Bug/missing exception handling: sampling returns nan equations

[lfrommelt]

To replicate:

import equation_tree
from equation_tree.prior import priors_from_space, structure_prior_from_max_depth, structure_prior_from_depth
import numpy as np

max_vars =  2

prior={'structures': {'[0, 1, 2, 3]': 0.25, 
                '[0, 1, 2, 2]': 0.25, 
                '[0, 1, 2, 1]': 0.25, 
                '[0, 1, 1, 2]': 0.25}, 
 'functions': {'squared': 1.0}, 
 'operators': {'+': 0.25, 
               '-': 0.25, 
               '/': 0.25, 
               '*': 0.25}, 
 'features': {'constants': 0.3, 
              'variables': 0.7}}

np.random.seed(864)
equation = equation_tree.sample(
            n=1, prior=prior, max_num_variables=max_vars
        )

print(equation)

np.random.seed(556)
equation = equation_tree.sample(
            n=1, prior=prior, max_num_variables=max_vars
        )

print(equation)

will output:

Processing: 100%|█████████████████████████████████████████████████████████████████| 1/1 [00:00<00:00, 45.17iteration/s]

[(c_1 + x_1)**2]

Processing: 100%|█████████████████████████████████████████████████████████████████| 1/1 [00:31<00:00, 31.80s/iteration]

[None]

So it seems the prior in general is not inconsistent, but a sampled structure + other spaces might be (not "enough leaves", maybe?). The much longer time that the failed run takes, makes it seem that some loop runs for a defined amount of iterations and, if not sucessfull, returns nan instead of raising an exception.