from typing import Any
import optuna
from optuna._experimental import experimental
from optuna._imports import try_import
with try_import() as _imports:
from catalyst.dl import Callback
if not _imports.is_successful():
Callback = object # NOQA
[docs]@experimental("2.0.0")
class CatalystPruningCallback(Callback):
"""Catalyst callback to prune unpromising trials.
See `the example <https://github.com/optuna/optuna/blob/master/
examples/catalyst_simple.py>`_ if you want to add a pruning callback
which observes the accuracy of Catalyst's ``SupervisedRunner``.
Args:
trial:
A :class:`~optuna.trial.Trial` corresponding to the current evaluation of the
objective function.
metric (str):
Name of a metric, which is passed to `catalyst.core.State.valid_metrics` dictionary to
fetch the value of metric computed on validation set. Pruning decision is made based
on this value.
"""
def __init__(self, trial: optuna.trial.Trial, metric: str = "loss") -> None:
# set order=1000 to run pruning callback after other callbacks
# refer to `catalyst.core.CallbackOrder`
_imports.check()
super(CatalystPruningCallback, self).__init__(order=1000)
self._trial = trial
self.metric = metric
def on_epoch_end(self, state: Any) -> None:
current_score = state.valid_metrics[self.metric]
self._trial.report(current_score, state.epoch)
if self._trial.should_prune():
message = "Trial was pruned at epoch {}.".format(state.epoch)
raise optuna.TrialPruned(message)