Outcome_constraints on ax search algorithm

Hello,

I am running tune with Asha scheduler and AX search algorithm . I would like to add some constraints on AX , a kind of stopping criteria if two metrics are jointly satisfied. I did the following:

algo = AxSearch(outcome_constraints=["loss_validation<= 0.005"])
algo = tune.suggest.ConcurrencyLimiter(algo, max_concurrent=4)
Scheduler = ASHAScheduler( max_t=params["nb_epochs"], reduction_factor=4, ) 
analysis = tune.run( train_fn_with_parameters, metric="loss_validation", mode="min", config=config)

I tried the following but it doesn’t work l get an error related to the format of outcome_constraints:

`1/ algo = AxSearch(outcome_constraints=["loss_validation<= 0.005"])`
2/ algo = AxSearch(outcome_constraints=["metric<= 0.005"])

3/algo = AxSearch(outcome_constraints=["loss_validation<= 0.005" and "loss_train<= 0.0001"])

Any cue ?

Could you try AxSearch(outcome_constraints=["loss_validation<= 0.005", "loss_train<= 0.0001"])?

loss_validation and metric are the same. I defined that in tune.run(....,metric='loss_validation')

Edited the last post.
outcome_constraints is expecting a list of constraints.