描述

🚀 Feature Request

I would like to be able:

to persist my study in order to resume my hyperparameter search from where I left it
set the gc_after_trial parameter of Optuna's study.optimize()

Motivation

Is your feature request related to a problem? Please describe. I'm always frustrated when my code crashes after 60 trials (out of 100). I suspect an OOM error. Being able to prevent the script from crashing in the first place with gc.collect() would be great. However, at least being able to resume my search from where it stopped would be a game changer.

Pitch

Describe the solution you'd like I would like to set gc_after_trial to True and a path to store my study parameters after each trial in my Optuna sweeper hydra config.

Describe alternatives you've considered I read Optuna's documentation but I'm not sure how to make their examples work with Hydra:

Are you willing to open a pull request? (See CONTRIBUTING) I'm not comfortable enough with Optuna's and Hydra's library to prepare a pull request.

貢獻者指南

技術棧: python
領域: machine learning
議題類型: feature
難度: 3
預計時間: half day
活動狀態: stale
清晰度: clear
前置要求: PythonHydra plugin basicsOptuna basics
新手友善度: 45
研究方向: Examine the existing Optuna sweeper plugin in Hydra (likely under `plugins/hydra optuna sweeper/`). Look at how `study.optimize()` is currently called and the plugin's configuration schema. Implement a new configuration parameter for `gc after trial` and a storage path for persistence. Use Optuna's `storage` argument to save the study (e.g., using `RDBStorage` or `pickle`). Ensure the study can be reloaded from the storage path to resume optimization. Refer to the Optuna FAQ links provided in the issue for examples on OOM prevention and study resumption.