Skip to content

Restart Hyperopt from pickle tries.pkl file #1815

@Cmurilochem

Description

@Cmurilochem

In restart_hyperopt and PR #1812, I've made some changes to allow us conduct hyperoptimization restarts. By adding some extra methods to the FileTrials class, we are for the moment able to restart jobs using a pickle tries.pkl file that is currently used to:

  • save final states of FileTrials after all trials in case of a normal job.
  • initiate the state of FileTrials to that of a previous run in case of restarts, e.g, using initial_filetrials = FileTrials.from_pkl(tries.pkl).

Would this fulfil our requirements for the moment ? I am also currently trying to add an extra @classmethod to the FileTrials class, e.g., filetrials = FileTrials.from_json(tries.json, *args, **kwargs) that could also be used for restarting in addition to tries.pkl. But I am facing some initial difficulties due to the special formatting and encodings present in tries.json. Not sure if the restart from json would possible or even sustainable as this could no longer work in case of any future changes in our current json structure. Please, I would appreciate any suggestions or comments.

Metadata

Metadata

Labels

questionFurther information is requested

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions