# SCVI hyperparameter optimization

**URL:** <https://discourse.scverse.org/t/scvi-hyperparameter-optimization/378>\
**Category:** scvi-tools\
**Tags:** scvi\
**Created:** [March 28, 2022, 6:35pm UTC](https://discourse.scverse.org/t/scvi-hyperparameter-optimization/378 "2022-03-28T18:35:15Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![vladie0](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/vladie0/32/187_2.png) [@vladie0](https://discourse.scverse.org/u/vladie0)\
**Post date:** [March 28, 2022, 6:35pm UTC](https://discourse.scverse.org/t/scvi-hyperparameter-optimization/378/1 "2022-03-28T18:35:15Z")

</div>

Dear all,

The SCVI autotune module (SCVI.inference) has been removed in the recent version updates.  
Is there a comparable approach building in scvi-tools for (hyper)parameter tuning or anything more convenient than a manual parameter sweep?

Any suggestion on best practices here?

Kr.

---

<div class="post-metadata">

**Author:** ![adamgayoso](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/adamgayoso/32/100_2.png) [@adamgayoso](https://discourse.scverse.org/u/adamgayoso)\
**Post date:** [March 28, 2022, 8:51pm UTC](https://discourse.scverse.org/t/scvi-hyperparameter-optimization/378/2 "2022-03-28T20:51:50Z")

</div>

You might consider using [Ray Tune](https://docs.ray.io/en/latest/tune/index.html).

We have a PR [here](https://github.com/scverse/scvi-tools/pull/997/files), with an experimental implementation.