Saving a study locally

Hi !

From what I read in the optuna.study.create_study documentation, the storage parameter only takes a database URL as input. Is there a way to persist a study locally? (I’m already using MLflow to log my metrics and model parameters, could it help?)

Thanks for your input 😃

[dianemarquette]

@nbkn865 Thanks for your input 😃. I wasn’t aware of the built-in SQLite database in Python that can run locally. I’m going to try to implement your solution 👍🏻

[nbkn865]

@dianemarquette, not familiar with Hydra, but it looks like a Python package, and it looks like there’s a parameter for Optuna storage. Given that, have you tried using the built-in SQLite database in Python?

From a Python call, you can set an Optuna study to store in SQLite that resides locally on whatever path your script is on.

For example:

study_name = “my_study” storage_name = f"sqlite:///{study_name}.db" study = optuna.create_study(study_name=study_name, storage=storage_name, load_if_exists=True)

This will create a file called my_study.db in the directory of your Python script that stores the study

You can then query its results using the following method call that creates a pandas dataframe:

df = study.trials_dataframe() # Get all columns of study

Based on the above, and using the Hydra documentation, maybe you can try the following parameter for storage and study_name:

sampler: target: optuna.samplers.TPESampler seed: 123 … warn_independent_sampling: true target: hydra_plugins.hydra_optuna_sweeper.optuna_sweeper.OptunaSweeper direction: minimize storage: sqlite:///my_test_study.db study_name: my_test_study