Closed nicolas-schreiber closed 1 year ago
I factored out the optuna config into a new config group so now it is called with python example/train.py +optim=optuna_clusterduck +platform=slurm_debug
. This way you can also run optuna with the basic launcher or clusterduck without optuna, without repeating any config information.
I tested also tested it on the Kluster too and it works like a charm :ok_hand: Let me know how you like the changes and we can merge.
Apparently with the Optuna Sweeper the
sweep_overrides
in the launcher are tuples not Lists, butload_sweep_config
doesn't like tuples... Otherwise added small example of how to use clusterduck together with the hydra optuna launcher, had to adjust the train script a bit to return a value optuna can optimize