Skopt `resume` incorrectly finished when trying to run more `num_samples`

Ray Tune
1

Hi, I have started using ray.tune to run some optimizations and have been running into issues have the resume work properly.

search_alg = SkOptSearch(
    metric='metric', 
    mode='min',
)

analysis = tune.run(
        func,
        config=tune_config,
        num_samples=20,
        search_alg=search_alg,
        raise_on_failed_trial=False,
        name='run_name',
        verbose=0,
        resume=False,
    )

And then trying to resume with more samples:

search_alg = SkOptSearch(
    metric='metric', 
    mode='min',
)

analysis = tune.run(
        func,
        config=tune_config,
        num_samples=50,
        search_alg=search_alg,
        raise_on_failed_trial=False,
        name='run_name',
        verbose=0,
        resume=True,
    )

But it ends up thinking that the run has already finished. Am I doing something funky here?

2

Hmm, would it be possible instead to try doing:

search_alg = SkOptSearch(
    metric='metric', 
    mode='min',
)

tune.run(func, search_alg=search_alg, ...)
search_alg.save("path")

Then upon restore:

search_alg2 = SkOptSearch(...)
search_alg2.restore("path")

tune.run(func, search_alg=search_alg2, ...)