Support numpy scalars in Trial.user_attrs

[toshihikoyanase]

Contributor License Agreement

This repository (optuna-dashboard) and Goptuna share common code.
This pull request may therefore be ported to Goptuna.
Make sure that you understand the consequences concerning licenses and check the box below if you accept the term before creating this pull request.

• I agree this patch may be ported to Goptuna by other Goptuna contributors.

Reference Issues/PRs

This PR addresses the following TODO comment:

optuna-dashboard/optuna_dashboard/_cached_extra_study_property.py

Line 88 in cd2159e

# TODO(c-bata): Support numpy-specific number types.

What does this implement/fix? Explain your changes.

• This change enables the conversion of numpy scalars to Python's native numerical values during the serialization of FrozenTrial.
• It recognizes np.integer and np.floating as sortable values.
• The process of checking for sortable values has been extracted into a private method for better clarity.

Discussion

numpy scalars can only stored in InMemoryStorage since they are not JSON-serializable.
I guess most of users use persistent storages such as RDBStorage and JournalFileStorage with the dashboard, so the change might benefit limited number of users.

Additional Context

Below is the code provided to verify the behavior described.

numpy-scalars-in-user-attrs.py

from __future__ import annotations

import logging
import os.path

import numpy as np
import optuna
from optuna.artifacts import FileSystemArtifactStore
from optuna_dashboard._app import create_app


host = "0.0.0.0"
port = 8080

optuna.logging.set_verbosity(logging.CRITICAL)


def create_optuna_storage() -> optuna.storages.InMemoryStorage:
    storage = optuna.storages.InMemoryStorage()

    # Single-objective study
    study = optuna.create_study(study_name="single-objective-user-attrs", storage=storage)

    def objective_single_user_attr(trial: optuna.Trial) -> float:
        x1 = trial.suggest_float("x1", 0, 10)
        x2 = trial.suggest_float("x2", 0, 10)
        if x1 < 5:
            trial.set_user_attr("X", "foo")
        else:
            trial.set_user_attr("X", "bar")
        trial.set_user_attr("Y", x1 + x2)
        trial.set_user_attr("np.float64", np.float64(x1 + x2))
        trial.set_user_attr("np.int32", np.int32(x1 + x2))
        trial.set_user_attr("int", int(x1 - x2))
        return (x1 - 2) ** 2 + (x2 - 5) ** 2

    study.optimize(objective_single_user_attr, n_trials=100)

    return storage


def main() -> None:
    storage = create_optuna_storage()
    base_path = os.path.join(os.path.dirname(__file__), "artifact")
    if not os.path.exists(base_path):
        os.mkdir(base_path)
    artifact_store = FileSystemArtifactStore(base_path=base_path)
    app = create_app(storage, artifact_store, debug=True)
    app.run(host=host, port=port, reloader=True)


if __name__ == "__main__":
    main()

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (c1c6adc) 64.35% compared to head (ad4d29f) 64.41%.
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #706      +/-   ##
==========================================
+ Coverage   64.35%   64.41%   +0.06%     
==========================================
  Files          35       35              
  Lines        2250     2254       +4     
==========================================
+ Hits         1448     1452       +4     
  Misses        802      802              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[c-bata]

LGTM. Thank you for your pull request.