Merge pull request optuna#1 from toshihikoyanase/prototype-auto-sampl…

…er-with-fallback

Support multi-objective and constrained optimization.

package/samplers/auto_sampler/README.md

@@ -19,8 +19,10 @@ This sampler currently accepts only `seed`.
 
 ## Installation
 
+This sampler requires optional dependencies of Optuna.
+
 ```shell
-$ pip install scipy torch cmaes
+$ pip install "optuna[optional]"
 ```
 
 ## Example

package/samplers/auto_sampler/sampler.py

@@ -2,12 +2,14 @@
 
 from collections.abc import Sequence
 from typing import Any
+from typing import Callable
 from typing import TYPE_CHECKING
 
 from optuna.distributions import CategoricalDistribution
 from optuna.samplers import BaseSampler
 from optuna.samplers import CmaEsSampler
 from optuna.samplers import GPSampler
+from optuna.samplers import NSGAIISampler
 from optuna.samplers import RandomSampler
 from optuna.samplers import TPESampler
 from optuna.samplers._lazy_random_state import LazyRandomState
@@ -31,30 +33,50 @@ class AutoSampler(BaseSampler):
         .. testcode::
 
             import optuna
-            from optuna.samplers import AutoSampler
+            import optunahub
 
 
             def objective(trial):
                 x = trial.suggest_float("x", -5, 5)
                 return x**2
 
-
-            study = optuna.create_study(sampler=AutoSampler())
-            study.optimize(objective, n_trials=10)
+            module = optunahub.load_module("samplers/auto_sampler")
+            study = optuna.create_study(sampler=module.AutoSampler())
+            study.optimize(objective, n_trials=300)
 
     .. note::
-        This sampler might require ``scipy``, ``torch``, and ``cmaes``.
-        You can install these dependencies with ``pip install scipy torch cmaes``.
+        This sampler requires optional dependencies of Optuna.
+        You can install them with ``pip install "optuna[optional]"``.
+        Alternatively, you can install them with ``pip install -r https://hub.optuna.org/samplers/auto_sampler/requirements.txt``.
 
     Args:
         seed: Seed for random number generator.
+        constraints_func:
+            An optional function that computes the objective constraints. It must take a
+            :class:`~optuna.trial.FrozenTrial` and return the constraints. The return value must
+            be a sequence of :obj:`float` s. A value strictly larger than 0 means that a
+            constraints is violated. A value equal to or smaller than 0 is considered feasible.
+            If ``constraints_func`` returns more than one value for a trial, that trial is
+            considered feasible if and only if all values are equal to 0 or smaller.
+
+            The ``constraints_func`` will be evaluated after each successful trial.
+            The function won't be called when trials fail or they are pruned, but this behavior is
+            subject to change in the future releases.
+
+            .. note::
+                If you enable this feature, Optuna's default sampler will be selected automatically.
 
     """
 
-    def __init__(self, seed: int | None = None) -> None:
+    def __init__(
+        self,
+        seed: int | None = None,
+        constraints_func: Callable[[FrozenTrial], Sequence[float]] | None = None,
+    ) -> None:
         self._rng = LazyRandomState(seed)
         seed_for_random_sampler = self._rng.rng.randint(1 << 32)
         self._sampler: BaseSampler = RandomSampler(seed=seed_for_random_sampler)
+        self._constraints_func = constraints_func
 
     def reseed_rng(self) -> None:
         self._rng.rng.seed()
@@ -75,10 +97,24 @@ def _include_conditional_param(self, study: Study) -> bool:
     def _determine_sampler(
         self, study: Study, trial: FrozenTrial, search_space: dict[str, BaseDistribution]
     ) -> None:
+        if len(study.directions) > 1:
+            # Fallback to the default sampler if the study has multiple objectives.
+            if isinstance(self._sampler, NSGAIISampler):
+                return
+            # TODO(toshihikoyanase): add warning message about fallback.
+            self._sampler = NSGAIISampler(constraints_func=self._constraints_func)
+            return
+
         if isinstance(self._sampler, TPESampler):
             return
 
         seed = self._rng.rng.randint(1 << 32)
+        if self._constraints_func is not None:
+            # Fallback to the default sampler if the study has constraints.
+            # TODO(toshihikoyanase): add warning message about fallback.
+            self._sampler = TPESampler(seed=seed, constraines_func=self._constraints_func)
+            return
+
         if any(
             isinstance(d, CategoricalDistribution) for d in search_space.values()
         ) or self._include_conditional_param(study):