Source code for pytorch_lightning.loggers.neptune
# Copyright The PyTorch Lightning team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
Neptune Logger
--------------
"""
__all__ = [
"NeptuneLogger",
]
import logging
import os
from argparse import Namespace
from typing import Any, Dict, Generator, List, Optional, Set, Union
from lightning_utilities.core.imports import RequirementCache
from torch import Tensor
import pytorch_lightning as pl
from pytorch_lightning.callbacks import Checkpoint
from pytorch_lightning.loggers.logger import Logger, rank_zero_experiment
from pytorch_lightning.utilities.logger import _add_prefix, _convert_params, _sanitize_callable_params
from pytorch_lightning.utilities.model_summary import ModelSummary
from pytorch_lightning.utilities.rank_zero import rank_zero_only
_NEPTUNE_AVAILABLE = RequirementCache("neptune-client")
if _NEPTUNE_AVAILABLE:
from neptune import new as neptune
from neptune.new.run import Run
else:
# needed for test mocks, and function signatures
neptune, Run = None, None
log = logging.getLogger(__name__)
_INTEGRATION_VERSION_KEY = "source_code/integrations/pytorch-lightning"
[docs]class NeptuneLogger(Logger):
r"""
Log using `Neptune <https://neptune.ai>`_.
Install it with pip:
.. code-block:: bash
pip install neptune-client
or conda:
.. code-block:: bash
conda install -c conda-forge neptune-client
**Quickstart**
Pass NeptuneLogger instance to the Trainer to log metadata with Neptune:
.. code-block:: python
from pytorch_lightning import Trainer
from pytorch_lightning.loggers import NeptuneLogger
neptune_logger = NeptuneLogger(
api_key="ANONYMOUS", # replace with your own
project="common/pytorch-lightning-integration", # format "<WORKSPACE/PROJECT>"
tags=["training", "resnet"], # optional
)
trainer = Trainer(max_epochs=10, logger=neptune_logger)
**How to use NeptuneLogger?**
Use the logger anywhere in your :class:`~pytorch_lightning.core.module.LightningModule` as follows:
.. code-block:: python
from neptune.new.types import File
from pytorch_lightning import LightningModule
class LitModel(LightningModule):
def training_step(self, batch, batch_idx):
# log metrics
acc = ...
self.log("train/loss", loss)
def any_lightning_module_function_or_hook(self):
# log images
img = ...
self.logger.experiment["train/misclassified_images"].log(File.as_image(img))
# generic recipe
metadata = ...
self.logger.experiment["your/metadata/structure"].log(metadata)
Note that syntax: ``self.logger.experiment["your/metadata/structure"].log(metadata)`` is specific to Neptune
and it extends logger capabilities. Specifically, it allows you to log various types of metadata
like scores, files, images, interactive visuals, CSVs, etc.
Refer to the `Neptune docs <https://docs.neptune.ai/you-should-know/logging-metadata#essential-logging-methods>`_
for more detailed explanations.
You can also use regular logger methods ``log_metrics()``, and ``log_hyperparams()`` with NeptuneLogger
as these are also supported.
**Log after fitting or testing is finished**
You can log objects after the fitting or testing methods are finished:
.. code-block:: python
neptune_logger = NeptuneLogger(project="common/pytorch-lightning-integration")
trainer = pl.Trainer(logger=neptune_logger)
model = ...
datamodule = ...
trainer.fit(model, datamodule=datamodule)
trainer.test(model, datamodule=datamodule)
# Log objects after `fit` or `test` methods
# model summary
neptune_logger.log_model_summary(model=model, max_depth=-1)
# generic recipe
metadata = ...
neptune_logger.experiment["your/metadata/structure"].log(metadata)
**Log model checkpoints**
If you have :class:`~pytorch_lightning.callbacks.ModelCheckpoint` configured,
Neptune logger automatically logs model checkpoints.
Model weights will be uploaded to the: "model/checkpoints" namespace in the Neptune Run.
You can disable this option:
.. code-block:: python
neptune_logger = NeptuneLogger(project="common/pytorch-lightning-integration", log_model_checkpoints=False)
**Pass additional parameters to the Neptune run**
You can also pass ``neptune_run_kwargs`` to specify the run in the greater detail, like ``tags`` or ``description``:
.. testcode::
:skipif: not _NEPTUNE_AVAILABLE
from pytorch_lightning import Trainer
from pytorch_lightning.loggers import NeptuneLogger
neptune_logger = NeptuneLogger(
project="common/pytorch-lightning-integration",
name="lightning-run",
description="mlp quick run with pytorch-lightning",
tags=["mlp", "quick-run"],
)
trainer = Trainer(max_epochs=3, logger=neptune_logger)
Check `run documentation <https://docs.neptune.ai/essentials/api-reference/run>`_
for more info about additional run parameters.
**Details about Neptune run structure**
Runs can be viewed as nested dictionary-like structures that you can define in your code.
Thanks to this you can easily organize your metadata in a way that is most convenient for you.
The hierarchical structure that you apply to your metadata will be reflected later in the UI.
You can organize this way any type of metadata - images, parameters, metrics, model checkpoint, CSV files, etc.
See Also:
- Read about
`what object you can log to Neptune <https://docs.neptune.ai/you-should-know/what-can-you-log-and-display>`_.
- Check `example run <https://app.neptune.ai/o/common/org/pytorch-lightning-integration/e/PTL-1/all>`_
with multiple types of metadata logged.
- For more detailed info check
`user guide <https://docs.neptune.ai/integrations-and-supported-tools/model-training/pytorch-lightning>`_.
Args:
api_key: Optional.
Neptune API token, found on https://neptune.ai upon registration.
Read: `how to find and set Neptune API token <https://docs.neptune.ai/administration/security-and-privacy/
how-to-find-and-set-neptune-api-token>`_.
It is recommended to keep it in the `NEPTUNE_API_TOKEN`
environment variable and then you can drop ``api_key=None``.
project: Optional.
Name of a project in a form of "my_workspace/my_project" for example "tom/mask-rcnn".
If ``None``, the value of `NEPTUNE_PROJECT` environment variable will be taken.
You need to create the project in https://neptune.ai first.
name: Optional. Editable name of the run.
Run name appears in the "all metadata/sys" section in Neptune UI.
run: Optional. Default is ``None``. The Neptune ``Run`` object.
If specified, this `Run`` will be used for logging, instead of a new Run.
When run object is passed you can't specify other neptune properties.
log_model_checkpoints: Optional. Default is ``True``. Log model checkpoint to Neptune.
Works only if ``ModelCheckpoint`` is passed to the ``Trainer``.
prefix: Optional. Default is ``"training"``. Root namespace for all metadata logging.
\**neptune_run_kwargs: Additional arguments like ``tags``, ``description``, ``capture_stdout``, etc.
used when run is created.
Raises:
ModuleNotFoundError:
If required Neptune package is not installed.
ValueError:
If argument passed to the logger's constructor is incorrect.
"""
LOGGER_JOIN_CHAR = "/"
PARAMETERS_KEY = "hyperparams"
ARTIFACTS_KEY = "artifacts"
def __init__(
self,
*, # force users to call `NeptuneLogger` initializer with `kwargs`
api_key: Optional[str] = None,
project: Optional[str] = None,
name: Optional[str] = None,
run: Optional["Run"] = None,
log_model_checkpoints: Optional[bool] = True,
prefix: str = "training",
**neptune_run_kwargs: Any,
):
if not _NEPTUNE_AVAILABLE:
raise ModuleNotFoundError(str(_NEPTUNE_AVAILABLE))
# verify if user passed proper init arguments
self._verify_input_arguments(api_key, project, name, run, neptune_run_kwargs)
super().__init__()
self._log_model_checkpoints = log_model_checkpoints
self._prefix = prefix
self._run_name = name
self._project_name = project
self._api_key = api_key
self._run_instance = run
self._neptune_run_kwargs = neptune_run_kwargs
self._run_short_id: Optional[str] = None
if self._run_instance is not None:
self._retrieve_run_data()
# make sure that we've log integration version for outside `Run` instances
self._run_instance[_INTEGRATION_VERSION_KEY] = pl.__version__
def _retrieve_run_data(self) -> None:
assert self._run_instance is not None
self._run_instance.wait()
if self._run_instance.exists("sys/id"):
self._run_short_id = self._run_instance["sys/id"].fetch()
self._run_name = self._run_instance["sys/name"].fetch()
else:
self._run_short_id = "OFFLINE"
self._run_name = "offline-name"
@property
def _neptune_init_args(self) -> Dict:
args: Dict = {}
# Backward compatibility in case of previous version retrieval
try:
args = self._neptune_run_kwargs
except AttributeError:
pass
if self._project_name is not None:
args["project"] = self._project_name
if self._api_key is not None:
args["api_token"] = self._api_key
if self._run_short_id is not None:
args["run"] = self._run_short_id
# Backward compatibility in case of previous version retrieval
try:
if self._run_name is not None:
args["name"] = self._run_name
except AttributeError:
pass
return args
def _construct_path_with_prefix(self, *keys: str) -> str:
"""Return sequence of keys joined by `LOGGER_JOIN_CHAR`, started with `_prefix` if defined."""
if self._prefix:
return self.LOGGER_JOIN_CHAR.join([self._prefix, *keys])
return self.LOGGER_JOIN_CHAR.join(keys)
@staticmethod
def _verify_input_arguments(
api_key: Optional[str],
project: Optional[str],
name: Optional[str],
run: Optional["Run"],
neptune_run_kwargs: dict,
) -> None:
# check if user passed the client `Run` object
if run is not None and not isinstance(run, Run):
raise ValueError("Run parameter expected to be of type `neptune.new.Run`.")
# check if user passed redundant neptune.init_run arguments when passed run
any_neptune_init_arg_passed = any(arg is not None for arg in [api_key, project, name]) or neptune_run_kwargs
if run is not None and any_neptune_init_arg_passed:
raise ValueError(
"When an already initialized run object is provided"
" you can't provide other neptune.init_run() parameters.\n"
)
def __getstate__(self) -> Dict[str, Any]:
state = self.__dict__.copy()
# Run instance can't be pickled
state["_run_instance"] = None
return state
def __setstate__(self, state: Dict[str, Any]) -> None:
self.__dict__ = state
self._run_instance = neptune.init_run(**self._neptune_init_args)
@property
@rank_zero_experiment
def experiment(self) -> Run:
r"""
Actual Neptune run object. Allows you to use neptune logging features in your
:class:`~pytorch_lightning.core.module.LightningModule`.
Example::
class LitModel(LightningModule):
def training_step(self, batch, batch_idx):
# log metrics
acc = ...
self.logger.experiment["train/acc"].log(acc)
# log images
img = ...
self.logger.experiment["train/misclassified_images"].log(File.as_image(img))
Note that syntax: ``self.logger.experiment["your/metadata/structure"].log(metadata)``
is specific to Neptune and it extends logger capabilities.
Specifically, it allows you to log various types of metadata like scores, files,
images, interactive visuals, CSVs, etc. Refer to the
`Neptune docs <https://docs.neptune.ai/you-should-know/logging-metadata#essential-logging-methods>`_
for more detailed explanations.
You can also use regular logger methods ``log_metrics()``, and ``log_hyperparams()``
with NeptuneLogger as these are also supported.
"""
return self.run
@property
@rank_zero_experiment
def run(self) -> Run:
if not self._run_instance:
self._run_instance = neptune.init_run(**self._neptune_init_args)
self._retrieve_run_data()
# make sure that we've log integration version for newly created
self._run_instance[_INTEGRATION_VERSION_KEY] = pl.__version__
return self._run_instance
[docs] @rank_zero_only
def log_hyperparams(self, params: Union[Dict[str, Any], Namespace]) -> None: # skipcq: PYL-W0221
r"""
Log hyper-parameters to the run.
Hyperparams will be logged under the "<prefix>/hyperparams" namespace.
Note:
You can also log parameters by directly using the logger instance:
``neptune_logger.experiment["model/hyper-parameters"] = params_dict``.
In this way you can keep hierarchical structure of the parameters.
Args:
params: `dict`.
Python dictionary structure with parameters.
Example::
from pytorch_lightning.loggers import NeptuneLogger
PARAMS = {
"batch_size": 64,
"lr": 0.07,
"decay_factor": 0.97
}
neptune_logger = NeptuneLogger(
api_key="ANONYMOUS",
project="common/pytorch-lightning-integration"
)
neptune_logger.log_hyperparams(PARAMS)
"""
params = _convert_params(params)
params = _sanitize_callable_params(params)
parameters_key = self.PARAMETERS_KEY
parameters_key = self._construct_path_with_prefix(parameters_key)
self.run[parameters_key] = params
[docs] @rank_zero_only
def log_metrics(self, metrics: Dict[str, Union[Tensor, float]], step: Optional[int] = None) -> None:
"""Log metrics (numeric values) in Neptune runs.
Args:
metrics: Dictionary with metric names as keys and measured quantities as values.
step: Step number at which the metrics should be recorded, currently ignored.
"""
if rank_zero_only.rank != 0:
raise ValueError("run tried to log from global_rank != 0")
metrics = _add_prefix(metrics, self._prefix, self.LOGGER_JOIN_CHAR)
for key, val in metrics.items():
# `step` is ignored because Neptune expects strictly increasing step values which
# Lightning does not always guarantee.
self.run[key].log(val)
[docs] @rank_zero_only
def finalize(self, status: str) -> None:
if not self._run_instance:
# When using multiprocessing, finalize() should be a no-op on the main process, as no experiment has been
# initialized there
return
if status:
self.run[self._construct_path_with_prefix("status")] = status
super().finalize(status)
@property
def save_dir(self) -> Optional[str]:
"""Gets the save directory of the experiment which in this case is ``None`` because Neptune does not save
locally.
Returns:
the root directory where experiment logs get saved
"""
return os.path.join(os.getcwd(), ".neptune")
@rank_zero_only
def log_model_summary(self, model: "pl.LightningModule", max_depth: int = -1) -> None:
model_str = str(ModelSummary(model=model, max_depth=max_depth))
self.run[self._construct_path_with_prefix("model/summary")] = neptune.types.File.from_content(
content=model_str, extension="txt"
)
[docs] @rank_zero_only
def after_save_checkpoint(self, checkpoint_callback: Checkpoint) -> None:
"""Automatically log checkpointed model. Called after model checkpoint callback saves a new checkpoint.
Args:
checkpoint_callback: the model checkpoint callback instance
"""
if not self._log_model_checkpoints:
return
file_names = set()
checkpoints_namespace = self._construct_path_with_prefix("model/checkpoints")
# save last model
if hasattr(checkpoint_callback, "last_model_path") and checkpoint_callback.last_model_path:
model_last_name = self._get_full_model_name(checkpoint_callback.last_model_path, checkpoint_callback)
file_names.add(model_last_name)
self.run[f"{checkpoints_namespace}/{model_last_name}"].upload(checkpoint_callback.last_model_path)
# save best k models
if hasattr(checkpoint_callback, "best_k_models"):
for key in checkpoint_callback.best_k_models.keys():
model_name = self._get_full_model_name(key, checkpoint_callback)
file_names.add(model_name)
self.run[f"{checkpoints_namespace}/{model_name}"].upload(key)
# log best model path and checkpoint
if hasattr(checkpoint_callback, "best_model_path") and checkpoint_callback.best_model_path:
self.run[self._construct_path_with_prefix("model/best_model_path")] = checkpoint_callback.best_model_path
model_name = self._get_full_model_name(checkpoint_callback.best_model_path, checkpoint_callback)
file_names.add(model_name)
self.run[f"{checkpoints_namespace}/{model_name}"].upload(checkpoint_callback.best_model_path)
# remove old models logged to experiment if they are not part of best k models at this point
if self.run.exists(checkpoints_namespace):
exp_structure = self.run.get_structure()
uploaded_model_names = self._get_full_model_names_from_exp_structure(exp_structure, checkpoints_namespace)
for file_to_drop in list(uploaded_model_names - file_names):
del self.run[f"{checkpoints_namespace}/{file_to_drop}"]
# log best model score
if hasattr(checkpoint_callback, "best_model_score") and checkpoint_callback.best_model_score:
self.run[self._construct_path_with_prefix("model/best_model_score")] = (
checkpoint_callback.best_model_score.cpu().detach().numpy()
)
@staticmethod
def _get_full_model_name(model_path: str, checkpoint_callback: Checkpoint) -> str:
"""Returns model name which is string `model_path` appended to `checkpoint_callback.dirpath`."""
if hasattr(checkpoint_callback, "dirpath"):
expected_model_path = f"{checkpoint_callback.dirpath}{os.path.sep}"
if not model_path.startswith(expected_model_path):
raise ValueError(f"{model_path} was expected to start with {expected_model_path}.")
# Remove extension from filepath
filepath, _ = os.path.splitext(model_path[len(expected_model_path) :])
else:
filepath = model_path
return filepath
@classmethod
def _get_full_model_names_from_exp_structure(cls, exp_structure: Dict[str, Any], namespace: str) -> Set[str]:
"""Returns all paths to properties which were already logged in `namespace`"""
structure_keys: List[str] = namespace.split(cls.LOGGER_JOIN_CHAR)
for key in structure_keys:
exp_structure = exp_structure[key]
uploaded_models_dict = exp_structure
return set(cls._dict_paths(uploaded_models_dict))
@classmethod
def _dict_paths(cls, d: Dict[str, Any], path_in_build: str = None) -> Generator:
for k, v in d.items():
path = f"{path_in_build}/{k}" if path_in_build is not None else k
if not isinstance(v, dict):
yield path
else:
yield from cls._dict_paths(v, path)
@property
def name(self) -> Optional[str]:
"""Return the experiment name or 'offline-name' when exp is run in offline mode."""
return self._run_name
@property
def version(self) -> Optional[str]:
"""Return the experiment version.
It's Neptune Run's short_id
"""
return self._run_short_id
