Optuna results change when rerun on Databricks

[Shwang]

The best trial results seem to change every time the same study is rerun. On Microsoft Azure this can be fixed by setting the sampler seed. However this solution doesn't seem to work on Databricks. Does anyone know why that is the case and how to make the results reproducible?

[Kaniz_Fatma]

Hi @Shwang, It's not uncommon for results to vary when re-running experiments, especially when using complex machine learning or optimization techniques. To better understand your situation, please provide more details about the changes you observed when re-running Optuna on Databricks. Any additional information you can provide about the specific experiment, configuration settings, data, and any differences in the environment would help diagnose the cause of the varying results. This will allow me to offer more accurate insights and suggestions to address your issue.