[Question] Multiple runs for same parameter values about syne-tune HOT 5 CLOSED

You use StoppingCriterion. If your training script only reports once at the end, you use max_num_evaluations.

One thing: GridSearcher does not have a method to return the grid size it is using. You can use len(scheduler.searcher.hp_values_combinations) to obtain the grid size.

Also note the order of evaluation. The grid searcher will run through all grid configs, then start again.

from syne-tune.

Hello, you can use allow_duplicates=True with GridSearcher. The searcher will then cycle through the grid multiple times. In this case, it will never stop returning configs. If you know the grid size, just query it K times grid size times.

from syne_tune.optimizer.baselines import GridSearch

scheduler = GridSearch(
   config_space,
   metric="...",
   search_options={"allow_duplicates": True},
)

from syne-tune.

Let me know if this works for you

from syne-tune.

Thanks! Where do you input the K * grid_size?

from syne-tune.

Thanks!!

from syne-tune.