Files
optuna-dashboard/optuna_dashboard/_cached_extra_study_property.py
T
2023-11-22 15:22:23 +09:00

112 lines
4.2 KiB
Python

from __future__ import annotations
import copy
import numbers
import threading
from typing import List
from typing import Optional
from typing import Set
from typing import Tuple
from typing import TYPE_CHECKING
from optuna.distributions import BaseDistribution
from optuna.trial import FrozenTrial
from optuna.trial import TrialState
# In-memory cache
cached_extra_study_property_cache_lock = threading.Lock()
cached_extra_study_property_cache: dict[int, "_CachedExtraStudyProperty"] = {}
if TYPE_CHECKING:
SearchSpaceSetT = Set[Tuple[str, BaseDistribution]]
SearchSpaceListT = List[Tuple[str, BaseDistribution]]
def get_cached_extra_study_property(
study_id: int, trials: list[FrozenTrial]
) -> tuple[SearchSpaceListT, SearchSpaceListT, list[tuple[str, bool]], bool]:
with cached_extra_study_property_cache_lock:
cached_extra_study_property = cached_extra_study_property_cache.get(study_id, None)
if cached_extra_study_property is None:
cached_extra_study_property = _CachedExtraStudyProperty()
cached_extra_study_property.update(trials)
cached_extra_study_property_cache[study_id] = cached_extra_study_property
return (
cached_extra_study_property.intersection_search_space,
cached_extra_study_property.union_search_space,
cached_extra_study_property.union_user_attrs,
cached_extra_study_property.has_intermediate_values,
)
class _CachedExtraStudyProperty:
def __init__(self) -> None:
self._cursor: int = -1
self._intersection_search_space: Optional[SearchSpaceSetT] = None
self._union_search_space: SearchSpaceSetT = set()
self._union_user_attrs: dict[str, bool] = {} # attr_name: is_sortable (= is_number)
self.has_intermediate_values: bool = False
@property
def intersection_search_space(self) -> SearchSpaceListT:
if self._intersection_search_space is None:
return []
intersection = list(self._intersection_search_space)
intersection.sort(key=lambda x: x[0])
return intersection
@property
def union_search_space(self) -> SearchSpaceListT:
union = list(self._union_search_space)
union.sort(key=lambda x: x[0])
return union
@property
def union_user_attrs(self) -> list[tuple[str, bool]]:
union = [(name, is_sortable) for name, is_sortable in self._union_user_attrs.items()]
sorted(union, key=lambda x: x[0])
return union
def update(self, trials: list[FrozenTrial]) -> None:
next_cursor = self._cursor
for trial in reversed(trials):
if self._cursor > trial.number:
break
if not trial.state.is_finished():
next_cursor = trial.number
self._update_user_attrs(trial)
if trial.state != TrialState.FAIL:
self._update_intermediate_values(trial)
self._update_search_space(trial)
self._cursor = next_cursor
def _update_user_attrs(self, trial: FrozenTrial) -> None:
current_user_attrs = {
k: not isinstance(v, bool) and isinstance(v, numbers.Real)
for k, v in trial.user_attrs.items()
}
for attr_name, current_is_sortable in current_user_attrs.items():
is_sortable = self._union_user_attrs.get(attr_name)
if is_sortable is None:
self._union_user_attrs[attr_name] = current_is_sortable
elif is_sortable and not current_is_sortable:
self._union_user_attrs[attr_name] = False
def _update_intermediate_values(self, trial: FrozenTrial) -> None:
if not self.has_intermediate_values and len(trial.intermediate_values) > 0:
self.has_intermediate_values = True
def _update_search_space(self, trial: FrozenTrial) -> None:
current = set([(n, d) for n, d in trial.distributions.items()])
self._union_search_space = self._union_search_space.union(current)
if self._intersection_search_space is None:
self._intersection_search_space = copy.copy(current)
else:
self._intersection_search_space = self._intersection_search_space.intersection(current)