Fix evolution tuner RuntimeError and IndexError #2695
Conversation
| for item in self.space: | ||
| is_rand[item] = True | ||
|
|
||
| config = json2parameter(self.searchspace_json, is_rand, self.random_state) |
There was a problem hiding this comment.
Is it possible to generate duplicate parameters?
There was a problem hiding this comment.
Yes, the first generation is random generated.
|
|
||
| def trial_end(self, parameter_id, success, **kwargs): | ||
| """ | ||
| To deal with trial failure. If a trial fails, readd the trial parameters into population. |
There was a problem hiding this comment.
I modify the docstring for clarity, please review on the latest version, thanks~
|
|
||
| total_config = config | ||
| if not self.population: | ||
| self.population.pop(1) |
There was a problem hiding this comment.
what if self.population only has one element?
There was a problem hiding this comment.
Thanks~ I have fixed it, please review on the latest version.
|
@colorjam UT is failed might because |
|
@colorjam please double check whether the generated configs are the same as the previous implementation when there is no failed trial. |
|
@QuanluZhang Thanks, I have checked it. I use the previous implementation to generate individuals. |
The current version of evolution tuner has several problems:
trialConcurrencyshould be larger thanpopulation_size, otherwise, the experiment willraise RuntimeError('The population is empty').Do not deal with trial failure. If the number of failed trails larger than
population_size, the experiment willraise RuntimeError('The population is empty').The end time of each trial may not be the same, so, the population length maybe 1. It can not make a comparison between two randomly selected individuals.