def log_metrics(self,
metrics: _OUT_DICT,
step: Optional[int] = None) -> None:
"""Logs the metric dict passed in. If `step` parameter is None and `step` key is presented is metrics, uses
metrics["step"] as a step.
Args:
metrics: Metric values
step: Step for which metrics should be logged. Default value is `self.global_step` during training or
the total validation / test log step count during validation and testing.
"""
if self.trainer.logger is None or not metrics:
return
# turn all tensors to scalars
scalar_metrics = metrics_to_scalars(metrics)
if step is None:
step = scalar_metrics.pop("step", None)
if step is None:
# added metrics for convenience
scalar_metrics.setdefault("epoch", self.trainer.current_epoch)
step = self.trainer.global_step
# log actual metrics
self.trainer.logger.agg_and_log_metrics(scalar_metrics, step=step)
self.trainer.logger.save()
self._logged_metrics.update(scalar_metrics)
def log_metrics(self, metrics: Dict[str, _METRIC], step: Optional[int] = None) -> None:
"""Logs the metric dict passed in.
If `step` parameter is None and `step` key is presented is metrics,
uses metrics["step"] as a step
Args:
metrics: Metric values
step: Step for which metrics should be logged. Default value is `self.global_step` during training or
the total validation / test log step count during validation and testing.
"""
if self.trainer.logger is None or not metrics:
return
# add gpu memory
if self.trainer._device_type == DeviceType.GPU and self.log_gpu_memory:
mem_map = memory.get_memory_profile(self.log_gpu_memory)
metrics.update(mem_map)
# turn all tensors to scalars
scalar_metrics = metrics_to_scalars(metrics)
if step is None:
step = scalar_metrics.pop("step", None)
if step is None:
# added metrics for convenience
scalar_metrics.setdefault("epoch", self.trainer.current_epoch)
step = self.trainer.global_step
# log actual metrics
if self.trainer.is_global_zero:
self.trainer.logger.agg_and_log_metrics(scalar_metrics, step=step)
self.trainer.logger.save()
self._logged_metrics.update(scalar_metrics)
def metrics(self, on_step: bool) -> Dict[MetricSource, Dict[str, _METRIC]]:
metrics = {k: {} for k in MetricSource}
for _, result_metric in self.valid_items():
# extract forward_cache or computed from the ResultMetric. ignore when the output is None
value = apply_to_collection(result_metric, ResultMetric, self._get_cache, on_step, include_none=False)
# check if the collection is empty
has_tensor = False
def any_tensor(_):
nonlocal has_tensor
has_tensor = True
apply_to_collection(value, torch.Tensor, any_tensor)
if not has_tensor:
continue
name, forked_name = self._forked_name(result_metric, on_step)
# populate logging metrics
if result_metric.meta.logger:
metrics[MetricSource.LOG][forked_name] = value
# populate callback metrics. callback metrics don't take `_step` forked metrics
if self.training or result_metric.meta.on_epoch and not on_step:
metrics[MetricSource.CALLBACK][name] = value
metrics[MetricSource.CALLBACK][forked_name] = value
# populate progress_bar metrics. convert tensors to numbers
if result_metric.meta.prog_bar:
metrics[MetricSource.PBAR][forked_name] = metrics_to_scalars(value)
return metrics
def metrics(self, on_step: bool) -> _METRICS:
metrics = _METRICS(callback={}, log={}, pbar={})
for _, result_metric in self.valid_items():
# extract forward_cache or computed from the _ResultMetric. ignore when the output is None
value = apply_to_collection(result_metric,
_ResultMetric,
self._get_cache,
on_step,
include_none=False)
# convert metric collection to dict container.
if isinstance(value, _ResultMetricCollection):
value = dict(value.items())
# check if the collection is empty
has_tensor = False
def any_tensor(_: Any) -> None:
nonlocal has_tensor
has_tensor = True
apply_to_collection(value, torch.Tensor, any_tensor)
if not has_tensor:
continue
name, forked_name = self._forked_name(result_metric, on_step)
# populate logging metrics
if result_metric.meta.logger:
metrics["log"][forked_name] = value
# populate callback metrics. callback metrics don't take `_step` forked metrics
if self.training or result_metric.meta.on_epoch and not on_step:
metrics["callback"][name] = value
metrics["callback"][forked_name] = value
# populate progress_bar metrics. convert tensors to numbers
if result_metric.meta.prog_bar:
metrics["pbar"][forked_name] = metrics_to_scalars(value)
return metrics
def log_metrics(self, metrics, grad_norm_dict, step=None):
"""Logs the metric dict passed in.
If `step` parameter is None and `step` key is presented is metrics,
uses metrics["step"] as a step
Args:
metrics (dict): Metric values
grad_norm_dict (dict): Gradient norms
step (int): Step for which metrics should be logged. Default value is `self.global_step` during training or
the total validation / test log step count during validation and testing.
"""
# add gpu memory
if self.trainer._device_type == DeviceType.GPU and self.log_gpu_memory:
mem_map = memory.get_memory_profile(self.log_gpu_memory)
metrics.update(mem_map)
# add norms
metrics.update(grad_norm_dict)
# turn all tensors to scalars
scalar_metrics = metrics_to_scalars(metrics)
if "step" in scalar_metrics and step is None:
step = scalar_metrics.pop("step")
elif step is None:
# added metrics by Lightning for convenience
scalar_metrics['epoch'] = self.trainer.current_epoch
step = self.trainer.global_step
# log actual metrics
if self.trainer.logger is not None:
if self.trainer.is_global_zero:
self.trainer.logger.agg_and_log_metrics(scalar_metrics,
step=step)
self.trainer.logger.save()
# track the logged metrics
self.logged_metrics.update(scalar_metrics)
self.trainer.dev_debugger.track_logged_metrics_history(
scalar_metrics)
