how to make use of existent experimental results? about optuna

[toshihikoyanase]

Hi @speedcell4,

Thank you for your interest in Optuna.
Currently, Optuna does not provide functions/methods to import existing experimental results, but we are investigating the best way to implement the trial-import feature.

Here, I want to share my current workaround to import existing results.

1. Please install Optuna from injected-trial branch:

https://github.com/pfnet/optuna/tree/injected-trial/optuna

2. Use InjectedTrial.suggest_xxx methods to save existing results. Please refer to the following example for further details:

https://gist.github.com/toshihikoyanase/c90209d0352a119b26e1be7f152dbf53

---

on the other hand, currently I think I have a workaround

Thank you for sharing your approach.
I think it is workable if you do not use suggest_categorical. Currently, it does not accept dynamic value spaces.

Here's an example to change ranges of suggest_xxx to import externally suggested parameter values.

https://gist.github.com/toshihikoyanase/99f4119f226da4ad227cd6f381136c72

Please note that TPESampler may raise ValueError when we set low==high as reported in #318. I used RandomSampler to avoid the issue in the above example.

If you have any questions, please let me know.

[speedcell4]

Thank you so much, that really helps