# Using ray tune as offline hyperparameter suggester

**URL:** <https://discuss.ray.io/t/using-ray-tune-as-offline-hyperparameter-suggester/7590>\
**Category:** Uncategorized\
**Created:** [September 17, 2022, 6:05pm UTC](https://discuss.ray.io/t/using-ray-tune-as-offline-hyperparameter-suggester/7590 "2022-09-17T18:05:48Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![kgmagnetic](https://avatars.discourse-cdn.com/v4/letter/k/6de8d8/32.png) [@kgmagnetic](https://discuss.ray.io/u/kgmagnetic)\
**Post date:** [September 17, 2022, 6:05pm UTC](https://discuss.ray.io/t/using-ray-tune-as-offline-hyperparameter-suggester/7590/1 "2022-09-17T18:05:48Z")

</div>

Hi all, is there a way to use ray tune to suggest a new set of hyperparameters based on an experiment history? The extensive suite of tools that come with ray for plugging into and orchestrating your training process are impressive, but for my use case I’d like to omit them at the moment. Instead, I’d like to make use of the search space optimization algorithms in ray tune by using it as an offline tuner. Essentially looking to set up a process that would act as:

`suggest_hyperparams(search_space, list_of_samples, list_of_scores, ...) -> next_hyperparams_sample_to_run`

Is this a supported flow? Or if not officially supported, any pointers on how to start hacking something together?