"""TensorBoard logger and its helper handlers."""
from typing import Any, Callable, List, Optional, Union
from torch.optim import Optimizer
from ignite.engine import Engine, Events
from ignite.handlers.base_logger import (
BaseLogger,
BaseOptimizerParamsHandler,
BaseOutputHandler,
BaseWeightsHandler,
BaseWeightsScalarHandler,
)
from ignite.handlers.utils import global_step_from_engine # noqa
__all__ = [
"TensorboardLogger",
"OptimizerParamsHandler",
"OutputHandler",
"WeightsScalarHandler",
"WeightsHistHandler",
"GradsScalarHandler",
"GradsHistHandler",
"global_step_from_engine",
]
[docs]class TensorboardLogger(BaseLogger):
"""
TensorBoard handler to log metrics, model/optimizer parameters, gradients during the training and validation.
By default, this class favors `tensorboardX <https://github.com/lanpa/tensorboardX>`_ package if installed:
.. code-block:: bash
pip install tensorboardX
otherwise, it falls back to using
`PyTorch's SummaryWriter
<https://pytorch.org/docs/stable/tensorboard.html>`_
(>=v1.2.0).
Args:
args: Positional arguments accepted from
`SummaryWriter
<https://pytorch.org/docs/stable/tensorboard.html>`_.
kwargs: Keyword arguments accepted from
`SummaryWriter
<https://pytorch.org/docs/stable/tensorboard.html>`_.
For example, `log_dir` to setup path to the directory where to log.
Examples:
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
# Create a logger
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Attach the logger to the trainer to log training loss at each iteration
tb_logger.attach_output_handler(
trainer,
event_name=Events.ITERATION_COMPLETED,
tag="training",
output_transform=lambda loss: {"loss": loss}
)
# Attach the logger to the evaluator on the training dataset and log NLL, Accuracy metrics after each epoch
# We setup `global_step_transform=global_step_from_engine(trainer)` to take the epoch
# of the `trainer` instead of `train_evaluator`.
tb_logger.attach_output_handler(
train_evaluator,
event_name=Events.EPOCH_COMPLETED,
tag="training",
metric_names=["nll", "accuracy"],
global_step_transform=global_step_from_engine(trainer),
)
# Attach the logger to the evaluator on the validation dataset and log NLL, Accuracy metrics after
# each epoch. We setup `global_step_transform=global_step_from_engine(trainer)` to take the epoch of the
# `trainer` instead of `evaluator`.
tb_logger.attach_output_handler(
evaluator,
event_name=Events.EPOCH_COMPLETED,
tag="validation",
metric_names=["nll", "accuracy"],
global_step_transform=global_step_from_engine(trainer)),
)
# Attach the logger to the trainer to log optimizer's parameters, e.g. learning rate at each iteration
tb_logger.attach_opt_params_handler(
trainer,
event_name=Events.ITERATION_STARTED,
optimizer=optimizer,
param_name='lr' # optional
)
# Attach the logger to the trainer to log model's weights norm after each iteration
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=WeightsScalarHandler(model)
)
# Attach the logger to the trainer to log model's weights as a histogram after each epoch
tb_logger.attach(
trainer,
event_name=Events.EPOCH_COMPLETED,
log_handler=WeightsHistHandler(model)
)
# Attach the logger to the trainer to log model's gradients norm after each iteration
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=GradsScalarHandler(model)
)
# Attach the logger to the trainer to log model's gradients as a histogram after each epoch
tb_logger.attach(
trainer,
event_name=Events.EPOCH_COMPLETED,
log_handler=GradsHistHandler(model)
)
# We need to close the logger when we are done
tb_logger.close()
It is also possible to use the logger as context manager:
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
with TensorboardLogger(log_dir="experiments/tb_logs") as tb_logger:
trainer = Engine(update_fn)
# Attach the logger to the trainer to log training loss at each iteration
tb_logger.attach_output_handler(
trainer,
event_name=Events.ITERATION_COMPLETED,
tag="training",
output_transform=lambda loss: {"loss": loss}
)
"""
def __init__(self, *args: Any, **kwargs: Any):
try:
from tensorboardX import SummaryWriter
except ImportError:
try:
from torch.utils.tensorboard import SummaryWriter
except ImportError:
raise ModuleNotFoundError(
"This contrib module requires either tensorboardX or torch >= 1.2.0. "
"You may install tensorboardX with command: \n pip install tensorboardX \n"
"or upgrade PyTorch using your package manager of choice (pip or conda)."
)
self.writer = SummaryWriter(*args, **kwargs)
def __getattr__(self, attr: Any) -> Any:
return getattr(self.writer, attr)
def close(self) -> None:
self.writer.close()
def _create_output_handler(self, *args: Any, **kwargs: Any) -> "OutputHandler":
return OutputHandler(*args, **kwargs)
def _create_opt_params_handler(self, *args: Any, **kwargs: Any) -> "OptimizerParamsHandler":
return OptimizerParamsHandler(*args, **kwargs)
[docs]class OutputHandler(BaseOutputHandler):
"""Helper handler to log engine's output, engine's state attributes and/or metrics
Args:
tag: common title for all produced plots. For example, "training"
metric_names: list of metric names to plot or a string "all" to plot all available
metrics.
output_transform: output transform function to prepare `engine.state.output` as a number.
For example, `output_transform = lambda output: output`
This function can also return a dictionary, e.g `{"loss": loss1, "another_loss": loss2}` to label the plot
with corresponding keys.
global_step_transform: global step transform function to output a desired global step.
Input of the function is `(engine, event_name)`. Output of function should be an integer.
Default is None, global_step based on attached engine. If provided,
uses function output as global_step. To setup global step from another engine, please use
:meth:`~ignite.handlers.tensorboard_logger.global_step_from_engine`.
state_attributes: list of attributes of the ``trainer.state`` to plot.
Examples:
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
# Create a logger
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Attach the logger to the evaluator on the validation dataset and log NLL, Accuracy metrics after
# each epoch. We setup `global_step_transform=global_step_from_engine(trainer)` to take the epoch
# of the `trainer`:
tb_logger.attach(
evaluator,
log_handler=OutputHandler(
tag="validation",
metric_names=["nll", "accuracy"],
global_step_transform=global_step_from_engine(trainer)
),
event_name=Events.EPOCH_COMPLETED
)
# or equivalently
tb_logger.attach_output_handler(
evaluator,
event_name=Events.EPOCH_COMPLETED,
tag="validation",
metric_names=["nll", "accuracy"],
global_step_transform=global_step_from_engine(trainer)
)
Another example, where model is evaluated every 500 iterations:
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
@trainer.on(Events.ITERATION_COMPLETED(every=500))
def evaluate(engine):
evaluator.run(validation_set, max_epochs=1)
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
def global_step_transform(*args, **kwargs):
return trainer.state.iteration
# Attach the logger to the evaluator on the validation dataset and log NLL, Accuracy metrics after
# every 500 iterations. Since evaluator engine does not have access to the training iteration, we
# provide a global_step_transform to return the trainer.state.iteration for the global_step, each time
# evaluator metrics are plotted on Tensorboard.
tb_logger.attach_output_handler(
evaluator,
event_name=Events.EPOCH_COMPLETED,
tag="validation",
metrics=["nll", "accuracy"],
global_step_transform=global_step_transform
)
Another example where the State Attributes ``trainer.state.alpha`` and ``trainer.state.beta``
are also logged along with the NLL and Accuracy after each iteration:
.. code-block:: python
tb_logger.attach(
trainer,
log_handler=OutputHandler(
tag="training",
metric_names=["nll", "accuracy"],
state_attributes=["alpha", "beta"],
),
event_name=Events.ITERATION_COMPLETED
)
Example of `global_step_transform`:
.. code-block:: python
def global_step_transform(engine, event_name):
return engine.state.get_event_attrib_value(event_name)
.. versionchanged:: 0.4.7
accepts an optional list of `state_attributes`
"""
def __init__(
self,
tag: str,
metric_names: Optional[List[str]] = None,
output_transform: Optional[Callable] = None,
global_step_transform: Optional[Callable[[Engine, Union[str, Events]], int]] = None,
state_attributes: Optional[List[str]] = None,
):
super(OutputHandler, self).__init__(
tag, metric_names, output_transform, global_step_transform, state_attributes
)
def __call__(self, engine: Engine, logger: TensorboardLogger, event_name: Union[str, Events]) -> None:
if not isinstance(logger, TensorboardLogger):
raise RuntimeError("Handler 'OutputHandler' works only with TensorboardLogger")
metrics = self._setup_output_metrics_state_attrs(engine, key_tuple=False)
global_step = self.global_step_transform(engine, event_name)
if not isinstance(global_step, int):
raise TypeError(
f"global_step must be int, got {type(global_step)}."
" Please check the output of global_step_transform."
)
for key, value in metrics.items():
logger.writer.add_scalar(key, value, global_step)
[docs]class OptimizerParamsHandler(BaseOptimizerParamsHandler):
"""Helper handler to log optimizer parameters
Args:
optimizer: torch optimizer or any object with attribute ``param_groups``
as a sequence.
param_name: parameter name
tag: common title for all produced plots. For example, "generator"
Examples:
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
# Create a logger
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Attach the logger to the trainer to log optimizer's parameters, e.g. learning rate at each iteration
tb_logger.attach(
trainer,
log_handler=OptimizerParamsHandler(optimizer),
event_name=Events.ITERATION_STARTED
)
# or equivalently
tb_logger.attach_opt_params_handler(
trainer,
event_name=Events.ITERATION_STARTED,
optimizer=optimizer
)
"""
def __init__(self, optimizer: Optimizer, param_name: str = "lr", tag: Optional[str] = None):
super(OptimizerParamsHandler, self).__init__(optimizer, param_name, tag)
def __call__(self, engine: Engine, logger: TensorboardLogger, event_name: Union[str, Events]) -> None:
if not isinstance(logger, TensorboardLogger):
raise RuntimeError("Handler OptimizerParamsHandler works only with TensorboardLogger")
global_step = engine.state.get_event_attrib_value(event_name)
tag_prefix = f"{self.tag}/" if self.tag else ""
params = {
f"{tag_prefix}{self.param_name}/group_{i}": float(param_group[self.param_name])
for i, param_group in enumerate(self.optimizer.param_groups)
}
for k, v in params.items():
logger.writer.add_scalar(k, v, global_step)
[docs]class WeightsScalarHandler(BaseWeightsScalarHandler):
"""Helper handler to log model's weights as scalars.
Handler, upon construction, iterates over named parameters of the model and keep
reference to ones permitted by `whitelist`. Then at every call, applies
reduction function to each parameter, produces a scalar and logs it.
Args:
model: model to log weights
reduction: function to reduce parameters into scalar
tag: common title for all produced plots. For example, "generator"
whitelist: specific weights to log. Should be list of model's submodules
or parameters names, or a callable which gets weight along with its name
and determines if it should be logged. Names should be fully-qualified.
For more information please refer to `PyTorch docs
<https://pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module.get_submodule>`_.
If not given, all of model's weights are logged.
Examples:
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
# Create a logger
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Attach the logger to the trainer to log model's weights norm after each iteration
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=WeightsScalarHandler(model, reduction=torch.norm)
)
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Log only `fc` weights
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=WeightsScalarHandler(
model,
whitelist=['fc']
)
)
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Log weights which have `bias` in their names
def has_bias_in_name(n, p):
return 'bias' in n
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=WeightsScalarHandler(model, whitelist=has_bias_in_name)
)
.. versionchanged:: 0.4.9
optional argument `whitelist` added.
"""
def __call__(self, engine: Engine, logger: TensorboardLogger, event_name: Union[str, Events]) -> None:
if not isinstance(logger, TensorboardLogger):
raise RuntimeError("Handler 'WeightsScalarHandler' works only with TensorboardLogger")
global_step = engine.state.get_event_attrib_value(event_name)
tag_prefix = f"{self.tag}/" if self.tag else ""
for name, p in self.weights:
name = name.replace(".", "/")
logger.writer.add_scalar(
f"{tag_prefix}weights_{self.reduction.__name__}/{name}",
self.reduction(p.data),
global_step,
)
[docs]class WeightsHistHandler(BaseWeightsHandler):
"""Helper handler to log model's weights as histograms.
Args:
model: model to log weights
tag: common title for all produced plots. For example, "generator"
whitelist: specific weights to log. Should be list of model's submodules
or parameters names, or a callable which gets weight along with its name
and determines if it should be logged. Names should be fully-qualified.
For more information please refer to `PyTorch docs
<https://pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module.get_submodule>`_.
If not given, all of model's weights are logged.
Examples:
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
# Create a logger
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Attach the logger to the trainer to log model's weights norm after each iteration
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=WeightsHistHandler(model)
)
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Log weights of `fc` layer
weights = ['fc']
# Attach the logger to the trainer to log weights norm after each iteration
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=WeightsHistHandler(model, whitelist=weights)
)
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Log weights which name include 'conv'.
weight_selector = lambda name, p: 'conv' in name
# Attach the logger to the trainer to log weights norm after each iteration
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=WeightsHistHandler(model, whitelist=weight_selector)
)
.. versionchanged:: 0.4.9
optional argument `whitelist` added.
"""
def __call__(self, engine: Engine, logger: TensorboardLogger, event_name: Union[str, Events]) -> None:
if not isinstance(logger, TensorboardLogger):
raise RuntimeError("Handler 'WeightsHistHandler' works only with TensorboardLogger")
global_step = engine.state.get_event_attrib_value(event_name)
tag_prefix = f"{self.tag}/" if self.tag else ""
for name, p in self.weights:
name = name.replace(".", "/")
logger.writer.add_histogram(
tag=f"{tag_prefix}weights/{name}", values=p.data.cpu().numpy(), global_step=global_step
)
[docs]class GradsScalarHandler(BaseWeightsScalarHandler):
"""Helper handler to log model's gradients as scalars.
Handler, upon construction, iterates over named parameters of the model and keep
reference to ones permitted by the `whitelist`. Then at every call, applies
reduction function to each parameter's gradient, produces a scalar and logs it.
Args:
model: model to log weights
reduction: function to reduce parameters into scalar
tag: common title for all produced plots. For example, "generator"
whitelist: specific gradients to log. Should be list of model's submodules
or parameters names, or a callable which gets weight along with its name
and determines if its gradient should be logged. Names should be
fully-qualified. For more information please refer to `PyTorch docs
<https://pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module.get_submodule>`_.
If not given, all of model's gradients are logged.
Examples:
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
# Create a logger
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Attach the logger to the trainer to log model's gradients norm after each iteration
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=GradsScalarHandler(model, reduction=torch.norm)
)
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Log gradient of `base`
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=GradsScalarHandler(
model,
reduction=torch.norm,
whitelist=['base']
)
)
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Log gradient of weights which belong to a `fc` layer
def is_in_fc_layer(n, p):
return 'fc' in n
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=GradsScalarHandler(model, whitelist=is_in_fc_layer)
)
.. versionchanged:: 0.4.9
optional argument `whitelist` added.
"""
def __call__(self, engine: Engine, logger: TensorboardLogger, event_name: Union[str, Events]) -> None:
if not isinstance(logger, TensorboardLogger):
raise RuntimeError("Handler 'GradsScalarHandler' works only with TensorboardLogger")
global_step = engine.state.get_event_attrib_value(event_name)
tag_prefix = f"{self.tag}/" if self.tag else ""
for name, p in self.weights:
if p.grad is None:
continue
name = name.replace(".", "/")
logger.writer.add_scalar(
f"{tag_prefix}grads_{self.reduction.__name__}/{name}", self.reduction(p.grad), global_step
)
[docs]class GradsHistHandler(BaseWeightsHandler):
"""Helper handler to log model's gradients as histograms.
Args:
model: model to log weights
tag: common title for all produced plots. For example, "generator"
whitelist: specific gradients to log. Should be list of model's submodules
or parameters names, or a callable which gets weight along with its name
and determines if its gradient should be logged. Names should be
fully-qualified. For more information please refer to `PyTorch docs
<https://pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module.get_submodule>`_.
If not given, all of model's gradients are logged.
Examples:
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
# Create a logger
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Attach the logger to the trainer to log model's weights norm after each iteration
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=GradsHistHandler(model)
)
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Log gradient of `fc.bias`
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=GradsHistHandler(model, whitelist=['fc.bias'])
)
.. code-block:: python
from ignite.handlers.tensorboard_logger import *
tb_logger = TensorboardLogger(log_dir="experiments/tb_logs")
# Log gradient of weights which have shape (2, 1)
def has_shape_2_1(n, p):
return p.shape == (2,1)
tb_logger.attach(
trainer,
event_name=Events.ITERATION_COMPLETED,
log_handler=GradsHistHandler(model, whitelist=has_shape_2_1)
)
.. versionchanged:: 0.4.9
optional argument `whitelist` added.
"""
def __call__(self, engine: Engine, logger: TensorboardLogger, event_name: Union[str, Events]) -> None:
if not isinstance(logger, TensorboardLogger):
raise RuntimeError("Handler 'GradsHistHandler' works only with TensorboardLogger")
global_step = engine.state.get_event_attrib_value(event_name)
tag_prefix = f"{self.tag}/" if self.tag else ""
for name, p in self.weights:
if p.grad is None:
continue
name = name.replace(".", "/")
logger.writer.add_histogram(
tag=f"{tag_prefix}grads/{name}", values=p.grad.cpu().numpy(), global_step=global_step
)