Description
First off: What an awesome library!
I've noticed some problems when trying to solve multi-objective problems though. The underlying problem seems to be the same - some check not being prepared to handle multi-objective problems (e.g. comparison of two numbers, where multi-objective problems would have to compare two lists of numbers)
Main Exception Message:
The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()
First Case: stop_criteria="saturate_{num}"
I tried using stop_criteria="saturate_100" in combination with a multi-objective problem - it fails excactly at generation 100. That holds true for any number {num} - when generation {num} is reached, the stop criterion is checked for the first time and fails because it is not ready to handle multi-objective problems. This is the resulting Exception Message:
Traceback (most recent call last):
File "C:\path\to\pygad\pygad.py", line 2143, in run
if (self.best_solutions_fitness[self.generations_completed - criterion[1]] - self.best_solutions_fitness[self.generations_completed - 1]) == 0:
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()
Second Case: mutation_type="adaptive"
Adaptive Mutation already fails on the first generation, but for the same underlying reason. When offspring_fitness is first evaluated, this Exception is thrown:
Traceback (most recent call last):
File "C:\path\to\pygad\pygad.py", line 2062, in run
self.last_generation_offspring_mutation = self.mutation(self.last_generation_offspring_crossover)
File "C:\path\to\pygad\utils\mutation.py", line 556, in adaptive_mutation
offspring = self.adaptive_mutation_probs_by_space(offspring)
File "C:\path\to\pygad\utils\mutation.py", line 794, in adaptive_mutation_probs_by_space
if offspring_fitness[offspring_idx] < average_fitness:
ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()
There may be more instances of the same Exception popping up at other places throughout PyGAD, I haven't really searched. Would be great to have this fixed, so adaptive mutation and other great features become usable with multi-objective problems.
Cheers, Matthias