Ray-Tune with Optuna and tune.sample_from

Question

I'm trying to run OptunaSearch with a config that looks like this

config = {"algorithm": tune.choice(list(search_space.keys())),
          "params": tune.sample_from(lambda spec: search_space[spec.config.algorithm]['params'])}

Where the search_space is a dict of like this

search_space = {
    "algorithm_1": {
        "params": {
            "a": tune.randint(50, 200),
            "b": tune.choice([3, 5, 7, 9]),
            "c": tune.loguniform(0.001, 0.1),
        },
    },
    "algorithm_2": {
        "params": {
            "a": tune.loguniform(0.1, 10),
            "d": tune.loguniform(0.001, 1),
        },
    },
}

but I get an error saying

ValueError: Optuna search does not support parameters of type `Function` with samplers of type `NoneType`

I'm aware of this restriction, but I'm not sure how to use that "define-by-run" interface so this is possible, and also couldn't find any similar example

I tried like this

analysis = tune.run(
            tune.with_parameters(self.optimize, X=X),
            search_alg=OptunaSearch(),
            metric="score",
            mode="max",
            num_samples=15,
            config=config
        )