When running a DEHB run with more than one worker/ in parallel and later continuing the replay of the history fails if the config_ids are mixed up due to asynchronous execution.
The Code fails in line 998 of the dehb.py file due to an index out of bounds error:
Line 998: self.config_repository.configs[job_info["config_id"]].config = job_info["config"]
In my instance I started a parallel run with an initial population of 15 (config indeces 0 to 14). When restarting and the replay portion of the tell method is called, self.config_repository.configs has 17 items: the 15 from the initial population and two more with config_id=15 and 16 but job_info["config_id"] is 17. The history recorded is also not consecutively in config_ids. I suspect the replay doesn't consider asynchronous execution, which leads to this situation.
When running a DEHB run with more than one worker/ in parallel and later continuing the replay of the history fails if the config_ids are mixed up due to asynchronous execution.
The Code fails in line 998 of the dehb.py file due to an index out of bounds error:
Line 998:
self.config_repository.configs[job_info["config_id"]].config = job_info["config"]In my instance I started a parallel run with an initial population of 15 (config indeces 0 to 14). When restarting and the replay portion of the tell method is called, self.config_repository.configs has 17 items: the 15 from the initial population and two more with config_id=15 and 16 but job_info["config_id"] is 17. The history recorded is also not consecutively in config_ids. I suspect the replay doesn't consider asynchronous execution, which leads to this situation.