.6 and will be removed in v1.8.
-        Directly pass a list of loggers to the Trainer and access the list via the `trainer.loggers` attribute.
-
-    Args:
-        logger_iterable: An iterable collection of loggers
-    """
-
-    def __init__(self, logger_iterable: Iterable[Logger]):
-        super().__init__()
-        self._logger_iterable = logger_iterable
-        rank_zero_deprecation(
-            "`LoggerCollection` is deprecated in v1.6 and will be removed in v1.8. Directly pass a list of loggers"
-            " to the Trainer and access the list via the `trainer.loggers` attribute."
-        )
-
-    def __getitem__(self, index: int) -> Logger:
-        return list(self._logger_iterable)[index]
-
-    def after_save_checkpoint(self, checkpoint_callback: "ReferenceType[Checkpoint]") -> None:
-        for logger in self._logger_iterable:
-            logger.after_save_checkpoint(checkpoint_callback)
-
-    def update_agg_funcs(
-        self,
-        agg_key_funcs: Optional[Mapping[str, Callable[[Sequence[float]], float]]] = None,
-        agg_default_func: Callable[[Sequence[float]], float] = np.mean,
-    ) -> None:
-        for logger in self._logger_iterable:
-            logger.update_agg_funcs(agg_key_funcs, agg_default_func)
-
-    @property
-    def experiment(self) -> List[Any]:
-        """Returns a list of experiment objects for all the loggers in the logger collection."""
-        return [logger.experiment for logger in self._logger_iterable]
-
-    def agg_and_log_metrics(self, metrics: Dict[str, float], step: Optional[int] = None) -> None:
-        for logger in self._logger_iterable:
-            logger.agg_and_log_metrics(metrics=metrics, step=step)
-
-    def log_metrics(self, metrics: Dict[str, float], step: Optional[int] = None) -> None:
-        for logger in self._logger_iterable:
-            logger.log_metrics(metrics=metrics, step=step)
-
-    def log_hyperparams(self, params: Union[Dict[str, Any], Namespace], *args: Any, **kwargs: Any) -> None:
-        for logger in self._logger_iterable:
-            logger.log_hyperparams(params, *args, **kwargs)
-
-    def log_graph(self, model: "pl.LightningModule", input_array: Optional[Tensor] = None) -> None:
-        for logger in self._logger_iterable:
-            logger.log_graph(model, input_array)
-
-    def log_text(self, *args: Any, **kwargs: Any) -> None:
-        for logger in self._logger_iterable:
-            logger.log_text(*args, **kwargs)
-
-    def log_image(self, *args: Any, **kwargs: Any) -> None:
-        for logger in self._logger_iterable:
-            logger.log_image(*args, **kwargs)
-
-    def save(self) -> None:
-        for logger in self._logger_iterable:
-            logger.save()
-
-    def finalize(self, status: str) -> None:
-        for logger in self._logger_iterable:
-            logger.finalize(status)
-
-    @property
-    def save_dir(self) -> Optional[str]:
-        """Returns ``None`` as checkpoints should be saved to default / chosen location when using multiple
-        loggers."""
-        # Checkpoints should be saved to default / chosen location when using multiple loggers
-        return None
-
-    @property
-    def name(self) -> str:
-        """Returns the unique experiment names for all the loggers in the logger collection joined by an
-        underscore."""
-        return "_".join(dict.fromkeys(str(logger.name) for logger in self._logger_iterable))
-
-    @property
-    def version(self) -> str:
-        """Returns the unique experiment versions for all the loggers in the logger collection joined by an
-        underscore."""
-        return "_".join(dict.fromkeys(str(logger.version) for logger in self._logger_iterable))
-
-
 class DummyExperiment:
     """Dummy experiment."""
 
@@ -355,10 +264,6 @@ def __getitem__(self, idx: int) -> "DummyLogger":
         # enables self.logger[0].experiment.add_image(...)
         return self
 
-    def __iter__(self) -> Generator[None, None, None]:
-        # if DummyLogger is substituting a logger collection, pretend it is empty
-        yield from ()
-
     def __getattr__(self, name: str) -> Callable:
         """Allows the DummyLogger to be called with arbitrary methods, to avoid AttributeErrors."""
 

src/pytorch_lightning/trainer/trainer.py

@@ -56,7 +56,7 @@
 from pytorch_lightning.core.datamodule import LightningDataModule
 from pytorch_lightning.core.optimizer import LightningOptimizer
 from pytorch_lightning.loggers import Logger
-from pytorch_lightning.loggers.logger import DummyLogger, LoggerCollection
+from pytorch_lightning.loggers.logger import DummyLogger
 from pytorch_lightning.loggers.tensorboard import TensorBoardLogger
 from pytorch_lightning.loops import PredictionLoop, TrainingEpochLoop
 from pytorch_lightning.loops.dataloader.evaluation_loop import EvaluationLoop
@@ -2604,28 +2604,12 @@ def _active_loop(self) -> Optional[Union[FitLoop, EvaluationLoop, PredictionLoop
 
     @property
     def logger(self) -> Optional[Logger]:
-        loggers = self.loggers
-        if len(loggers) == 0:
-            return None
-        if len(loggers) == 1:
-            return loggers[0]
-        else:
-            rank_zero_deprecation(
-                "Using `trainer.logger` when multiple loggers are configured."
-                " This behavior will change in v1.8 when `LoggerCollection` is removed, and"
-                " `trainer.logger` will return the first logger available.",
-                stacklevel=5,
-            )
-            with warnings.catch_warnings():
-                warnings.simplefilter("ignore")
-                return LoggerCollection(loggers)
+        return self.loggers[0] if len(self.loggers) > 0 else None
 
     @logger.setter
     def logger(self, logger: Optional[Logger]) -> None:
         if not logger:
             self.loggers = []
-        elif isinstance(logger, LoggerCollection):
-            self.loggers = list(logger)
         else:
             self.loggers = [logger]
 

tests/tests_pytorch/core/test_lightning_module.py

@@ -79,7 +79,7 @@ def test_property_logger(tmpdir):
     assert model.logger is None
 
     logger = TensorBoardLogger(tmpdir)
-    trainer = Mock(loggers=[logger])
+    trainer = Trainer(logger=logger)
     model.trainer = trainer
     assert model.logger == logger
 
@@ -94,6 +94,12 @@ def test_property_loggers(tmpdir):
     model.trainer = trainer
     assert model.loggers == [logger]
 
+    logger0 = TensorBoardLogger(tmpdir)
+    logger1 = TensorBoardLogger(tmpdir)
+    trainer = Trainer(logger=[logger0, logger1])
+    model.trainer = trainer
+    assert model.loggers == [logger0, logger1]
+
 
 def test_1_optimizer_toggle_model():
     """Test toggle_model runs when only one optimizer is used."""

tests/tests_pytorch/deprecated_api/test_remove_1-8.py

@@ -26,7 +26,7 @@
 from pytorch_lightning import Callback, Trainer
 from pytorch_lightning.callbacks import ModelCheckpoint
 from pytorch_lightning.demos.boring_classes import BoringDataModule, BoringModel
-from pytorch_lightning.loggers import CSVLogger, Logger, LoggerCollection
+from pytorch_lightning.loggers import CSVLogger, Logger
 from pytorch_lightning.plugins.precision.precision_plugin import PrecisionPlugin
 from pytorch_lightning.profiler import AbstractProfiler, BaseProfiler
 from pytorch_lightning.profilers import AdvancedProfiler, Profiler, SimpleProfiler
@@ -441,30 +441,6 @@ def _get_python_cprofile_total_duration(profile):
     np.testing.assert_allclose(recorded_total_duration, expected_total_duration, rtol=0.2)
 
 
-def test_v1_8_0_logger_collection(tmpdir):
-    logger1 = CSVLogger(tmpdir)
-    logger2 = CSVLogger(tmpdir)
-
-    trainer1 = Trainer(logger=logger1)
-    trainer2 = Trainer(logger=[logger1, logger2])
-
-    # Should have no deprecation warning
-    trainer1.logger
-    trainer1.loggers
-    trainer2.loggers
-
-    with pytest.deprecated_call(match="logger` will return the first logger"):
-        _ = trainer2.logger
-    with pytest.deprecated_call(match="`LoggerCollection` is deprecated in v1.6"):
-        _ = LoggerCollection([logger1, logger2])
-
-    model = BoringModel()
-    trainer = Trainer(logger=[logger1, logger2])
-    model.trainer = trainer
-    with pytest.deprecated_call(match="logger` will return the first logger"):
-        _ = model.logger
-
-
 def test_v1_8_0_precision_plugin_checkpoint_hooks(tmpdir):
     class PrecisionPluginSaveHook(PrecisionPlugin):
         def on_save_checkpoint(self, checkpoint):