Fix adding mutation as parent operator (#262)

* Fix bandit

* Review fixes

* Review fixes

* Fix action name

---------

Co-authored-by: Sergey Kasyanov <[email protected]>

golem/core/optimisers/adaptive/mab_agents/contextual_mab_agent.py

@@ -85,7 +85,7 @@ def partial_fit(self, experience: ExperienceBuffer):
     def _get_experience(self, experience: ExperienceBuffer):
         """ Get experience from ExperienceBuffer, process rewards and log. """
         obs, actions, rewards = experience.retrieve_experience()
-        arms = [self._arm_by_action[action] for action in actions]
+        arms = [self._arm_by_action[action.__name__] for action in actions]
         # there is no need to process rewards as in MAB, since this processing unifies rewards for all contexts
         self._dbg_log(obs, actions, rewards)
         return obs, arms, rewards

golem/core/optimisers/adaptive/mab_agents/mab_agent.py

@@ -2,7 +2,8 @@
 import _pickle as pickle
 import random
 import re
-from typing import Union, Sequence, Optional
+from functools import partial
+from typing import Union, Sequence, Optional, Callable
 
 from mabwiser.mab import MAB, LearningPolicy
 from scipy.special import softmax
@@ -25,7 +26,8 @@ def __init__(self,
         super().__init__(actions=actions, enable_logging=enable_logging)
         self.actions = list(actions)
         self._indices = list(range(len(actions)))
-        self._arm_by_action = dict(zip(actions, self._indices))
+        # str because parent operator for mutation is stored as string for custom mutations serialisation
+        self._arm_by_action = dict(map(lambda x, y: (self._get_callable_name(x), y), actions, self._indices))
         self._agent = MAB(arms=self._indices,
                           learning_policy=LearningPolicy.EpsilonGreedy(epsilon=0.4),
                           n_jobs=n_jobs)
@@ -34,6 +36,16 @@ def __init__(self,
             self._initial_fit()
         self._path_to_save = path_to_save
 
+    @staticmethod
+    def _get_callable_name(action: Callable):
+        if isinstance(action, partial):
+            return action.func.__name__
+        else:
+            try:
+                return action.__name__
+            except AttributeError:
+                return str(action)
+
     def _initial_fit(self):
         n = len(self.actions)
         uniform_rewards = [1. / n] * n

golem/core/optimisers/genetic/operators/base_mutations.py

@@ -39,6 +39,10 @@ class MutationTypesEnum(Enum):
 
     none = 'none'
 
+    @property
+    def __name__(self):
+        return self.name
+
 
 def get_mutation_prob(mut_id: MutationStrengthEnum, node: Optional[GraphNode],
                       default_mutation_prob: float = 0.7) -> float:

golem/core/optimisers/genetic/operators/mutation.py

@@ -103,8 +103,9 @@ def _mutation(self, individual: Individual) -> Tuple[Individual, bool]:
                 new_graph = self._apply_mutations(new_graph, mutation_type)
                 is_correct_graph = self.graph_generation_params.verifier(new_graph)
                 if is_correct_graph:
+                    # str for custom mutations serialisation
                     parent_operator = ParentOperator(type_='mutation',
-                                                     operators=mutation_type,
+                                                     operators=mutation_type.__name__,
                                                      parent_individuals=individual)
                     individual = Individual(new_graph, parent_operator,
                                             metadata=self.requirements.static_individual_metadata)