import optuna
from optuna._imports import try_import
with try_import() as _imports:
import tensorflow as tf
from tensorflow.estimator import SessionRunHook
from tensorflow_estimator.python.estimator.early_stopping import read_eval_metrics
if not _imports.is_successful():
SessionRunHook = object # NOQA
[docs]class TensorFlowPruningHook(SessionRunHook):
"""TensorFlow SessionRunHook to prune unpromising trials.
See `the example <https://github.com/optuna/optuna-examples/tree/main/
tensorflow/tensorflow_estimator_integration.py>`_
if you want to add a pruning hook to TensorFlow's estimator.
Args:
trial:
A :class:`~optuna.trial.Trial` corresponding to the current evaluation of
the objective function.
estimator:
An estimator which you will use.
metric:
An evaluation metric for pruning, e.g., ``accuracy`` and ``loss``.
run_every_steps:
An interval to watch the summary file.
"""
def __init__(
self,
trial: optuna.trial.Trial,
estimator: "tf.estimator.Estimator",
metric: str,
run_every_steps: int,
) -> None:
_imports.check()
self._trial = trial
self._estimator = estimator
self._current_summary_step = -1
self._metric = metric
self._global_step_tensor = None
self._timer = tf.estimator.SecondOrStepTimer(every_secs=None, every_steps=run_every_steps)
def begin(self) -> None:
self._global_step_tensor = tf.compat.v1.train.get_global_step()
def before_run(
self, run_context: "tf.estimator.SessionRunContext"
) -> "tf.estimator.SessionRunArgs":
del run_context
return tf.estimator.SessionRunArgs(self._global_step_tensor)
def after_run(
self,
run_context: "tf.estimator.SessionRunContext",
run_values: "tf.estimator.SessionRunValues",
) -> None:
global_step = run_values.results
# Get eval metrics every n steps.
if self._timer.should_trigger_for_step(global_step):
self._timer.update_last_triggered_step(global_step)
eval_metrics = read_eval_metrics(self._estimator.eval_dir())
else:
eval_metrics = None
if eval_metrics:
summary_step = next(reversed(eval_metrics))
latest_eval_metrics = eval_metrics[summary_step]
# If there exists a new evaluation summary.
if summary_step > self._current_summary_step:
current_score = latest_eval_metrics[self._metric]
if current_score is None:
current_score = float("nan")
self._trial.report(float(current_score), step=summary_step)
self._current_summary_step = summary_step
if self._trial.should_prune():
message = "Trial was pruned at iteration {}.".format(self._current_summary_step)
raise optuna.TrialPruned(message)