In my setting I have an abstract situation like the following, which shall only function as an example case:
base = trial.suggest_int(1, 3)
power = trial.suggest_int(1, 10)
# value = base ** power
As when the base == 1 the power parameter becomes irrelevant and I would like to fix it to 1.
For example:
base = trial.suggest_int("base", 1, 3)
if base == 1:
# Different distribution! But still inside the other.
power = trial.suggest_int("power", 1, 1)
else:
power = trial.suggest_int("power", 1, 10)
While this works it creates later problems in the form of ValueErrors because the underlying distributions are not the same.
How can I suggest a fixed value with the same parameter name that depends on another value that is sampled within the trial?