def _merge_clusters(scheme, O, S, add_random=True):
"""Merge bad clusters into 1"""
clusters = construct_clusters(scheme, S)
new_clusters = []
while clusters:
p = random.randint(0, 1)
updated_cluster = clusters.pop(0)
if not add_random or p:
# randomly decide if want to merge curr cluster
# always expect split if len(clusters) == 1
cluster_length = len(clusters)
for _ in range(cluster_length):
if not clusters:
break
next_cluster = clusters.pop(0)
merged = deepcopy(updated_cluster)
merged.merge(next_cluster)
# if merged cluster is worse than split clusters, do not merge
# else, merge
if merged.value <= updated_cluster.value + next_cluster.value:
clusters.append(next_cluster)
else:
updated_cluster = merged
new_clusters.append(updated_cluster)
# fix ids
for i in range(len(new_clusters)):
new_clusters[i].id = i
new_S = Solution.from_clusters(scheme, new_clusters)
if satisfies_constraints(scheme, new_S):
S = new_S
else: # shouldn't happen
raise ValueError('infeasible solution in split')
return S
def _move_machines(scheme, O, S):
"""Perform movement of machines between clusters within a solution"""
clusters = construct_clusters(scheme, S)
rated_machines = _find_best_fit_for_machines(scheme, clusters)
while rated_machines:
stat = rated_machines.pop(0)
curr_id, new_id = stat.curr_cluster, stat.new_cluster
if clusters[curr_id].near_empty:
# can't move anything out of near empty cluster
continue
curr_cluster = deepcopy(clusters[curr_id])
new_cluster = deepcopy(clusters[new_id])
curr_cluster.machines, new_cluster.machines = _move(
stat.machine, curr_cluster.machines, new_cluster.machines)
old_value = clusters[curr_id].value + clusters[new_id].value
new_value = curr_cluster.value + new_cluster.value
# if new clusters have better objective than old ones, approve move
if new_value > old_value:
clusters[curr_id] = curr_cluster
clusters[new_id] = new_cluster
rated_machines = _find_best_fit_for_machines(scheme, clusters)
# construct new S
# if new S is better and satisfies constraints, approve changes
new_S = Solution.from_clusters(scheme, clusters)
if O(scheme, new_S) > O(scheme, S) and satisfies_constraints(
scheme, new_S):
S = new_S
return S
def _split_clusters(scheme, O, S, add_random=True):
"""Split bad clusters in solution"""
clusters = construct_clusters(scheme, S)
new_clusters = []
for cluster in clusters:
p = random.randint(0, 1)
if not cluster.can_split:
# copy "as is" if can't split the cluster
new_clusters.append(cluster)
continue
if add_random and len(clusters) > 1 and p:
# randomly decide if want to split curr cluster
# always expect split if len(clusters) == 1
new_clusters.append(cluster)
continue
new_clusters += _split(scheme, cluster)
# fix ids:
for i in range(len(new_clusters)):
new_clusters[i].id = i
new_S = Solution.from_clusters(scheme, new_clusters)
if satisfies_constraints(scheme, new_S):
S = new_S
else: # shouldn't happen
raise ValueError('infeasible solution in split')
return S