def mutateLearners(self, team):
# Since we will remove Learners from the Team while iterating over them,
# grab a copy of the list of Learner references. We will iterate through
# the copy while mutating the original list
learners = team.learners.copy()
# Probabalistically mutate Learners in the Team
for learner in learners:
# Most Learners will be skipped and not mutated
if not weightedCoinFlip(Trainer.P_MUT_LEARNER):
continue
# Sanity check
if team.countAtomicActions() == 0:
print(
"WARNING - Trainer::mutateLearners - No atomic actions in Team!"
)
# If this Team only has one Learner with an atomic action and it
# happens to be this Learner, remove the possibility of setting the
# new action to a Team pointer
p_atomic = Trainer.P_ATOMIC
if learner.isActionAtomic() and team.countAtomicActions() == 1:
p_atomic = 1.0
# Copy Learner. The copy will be mutated and added to the pool of
# Learners. This is done to ensure that, in the event that this Learner
# is also used to great success in another Team, that other Team
# still points to the original functional Learner.
learner_prime = Learner(learner=learner)
# Remove original Learner from Team and replace it with the new one.
team.removeLearner(learner)
team.addLearner(learner_prime)
# Mutate the new Learner
learner_prime.mutateProgram()
# Probabalistically mutate the new Learner's action
if weightedCoinFlip(Trainer.P_MUT_LEARNER_ACTION):
self.mutateLearnerAction(learner_prime, p_atomic)
# Sanity check
if team.countAtomicActions() == 0:
print(
"WARNING - Trainer::mutateLearners - No atomic actions in Team after mutation!"
)
# Add new Learner to the pool of Learners
self.learner_pop.append(learner_prime)
