move tensorboardX to extra

.github/workflows/ci-pkg-install.yml

@@ -99,4 +99,4 @@ jobs:
           pip install -q "pytest-doctestplus>=0.9.0"
           pip list
           PKG_NAME=$(python -c "print({'app': 'lit_app', 'fabric': 'lit_fabric', 'pytorch': 'pl'}.get('${{matrix.pkg-name}}', ''))")
-          python -m pytest src/${PKG_NAME} --ignore-glob="**/cli/*-template/**"
+          python -m pytest src/${PKG_NAME} --ignore-glob="**/cli/*-template/**" --doctest-plus

docs/source-pytorch/common/trainer.rst

@@ -817,10 +817,11 @@ logger
 :doc:`Logger <../visualize/loggers>` (or iterable collection of loggers) for experiment tracking. A ``True`` value uses the default ``TensorBoardLogger`` shown below. ``False`` will disable logging.
 
 .. testcode::
+    :skipif: not _TENSORBOARD_AVAILABLE and not _TENSORBOARDX_AVAILABLE
 
     from pytorch_lightning.loggers import TensorBoardLogger
 
-    # default logger used by trainer
+    # default logger used by trainer (if tensorboard is installed)
     logger = TensorBoardLogger(save_dir=os.getcwd(), version=1, name="lightning_logs")
     Trainer(logger=logger)
 

docs/source-pytorch/conf.py

@@ -400,6 +400,7 @@ def package_list_from_file(file):
 from pytorch_lightning.utilities import (
     _TORCHVISION_AVAILABLE,
 )
+from lightning_fabric.loggers.tensorboard import _TENSORBOARD_AVAILABLE, _TENSORBOARDX_AVAILABLE
 from pytorch_lightning.loggers.neptune import _NEPTUNE_AVAILABLE
 from pytorch_lightning.loggers.comet import _COMET_AVAILABLE
 from pytorch_lightning.loggers.mlflow import _MLFLOW_AVAILABLE

docs/source-pytorch/extensions/logging.rst

@@ -62,6 +62,7 @@ To visualize tensorboard in a jupyter notebook environment, run the following co
 You can also pass a custom Logger to the :class:`~pytorch_lightning.trainer.trainer.Trainer`.
 
 .. testcode::
+    :skipif: not _TENSORBOARD_AVAILABLE and not _TENSORBOARDX_AVAILABLE
 
     from pytorch_lightning import loggers as pl_loggers
 
@@ -79,7 +80,7 @@ Choose from any of the others such as MLflow, Comet, Neptune, WandB, etc.
 To use multiple loggers, simply pass in a ``list`` or ``tuple`` of loggers.
 
 .. testcode::
-    :skipif: not _COMET_AVAILABLE
+    :skipif: (not _TENSORBOARD_AVAILABLE and not _TENSORBOARDX_AVAILABLE) or not _COMET_AVAILABLE
 
     tb_logger = pl_loggers.TensorBoardLogger(save_dir="logs/")
     comet_logger = pl_loggers.CometLogger(save_dir="logs/")
@@ -378,7 +379,7 @@ When Lightning creates a checkpoint, it stores a key ``"hyper_parameters"`` with
 
 Some loggers also allow logging the hyperparams used in the experiment. For instance,
 when using the ``TensorBoardLogger``, all hyperparams will show
-in the `hparams tab <https://pytorch.org/docs/stable/tensorboard.html#torch.utils.tensorboard.writer.SummaryWriter.add_hparams>`_.
+in the hparams tab at :meth:`torch.utils.tensorboard.writer.SummaryWriter.add_hparams`.
 
 .. note::
     If you want to track a metric in the tensorboard hparams tab, log scalars to the key ``hp_metric``. If tracking multiple metrics, initialize ``TensorBoardLogger`` with ``default_hp_metric=False`` and call ``log_hyperparams`` only once with your metric keys and initial values. Subsequent updates can simply be logged to the metric keys. Refer to the examples below for setting up proper hyperparams metrics tracking within the :doc:`LightningModule <../common/lightning_module>`.

docs/source-pytorch/starter/introduction.rst

@@ -139,7 +139,7 @@ A LightningModule enables your PyTorch nn.Module to play together in complex way
             z = self.encoder(x)
             x_hat = self.decoder(z)
             loss = nn.functional.mse_loss(x_hat, x)
-            # Logging to TensorBoard by default
+            # Logging to TensorBoard (if installed) by default
             self.log("train_loss", loss)
             return loss
 
@@ -218,7 +218,7 @@ Once you've trained the model you can export to onnx, torchscript and put it int
 *********************
 6: Visualize training
 *********************
-Lightning comes with a *lot* of batteries included. A helpful one is Tensorboard for visualizing experiments.
+If you have tensorboard installed, you can use it for visualizing experiments.
 
 Run this on your commandline and open your browser to **http://localhost:6006/**
 

docs/source-pytorch/visualize/logging_basic.rst

@@ -58,13 +58,13 @@ TODO: need progress bar here
 
 View in the browser
 ===================
-To view metrics in the browser you need to use an *experiment manager* with these capabilities. By Default, Lightning uses Tensorboard which is free and opensource.
+To view metrics in the browser you need to use an *experiment manager* with these capabilities.
 
-Tensorboard is already enabled by default
+By Default, Lightning uses Tensorboard (if available) and a simple CSV logger otherwise.
 
 .. code-block:: python
 
-    # every trainer already has tensorboard enabled by default
+    # every trainer already has tensorboard enabled by default (if the dependency is available)
     trainer = Trainer()
 
 To launch the tensorboard dashboard run the following command on the commandline.

docs/source-pytorch/visualize/supported_exp_managers.rst

@@ -104,7 +104,7 @@ Here's the full documentation for the :class:`~pytorch_lightning.loggers.Neptune
 
 Tensorboard
 ===========
-`TensorBoard <https://pytorch.org/docs/stable/tensorboard.html>`_ already comes installed with Lightning. If you removed the install install the following package.
+`TensorBoard <https://pytorch.org/docs/stable/tensorboard.html>`_ can be installed with:
 
 .. code-block:: bash
 
@@ -179,7 +179,7 @@ Use multiple exp managers
 To use multiple experiment managers at the same time, pass a list to the *logger* :class:`~pytorch_lightning.trainer.trainer.Trainer` argument.
 
 .. testcode::
-    :skipif: not _WANDB_AVAILABLE
+    :skipif: (not _TENSORBOARD_AVAILABLE and not _TENSORBOARDX_AVAILABLE) or not _WANDB_AVAILABLE
 
     from pytorch_lightning.loggers import TensorBoardLogger, WandbLogger
 

requirements/pytorch/base.txt

@@ -6,7 +6,6 @@ torch>=1.10.0, <=1.13.1
 tqdm>=4.57.0, <4.65.0
 PyYAML>=5.4, <=6.0
 fsspec[http]>2021.06.0, <2022.8.0
-tensorboardX>=2.2, <=2.5.1  # min version is set by torch.onnx missing attribute
 torchmetrics>=0.7.0, <0.10.1  # needed for using fixed compare_version
 packaging>=17.0, <=21.3
 typing-extensions>=4.0.0, <=4.4.0

requirements/pytorch/extra.txt

@@ -7,3 +7,4 @@ omegaconf>=2.0.5, <2.3.0
 hydra-core>=1.0.5, <1.3.0
 jsonargparse[signatures]>=4.18.0, <4.19.0
 rich>=10.14.0, !=10.15.0.a, <13.0.0
+tensorboardX>=2.2, <=2.5.1  # min version is set by torch.onnx missing attribute

requirements/pytorch/loggers.info

@@ -3,3 +3,4 @@ neptune-client
 comet-ml
 mlflow>=1.0.0
 wandb
+tensorboard>=2.9.1

src/pytorch_lightning/CHANGELOG.md

@@ -20,7 +20,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Added support for colossalai 0.1.11 ([#15888](https://github.com/Lightning-AI/lightning/pull/15888))
 - Added `LightningCLI` support for optimizer and learning schedulers via callable type dependency injection ([#15869](https://github.com/Lightning-AI/lightning/pull/15869))
 - Added support for activation checkpointing for the `DDPFullyShardedNativeStrategy` strategy ([#15826](https://github.com/Lightning-AI/lightning/pull/15826))
-- Added the option to set `DDPFullyShardedNativeStrategy(cpu_offload=True|False)` via bool instead of needing to pass a configufation object ([#15832](https://github.com/Lightning-AI/lightning/pull/15832))
+- Added the option to set `DDPFullyShardedNativeStrategy(cpu_offload=True|False)` via bool instead of needing to pass a configuration object ([#15832](https://github.com/Lightning-AI/lightning/pull/15832))
 - Added info message for Ampere CUDA GPU users to enable tf32 matmul precision ([#16037](https://github.com/Lightning-AI/lightning/pull/16037))
 - Added support for returning optimizer-like classes in `LightningModule.configure_optimizers` ([#16189](https://github.com/Lightning-AI/lightning/pull/16189))
 
@@ -35,6 +35,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - The Trainer now raises an error if it is given multiple stateful callbacks of the same time with colliding state keys ([#15634](https://github.com/Lightning-AI/lightning/pull/15634))
 - `MLFlowLogger` now logs hyperparameters and metrics in batched API calls ([#15915](https://github.com/Lightning-AI/lightning/pull/15915))
 - Overriding the `on_train_batch_{start,end}` hooks in conjunction with taking a `dataloader_iter` in the `training_step` no longer errors out and instead shows a warning ([#16062](https://github.com/Lightning-AI/lightning/pull/16062))
+- Move `tensorboardX` to extra dependencies. Use the `CSVLogger` by default ([#16349](https://github.com/Lightning-AI/lightning/pull/16349))
 
 
 ### Deprecated

src/pytorch_lightning/loggers/tensorboard.py

@@ -24,7 +24,7 @@
 from torch import Tensor
 
 import pytorch_lightning as pl
-from lightning_fabric.loggers.tensorboard import _TENSORBOARD_AVAILABLE
+from lightning_fabric.loggers.tensorboard import _TENSORBOARD_AVAILABLE, _TENSORBOARDX_AVAILABLE
 from lightning_fabric.loggers.tensorboard import TensorBoardLogger as FabricTensorBoardLogger
 from lightning_fabric.utilities.logger import _convert_params
 from lightning_fabric.utilities.types import _PATH
@@ -39,6 +39,10 @@
 if _OMEGACONF_AVAILABLE:
     from omegaconf import Container, OmegaConf
 
+# Skip doctests if requirements aren't available
+if not (_TENSORBOARD_AVAILABLE or _TENSORBOARDX_AVAILABLE):
+    __doctest_skip__ = ["TensorBoardLogger", "TensorBoardLogger.*"]
+
 
 class TensorBoardLogger(Logger, FabricTensorBoardLogger):
     r"""

src/pytorch_lightning/trainer/connectors/logger_connector/logger_connector.py

@@ -14,15 +14,20 @@
 from typing import Any, Iterable, Optional, Union
 
 from lightning_utilities.core.apply_func import apply_to_collection
+from lightning_utilities.core.rank_zero import WarningCache
 from torch import Tensor
 
 import pytorch_lightning as pl
+from lightning_fabric.loggers import CSVLogger
+from lightning_fabric.loggers.tensorboard import _TENSORBOARD_AVAILABLE, _TENSORBOARDX_AVAILABLE
 from lightning_fabric.plugins.environments import SLURMEnvironment
 from lightning_fabric.utilities import move_data_to_device
 from lightning_fabric.utilities.apply_func import convert_tensors_to_scalars
 from pytorch_lightning.loggers import Logger, TensorBoardLogger
 from pytorch_lightning.trainer.connectors.logger_connector.result import _METRICS, _OUT_DICT, _PBAR_DICT
 
+warning_cache = WarningCache()
+
 
 class LoggerConnector:
     def __init__(self, trainer: "pl.Trainer") -> None:
@@ -57,9 +62,18 @@ def configure_logger(self, logger: Union[bool, Logger, Iterable[Logger]]) -> Non
             self.trainer.loggers = []
         elif logger is True:
             # default logger
-            self.trainer.loggers = [
-                TensorBoardLogger(save_dir=self.trainer.default_root_dir, version=SLURMEnvironment.job_id())
-            ]
+            if _TENSORBOARD_AVAILABLE or _TENSORBOARDX_AVAILABLE:
+                logger_ = TensorBoardLogger(save_dir=self.trainer.default_root_dir, version=SLURMEnvironment.job_id())
+            else:
+                warning_cache.warn(
+                    "Starting from v1.9.0, `tensorboardX` has been removed as a dependency of the `pytorch_lightning`"
+                    " package, due to potential conflicts with other packages in the ML ecosystem. For this reason,"
+                    " `logger=True` will use `CSVLogger` as the default logger, unless the `tensorboard`"
+                    " or `tensorboardX` packages are found."
+                    " Please `pip install lightning[extra]` or one of them to enable TensorBoard support by default"
+                )
+                logger_ = CSVLogger(root_dir=self.trainer.default_root_dir)  # type: ignore[assignment]
+            self.trainer.loggers = [logger_]
         elif isinstance(logger, Iterable):
             self.trainer.loggers = list(logger)
         else:

src/pytorch_lightning/trainer/trainer.py

@@ -286,9 +286,9 @@ def __init__(
                 Default: ``1.0``.
 
             logger: Logger (or iterable collection of loggers) for experiment tracking. A ``True`` value uses
-                the default ``TensorBoardLogger``. ``False`` will disable logging. If multiple loggers are
-                provided, local files (checkpoints, profiler traces, etc.) are saved in the ``log_dir`` of
-                the first logger.
+                the default ``TensorBoardLogger`` if it is installed, otherwise ``CSVLogger``.
+                ``False`` will disable logging. If multiple loggers are provided, local files
+                (checkpoints, profiler traces, etc.) are saved in the ``log_dir`` of he first logger.
                 Default: ``True``.
 
             log_every_n_steps: How often to log within steps.