Table of Contents

Introduction

OptunaSharp is a .NET port of Optuna, the define-by-run hyperparameter optimization framework. It aims for 1-to-1 behavioral parity with Optuna 5.x and is built on top of NumSharp for numerical work.

Status: core surface ported. The core algorithm surface is in place — studies, trials, distributions, samplers, pruners, storages, importance, terminator, and visualization — with more landing module-by-module.

The study / trial model

Optuna's two core concepts, which OptunaSharp mirrors:

  • Study — an optimization session over an objective function, with a direction (minimize/maximize), a sampler, a pruner, and a storage.
  • Trial — a single evaluation of the objective. During a trial the objective suggests hyperparameters (suggest_float, suggest_int, suggest_categorical, ...), which the sampler draws from the search space it has inferred so far (define-by-run).
# Optuna (Python) — the behavior OptunaSharp targets
import optuna

def objective(trial):
    x = trial.suggest_float("x", -10, 10)
    return (x - 2) ** 2

study = optuna.create_study(direction="minimize")
study.optimize(objective, n_trials=100)
print(study.best_params)   # -> {'x': ~2.0}

The same study in OptunaSharp: names take .NET casing, everything else keeps Optuna's shape.

using OptunaSharp;

var study = Optuna.CreateStudy(direction: StudyDirection.Minimize);

study.Optimize(trial =>
{
    double x = trial.SuggestFloat("x", -10, 10);
    return (x - 2) * (x - 2);
}, nTrials: 100);

Console.WriteLine(study.BestParams["x"]);   // -> ~2.0

Why a .NET port

  • Use Optuna's ergonomics from C#/F# without embedding CPython in production.
  • Interoperate with the rest of the SciSharp STACK (NumSharp, TensorFlow.NET, ML.NET) in one runtime.
  • Deterministic, reproducible optimization runs under a fixed seed, matching Optuna.

How parity is verified

Optuna is the source of truth. OptunaSharp is checked against the real library through the OptunaSharp.Tests.Interop project, which runs live Optuna in-process via pythonnet and asserts that OptunaSharp produces matching results — the same approach NumSharp uses against live NumPy.

Connect your trainer, storage, Python optimizer, cloud artifacts, or charts through the Integrations catalog. It covers every optional package and the training integrations built into Core.

Continue to Getting Started.

Edit this page