Search Spaces

Because OptunaSharp is define-by-run, there is no static search-space declaration — the space is whatever the objective happened to suggest. Different trials can suggest different parameters (a conditional if branch, a variable number of layers), so the space is dynamic. Several samplers need a stable, shared space to reason about; that is what search-space inference computes. OptunaSharp ports optuna.search_space into OptunaSharp.SearchSpace.

Intersection search space

The intersection search space is the set of parameters that appear in every completed trial, with a distribution compatible across all of them. It is the safe common ground a relative sampler can model jointly.

using OptunaSharp.Collections;     // OrderedDict
using OptunaSharp.Distributions;   // BaseDistribution
using OptunaSharp.SearchSpace;

var searchSpace = new IntersectionSearchSpace();
OrderedDict<string, BaseDistribution> space = searchSpace.Calculate(study);

foreach (var (name, dist) in space)
    Console.WriteLine($"{name}: {dist}");

IntersectionSearchSpace can cache incrementally across calls within a run:

public IntersectionSearchSpace(bool includePruned = false);
public OrderedDict<string, BaseDistribution> Calculate(Study study, bool useCache = false);

Set includePruned: true to also fold in parameters from pruned trials. Pass useCache: true to reuse the result of previous calls and only fold in trials seen since.

For a one-shot computation over a specific list of trials, use the static helper:

OrderedDict<string, BaseDistribution> space =
    SearchSpaceFunctions.IntersectionSearchSpaceOf(trials, includePruned: false);

The result preserves the order in which parameters were first seen (OrderedDict), matching Optuna.

Grouped search space

When a sampler is configured to model parameters in independent groups (for example new TPESampler(group: true)), OptunaSharp decomposes the space into maximal subsets of parameters that always co-occur, so each group is modeled jointly and groups are modeled independently. That decomposition is an internal detail driven by the sampler — you enable it through the sampler's group option rather than calling it directly. See Samplers.

Relationship to Optuna

Optuna (Python)	OptunaSharp
`optuna.search_space.IntersectionSearchSpace()`	`new IntersectionSearchSpace()`
`.calculate(study)`	`.Calculate(study)`
`intersection_search_space(trials)`	`SearchSpaceFunctions.IntersectionSearchSpaceOf(trials)`

Table of Contents

Search Spaces

Intersection search space

Grouped search space

Relationship to Optuna

See also