from dataclasses import dataclass
import decimal
from typing import Any
from typing import Dict
from typing import Iterable
from typing import List
from typing import Optional
from typing import Sequence
from typing import Tuple
import numpy as np
from optuna._experimental import experimental_class
from optuna.distributions import BaseDistribution
from optuna.distributions import CategoricalDistribution
from optuna.distributions import FloatDistribution
from optuna.distributions import IntDistribution
from optuna.samplers import BaseSampler
from optuna.samplers._lazy_random_state import LazyRandomState
from optuna.study import Study
from optuna.trial import create_trial
from optuna.trial import FrozenTrial
from optuna.trial import TrialState
@dataclass
class _TreeNode:
# This is a class to represent the tree of search space.
# A tree node has three states:
# 1. Unexpanded. This is represented by children=None.
# 2. Leaf. This is represented by children={} and param_name=None.
# 3. Normal node. It has a param_name and non-empty children.
param_name: Optional[str] = None
children: Optional[Dict[float, "_TreeNode"]] = None
def expand(self, param_name: Optional[str], search_space: Iterable[float]) -> None:
# If the node is unexpanded, expand it.
# Otherwise, check if the node is compatible with the given search space.
if self.children is None:
# Expand the node
self.param_name = param_name
self.children = {value: _TreeNode() for value in search_space}
else:
if self.param_name != param_name:
raise ValueError(f"param_name mismatch: {self.param_name} != {param_name}")
if self.children.keys() != set(search_space):
raise ValueError(
f"search_space mismatch: {set(self.children.keys())} != {set(search_space)}"
)
def set_leaf(self) -> None:
self.expand(None, [])
def add_path(
self, params_and_search_spaces: Iterable[Tuple[str, Iterable[float], float]]
) -> Optional["_TreeNode"]:
# Add a path (i.e. a list of suggested parameters in one trial) to the tree.
current_node = self
for param_name, search_space, value in params_and_search_spaces:
current_node.expand(param_name, search_space)
assert current_node.children is not None
if value not in current_node.children:
return None
current_node = current_node.children[value]
return current_node
def count_unexpanded(self) -> int:
# Count the number of unexpanded nodes in the subtree.
return (
1
if self.children is None
else sum(child.count_unexpanded() for child in self.children.values())
)
def sample_child(self, rng: np.random.RandomState) -> float:
assert self.children is not None
# Sample an unexpanded node in the subtree uniformly, and return the first
# parameter value in the path to the node.
# Equivalently, we sample the child node with weights proportional to the number
# of unexpanded nodes in the subtree.
weights = np.array(
[child.count_unexpanded() for child in self.children.values()], dtype=np.float64
)
weights /= weights.sum()
return rng.choice(list(self.children.keys()), p=weights)
[docs]
@experimental_class("3.1.0")
class BruteForceSampler(BaseSampler):
"""Sampler using brute force.
This sampler performs exhaustive search on the defined search space.
Example:
.. testcode::
import optuna
def objective(trial):
c = trial.suggest_categorical("c", ["float", "int"])
if c == "float":
return trial.suggest_float("x", 1, 3, step=0.5)
elif c == "int":
a = trial.suggest_int("a", 1, 3)
b = trial.suggest_int("b", a, 3)
return a + b
study = optuna.create_study(sampler=optuna.samplers.BruteForceSampler())
study.optimize(objective)
Note:
The defined search space must be finite. Therefore, when using
:class:`~optuna.distributions.FloatDistribution` or
:func:`~optuna.trial.Trial.suggest_float`, ``step=None`` is not allowed.
Note:
The sampler may fail to try the entire search space in when the suggestion ranges or
parameters are changed in the same :class:`~optuna.study.Study`.
Args:
seed:
A seed to fix the order of trials as the search order randomly shuffled. Please note
that it is not recommended using this option in distributed optimization settings since
this option cannot ensure the order of trials and may increase the number of duplicate
suggestions during distributed optimization.
"""
def __init__(self, seed: Optional[int] = None) -> None:
self._rng = LazyRandomState(seed)
[docs]
def infer_relative_search_space(
self, study: Study, trial: FrozenTrial
) -> Dict[str, BaseDistribution]:
return {}
[docs]
def sample_relative(
self, study: Study, trial: FrozenTrial, search_space: Dict[str, BaseDistribution]
) -> Dict[str, Any]:
return {}
@staticmethod
def _populate_tree(
tree: _TreeNode, trials: Iterable[FrozenTrial], params: Dict[str, Any]
) -> None:
# Populate tree under given params from the given trials.
incomplete_leaves: List[_TreeNode] = []
for trial in trials:
if not all(p in trial.params and trial.params[p] == v for p, v in params.items()):
continue
leaf = tree.add_path(
(
(
param_name,
_enumerate_candidates(param_distribution),
param_distribution.to_internal_repr(trial.params[param_name]),
)
for param_name, param_distribution in trial.distributions.items()
if param_name not in params
)
)
if leaf is not None:
# The parameters are on the defined grid.
if trial.state.is_finished():
leaf.set_leaf()
else:
incomplete_leaves.append(leaf)
# Add all incomplete leaf nodes at the end because they may not have complete search space.
for leaf in incomplete_leaves:
if leaf.children is None:
leaf.set_leaf()
[docs]
def sample_independent(
self,
study: Study,
trial: FrozenTrial,
param_name: str,
param_distribution: BaseDistribution,
) -> Any:
trials = study.get_trials(
deepcopy=False,
states=(
TrialState.COMPLETE,
TrialState.PRUNED,
TrialState.RUNNING,
TrialState.FAIL,
),
)
tree = _TreeNode()
candidates = _enumerate_candidates(param_distribution)
tree.expand(param_name, candidates)
# Populating must happen after the initialization above to prevent `tree` from
# being initialized as an empty graph, which is created with n_jobs > 1
# where we get trials[i].params = {} for some i.
self._populate_tree(tree, (t for t in trials if t.number != trial.number), trial.params)
if tree.count_unexpanded() == 0:
return param_distribution.to_external_repr(self._rng.rng.choice(candidates))
else:
return param_distribution.to_external_repr(tree.sample_child(self._rng.rng))
[docs]
def after_trial(
self,
study: Study,
trial: FrozenTrial,
state: TrialState,
values: Optional[Sequence[float]],
) -> None:
trials = study.get_trials(
deepcopy=False,
states=(
TrialState.COMPLETE,
TrialState.PRUNED,
TrialState.RUNNING,
TrialState.FAIL,
),
)
tree = _TreeNode()
self._populate_tree(
tree,
(
(
t
if t.number != trial.number
else create_trial(
state=state, # Set current trial as complete.
values=values,
params=trial.params,
distributions=trial.distributions,
)
)
for t in trials
),
{},
)
if tree.count_unexpanded() == 0:
study.stop()
def _enumerate_candidates(param_distribution: BaseDistribution) -> Sequence[float]:
if isinstance(param_distribution, FloatDistribution):
if param_distribution.step is None:
raise ValueError(
"FloatDistribution.step must be given for BruteForceSampler"
" (otherwise, the search space will be infinite)."
)
low = decimal.Decimal(str(param_distribution.low))
high = decimal.Decimal(str(param_distribution.high))
step = decimal.Decimal(str(param_distribution.step))
ret = []
value = low
while value <= high:
ret.append(float(value))
value += step
return ret
elif isinstance(param_distribution, IntDistribution):
return list(
range(param_distribution.low, param_distribution.high + 1, param_distribution.step)
)
elif isinstance(param_distribution, CategoricalDistribution):
return list(range(len(param_distribution.choices))) # Internal representations.
else:
raise ValueError(f"Unknown distribution {param_distribution}.")