File size: 13,049 Bytes
9c6594c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 |
# Copyright The Lightning AI 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.
import os
from argparse import Namespace
from collections.abc import Mapping
from typing import TYPE_CHECKING, Any, Optional, Union
from lightning_utilities.core.imports import RequirementCache
from torch import Tensor
from torch.nn import Module
from typing_extensions import override
from lightning_fabric.loggers.logger import Logger, rank_zero_experiment
from lightning_fabric.utilities.cloud_io import _is_dir, get_filesystem
from lightning_fabric.utilities.logger import _add_prefix, _convert_params, _flatten_dict
from lightning_fabric.utilities.logger import _sanitize_params as _utils_sanitize_params
from lightning_fabric.utilities.rank_zero import rank_zero_only, rank_zero_warn
from lightning_fabric.utilities.types import _PATH
from lightning_fabric.wrappers import _unwrap_objects
_TENSORBOARD_AVAILABLE = RequirementCache("tensorboard")
_TENSORBOARDX_AVAILABLE = RequirementCache("tensorboardX")
if TYPE_CHECKING:
# assumes at least one will be installed when type checking
if _TENSORBOARD_AVAILABLE:
from torch.utils.tensorboard import SummaryWriter
else:
from tensorboardX import SummaryWriter # type: ignore[no-redef]
class TensorBoardLogger(Logger):
r"""Log to local file system in `TensorBoard <https://www.tensorflow.org/tensorboard>`_ format.
Implemented using :class:`~tensorboardX.SummaryWriter`. Logs are saved to
``os.path.join(root_dir, name, version)``. This is the recommended logger in Lightning Fabric.
Args:
root_dir: The root directory in which all your experiments with different names and versions will be stored.
name: Experiment name. Defaults to ``'lightning_logs'``. If it is the empty string then no per-experiment
subdirectory is used.
version: Experiment version. If version is not specified the logger inspects the save
directory for existing versions, then automatically assigns the next available version.
If it is a string then it is used as the run-specific subdirectory name,
otherwise ``'version_${version}'`` is used.
default_hp_metric: Enables a placeholder metric with key `hp_metric` when `log_hyperparams` is
called without a metric (otherwise calls to ``log_hyperparams`` without a metric are ignored).
prefix: A string to put at the beginning of all metric keys.
sub_dir: Sub-directory to group TensorBoard logs. If a ``sub_dir`` argument is passed
then logs are saved in ``/root_dir/name/version/sub_dir/``. Defaults to ``None`` in which case
logs are saved in ``/root_dir/name/version/``.
\**kwargs: Additional arguments used by :class:`tensorboardX.SummaryWriter` can be passed as keyword
arguments in this logger. To automatically flush to disk, `max_queue` sets the size
of the queue for pending logs before flushing. `flush_secs` determines how many seconds
elapses before flushing.
Example::
from lightning_fabric.loggers import TensorBoardLogger
logger = TensorBoardLogger("path/to/logs/root", name="my_model")
logger.log_hyperparams({"epochs": 5, "optimizer": "Adam"})
logger.log_metrics({"acc": 0.75})
logger.finalize("success")
"""
LOGGER_JOIN_CHAR = "-"
def __init__(
self,
root_dir: _PATH,
name: Optional[str] = "lightning_logs",
version: Optional[Union[int, str]] = None,
default_hp_metric: bool = True,
prefix: str = "",
sub_dir: Optional[_PATH] = None,
**kwargs: Any,
):
if not _TENSORBOARD_AVAILABLE and not _TENSORBOARDX_AVAILABLE:
raise ModuleNotFoundError(
"Neither `tensorboard` nor `tensorboardX` is available. Try `pip install`ing either.\n"
f"{str(_TENSORBOARDX_AVAILABLE)}\n{str(_TENSORBOARD_AVAILABLE)}"
)
super().__init__()
root_dir = os.fspath(root_dir)
self._root_dir = root_dir
self._name = name or ""
self._version = version
self._sub_dir = None if sub_dir is None else os.fspath(sub_dir)
self._default_hp_metric = default_hp_metric
self._prefix = prefix
self._fs = get_filesystem(root_dir)
self._experiment: Optional[SummaryWriter] = None
self._kwargs = kwargs
@property
@override
def name(self) -> str:
"""Get the name of the experiment.
Returns:
The name of the experiment.
"""
return self._name
@property
@override
def version(self) -> Union[int, str]:
"""Get the experiment version.
Returns:
The experiment version if specified else the next version.
"""
if self._version is None:
self._version = self._get_next_version()
return self._version
@property
@override
def root_dir(self) -> str:
"""Gets the save directory where the TensorBoard experiments are saved.
Returns:
The local path to the save directory where the TensorBoard experiments are saved.
"""
return self._root_dir
@property
@override
def log_dir(self) -> str:
"""The directory for this run's tensorboard checkpoint.
By default, it is named ``'version_${self.version}'`` but it can be overridden by passing a string value for the
constructor's version parameter instead of ``None`` or an int.
"""
version = self.version if isinstance(self.version, str) else f"version_{self.version}"
log_dir = os.path.join(self.root_dir, self.name, version)
if isinstance(self.sub_dir, str):
log_dir = os.path.join(log_dir, self.sub_dir)
log_dir = os.path.expandvars(log_dir)
log_dir = os.path.expanduser(log_dir)
return log_dir
@property
def sub_dir(self) -> Optional[str]:
"""Gets the sub directory where the TensorBoard experiments are saved.
Returns:
The local path to the sub directory where the TensorBoard experiments are saved.
"""
return self._sub_dir
@property
@rank_zero_experiment
def experiment(self) -> "SummaryWriter":
"""Actual tensorboard object. To use TensorBoard features anywhere in your code, do the following.
Example::
logger.experiment.some_tensorboard_function()
"""
if self._experiment is not None:
return self._experiment
assert rank_zero_only.rank == 0, "tried to init log dirs in non global_rank=0"
if self.root_dir:
self._fs.makedirs(self.root_dir, exist_ok=True)
if _TENSORBOARD_AVAILABLE:
from torch.utils.tensorboard import SummaryWriter
else:
from tensorboardX import SummaryWriter # type: ignore[no-redef]
self._experiment = SummaryWriter(log_dir=self.log_dir, **self._kwargs)
return self._experiment
@override
@rank_zero_only
def log_metrics(self, metrics: Mapping[str, float], step: Optional[int] = None) -> None:
assert rank_zero_only.rank == 0, "experiment tried to log from global_rank != 0"
metrics = _add_prefix(metrics, self._prefix, self.LOGGER_JOIN_CHAR)
for k, v in metrics.items():
if isinstance(v, Tensor):
v = v.item()
if isinstance(v, dict):
self.experiment.add_scalars(k, v, step)
else:
try:
self.experiment.add_scalar(k, v, step)
# TODO(fabric): specify the possible exception
except Exception as ex:
raise ValueError(
f"\n you tried to log {v} which is currently not supported. Try a dict or a scalar/tensor."
) from ex
@override
@rank_zero_only
def log_hyperparams(
self,
params: Union[dict[str, Any], Namespace],
metrics: Optional[dict[str, Any]] = None,
step: Optional[int] = None,
) -> None:
"""Record hyperparameters. TensorBoard logs with and without saved hyperparameters are incompatible, the
hyperparameters are then not displayed in the TensorBoard. Please delete or move the previously saved logs to
display the new ones with hyperparameters.
Args:
params: A dictionary-like container with the hyperparameters
metrics: Dictionary with metric names as keys and measured quantities as values
step: Optional global step number for the logged metrics
"""
params = _convert_params(params)
# format params into the suitable for tensorboard
params = _flatten_dict(params)
params = self._sanitize_params(params)
if metrics is None:
if self._default_hp_metric:
metrics = {"hp_metric": -1}
elif not isinstance(metrics, dict):
metrics = {"hp_metric": metrics}
if metrics:
self.log_metrics(metrics, step)
if _TENSORBOARD_AVAILABLE:
from torch.utils.tensorboard.summary import hparams
else:
from tensorboardX.summary import hparams # type: ignore[no-redef]
exp, ssi, sei = hparams(params, metrics)
writer = self.experiment._get_file_writer()
writer.add_summary(exp, step)
writer.add_summary(ssi, step)
writer.add_summary(sei, step)
@override
@rank_zero_only
def log_graph(self, model: Module, input_array: Optional[Tensor] = None) -> None:
model_example_input = getattr(model, "example_input_array", None)
input_array = model_example_input if input_array is None else input_array
model = _unwrap_objects(model)
if input_array is None:
rank_zero_warn(
"Could not log computational graph to TensorBoard: The `model.example_input_array` attribute"
" is not set or `input_array` was not given."
)
elif not isinstance(input_array, (Tensor, tuple)):
rank_zero_warn(
"Could not log computational graph to TensorBoard: The `input_array` or `model.example_input_array`"
f" has type {type(input_array)} which can't be traced by TensorBoard. Make the input array a tuple"
f" representing the positional arguments to the model's `forward()` implementation."
)
elif callable(getattr(model, "_on_before_batch_transfer", None)) and callable(
getattr(model, "_apply_batch_transfer_handler", None)
):
# this is probably is a LightningModule
input_array = model._on_before_batch_transfer(input_array)
input_array = model._apply_batch_transfer_handler(input_array)
self.experiment.add_graph(model, input_array)
else:
self.experiment.add_graph(model, input_array)
@override
@rank_zero_only
def save(self) -> None:
self.experiment.flush()
@override
@rank_zero_only
def finalize(self, status: str) -> None:
if self._experiment is not None:
self.experiment.flush()
self.experiment.close()
def _get_next_version(self) -> int:
save_dir = os.path.join(self.root_dir, self.name)
try:
listdir_info = self._fs.listdir(save_dir)
except OSError:
return 0
existing_versions = []
for listing in listdir_info:
d = listing["name"]
bn = os.path.basename(d)
if _is_dir(self._fs, d) and bn.startswith("version_"):
dir_ver = bn.split("_")[1].replace("/", "")
if dir_ver.isdigit():
existing_versions.append(int(dir_ver))
if len(existing_versions) == 0:
return 0
return max(existing_versions) + 1
@staticmethod
def _sanitize_params(params: dict[str, Any]) -> dict[str, Any]:
params = _utils_sanitize_params(params)
# logging of arrays with dimension > 1 is not supported, sanitize as string
return {k: str(v) if hasattr(v, "ndim") and v.ndim > 1 else v for k, v in params.items()}
def __getstate__(self) -> dict[str, Any]:
state = self.__dict__.copy()
state["_experiment"] = None
return state
|