How to convert study.pkl to study.db?

Note to the questioner

1.As a new user of Optuna, I really love it. However, in the beginning I saved an existing study to .pkl file, but now I wish to convert it to RDB backend in order to perform the parallelization . How can I have that?

I may wish something like

existing_study=joblib.load("existing_study.pkl")

#save it to .db file
existing_study.save( 'sql_base.db')

Is there any solution?

1. In addition, how can I list all study names which have been saved in .db file? Sorry, I do not have so much knowledge in RDB also…

Thank you in advance.

[nzw0301]

I think the following way should work.

Let’s assume that we have finished a few trials and saved the study as existing_study.pkl.

import optuna
import joblib


def objective(trial):
    x = trial.suggest_float("x", -1, 1)
    return x ** 2

study = optuna.create_study()
study.optimize(objective, n_trials=3)

joblib.dump(study, "existing_study.pkl")

Then we load the existing_study.pkl and save the study into a database, example.db. I think that can be done by using create_new_study and create_new_trial.

import optuna
import joblib


study = joblib.load("existing_study.pkl")

storage = optuna.storages.RDBStorage(
    url="sqlite:///example.db"
)
study_id = storage.create_new_study(study_name="study_on_db")

for trial in study.get_trials():
    storage.create_new_trial(study_id=study_id, template_trial=trial)

study = optuna.load_study(study_name="study_on_db", storage=storage)

study.optimize(objective, n_trials=10)

len(study.get_trials())  # == 13

[yywangvr]

Thank you again. I guess this may because I run Optuna in two processes as indicated in here, and therefore I do not have values but obtain intermediate_values instead.

When I retry my code in one process, everything seems fine.