Saving/Resuming a Study with an RDB backend

A relational-database (RDB) backend makes a study persistent: you can save it, resume it later, and inspect the history of every trial. It is also what lets several processes optimize the same study at once. This recipe uses SQLite, a single-file database that needs no server; install OptunaSharp.Storage.Sqlite to get RDBStorage.

SQLite is one host only. For PostgreSQL or MySQL use OptunaSharp.Storage.EFCore; to share a study across machines, put it behind the gRPC proxy or use journal-file storage. See the Storage guide.

New study

Create a persistent study by passing a storage. The SQLite file example.db is created automatically and a new study record is written into it.

using OptunaSharp;
using OptunaSharp.Storages;

using var storage = new RDBStorage("sqlite:///example.db");
var study = Optuna.CreateStudy(storage: storage, studyName: "example-study");

study.Optimize(trial =>
{
    double x = trial.SuggestFloat("x", -10, 10);
    return (x - 2) * (x - 2);
}, nTrials: 3);

studyName is the unique identifier you resume by. Because the same schema Optuna writes is used, Python Optuna can open example.db and vice versa.

Resume study

To resume, construct a study with the same name and storage and pass loadIfExists: true — it attaches to the existing record instead of throwing that the name is taken.

using var storage = new RDBStorage("sqlite:///example.db");
var study = Optuna.CreateStudy(
    storage: storage, studyName: "example-study", loadIfExists: true);

study.Optimize(trial =>
{
    double x = trial.SuggestFloat("x", -10, 10);
    return (x - 2) * (x - 2);
}, nTrials: 3);   // these 3 trials are appended to the 3 already stored

The storage does not persist the sampler's or pruner's in-memory state. If you seeded the sampler for reproducibility, construct a fresh seeded sampler and pass it to CreateStudy when you resume.

Reading the stored history

Every property of a study reads through the storage, so a resumed study sees all prior trials:

using var storage = new RDBStorage("sqlite:///example.db");
var study = Optuna.CreateStudy(
    storage: storage, studyName: "example-study", loadIfExists: true);

Console.WriteLine($"Best params: {study.BestParams["x"]}");
Console.WriteLine($"Best value:  {study.BestValue}");
Console.WriteLine($"Best trial:  {study.BestTrial}");
Console.WriteLine($"Trials:      {study.Trials.Count}");

// Tabular export (the analog of Optuna's trials_dataframe):
TrialsTable df = study.TrialsDataframe(attrs: new[] { "number", "value", "params", "state" });

TrialsDataframe returns a TrialsTable with a column per field — no pandas runtime required.

Relationship to Optuna

Optuna (Python)	OptunaSharp
`optuna.create_study(study_name=…, storage="sqlite:///example.db")`	`Optuna.CreateStudy(studyName: …, storage: new RDBStorage("sqlite:///example.db"))`
`optuna.create_study(…, load_if_exists=True)`	`Optuna.CreateStudy(…, loadIfExists: true)`
`study.trials_dataframe(attrs=("number","value","params","state"))`	`study.TrialsDataframe(attrs: new[] { "number", "value", "params", "state" })`
`study.best_params` / `study.best_value`	`study.BestParams` / `study.BestValue`

Table of Contents

Saving/Resuming a Study with an RDB backend

New study

Resume study

Reading the stored history

Relationship to Optuna

See also