currently there are 3 parameters which affect this feature:
experiment.seed
eprunner.fix_seed_pergeneration (or something like that)
optimizer.fix_seed_per_generation (or something like that)
One of them is not used iirc and should be removed. The experiment.seed is only indirectly related. And the remaining one should default to "true" and it should throw an error (or a warning?), if experiment.seed==0, because the the fix_seed_per_generation can only work when there is a global seed to begin with.
currently there are 3 parameters which affect this feature:
One of them is not used iirc and should be removed. The experiment.seed is only indirectly related. And the remaining one should default to "true" and it should throw an error (or a warning?), if experiment.seed==0, because the the fix_seed_per_generation can only work when there is a global seed to begin with.