How to Persist and Resume Studies¶

Interactive notebook

See the companion notebook for a runnable example. View · Open in marimo

This guide shows you how to save optimization results to a database and resume a search from where it left off. Use this when experiments are long-running or you want to build on previous results.

Prerequisites¶

Sklearn-Optuna installed (Getting Started)
A working OptunaSearchCV search

Save Trials to a Database¶

Wrap an Optuna storage backend with the Storage class and pass it to OptunaSearchCV:

import optuna
from sklearn_optuna import OptunaSearchCV, Storage

search = OptunaSearchCV(
    estimator, param_distributions,
    storage=Storage(storage=optuna.storages.RDBStorage, url="sqlite:///study.db"),
)
search.fit(X, y)

Trials are persisted to study.db as they complete. If the process is interrupted, completed trials are not lost.

Resume a Previous Search¶

After fitting, the study is available as search.study_. Pass it back to fit() to continue from where the previous run stopped:

# First run: 50 trials
search.fit(X, y)

# Resume: 50 more trials appended to the same study
search.fit(X, y, study=search.study_)

If you are using a Storage backend, you can also resume across sessions by loading the study from the database.

Use a Study Name¶

Pass study_name to organize multiple studies in the same database:

search = OptunaSearchCV(
    estimator, param_distributions,
    storage=Storage(storage=optuna.storages.RDBStorage, url="sqlite:///experiments.db"),
    study_name="logistic_regression_v2",
)