[BUG]: nested constraint on exp + tanh not applied

[astro-nova]

What happened?

I have a few nested constraints, basically want to force it that each function can only take variables/constants as arguments, instead of more functions, here:

model = SRRegressor(
    binary_operators=[+,-,*,/], 
    unary_operators=[exp,tanh,erf,log10], 
    elementwise_loss=L1DistLoss(),
    batching=true,
    batch_size=200,
    turbo=true,
    constraints=[(^)=>(-1, 1), tanh=>5, erf=>5, exp=>5, log10=>5],
    complexity_of_variables=[3,1,1],
    complexity_of_constants=1,
    nested_constraints= [exp => [exp => 0, tanh => 0, erf => 0], tanh => [exp => 0, tanh => 0, erf => 0],
                        erf => [exp => 0, tanh => 0, erf => 0]],
    dimensional_constraint_penalty=0,
    maxsize=20,
    populations=50,
    population_size=100,
    niterations=1000,
    parsimony=0.0001,
    bumper=false,
    output_directory="fits/$param",
    ncycles_per_iteration=500,
);

In my hall of fame, it keeps optimizing to tanh(exp(x₃)) * x₁. It doesn't look like any other functions are improperly nested, just the exp appearing inside the tanh. Am I doing something wrong?

Cheers,
Liza

Version

1.5.2

Operating System

macOS

Interface

Julia REPL, Script (i.e., python my_script.py)

Relevant log output

Extra Info

No response