def unite_recursively(self):
threshold = constants.THRESHOLD_FOR_REMOVING_FROM_PART_OF # 0.2くらい
try:
cluster_a, cluster_b = self._find_clusters_to_be_united()
except ValueError: # []つまり統合すべきタスクなし
return
united_new_cluster = cluster_a.union(cluster_b)
not_shared_tasks = cluster_a.difference(cluster_b).union(
cluster_b.difference(cluster_a))
evaluator = TaskGraphEvaluator(self.graph)
for task_name in not_shared_tasks:
contribution = float(
evaluator.contribution_with_task_name(task_name)[0])
average_of_contribution = evaluator.contribution_with_cluster(
united_new_cluster) / len(united_new_cluster)
if contribution < average_of_contribution * threshold:
print('%sは別ルートの可能性が高いので排除しました' % task_name)
united_new_cluster.remove(task_name)
self.task_distance_pairs[constants.SUPERTYPE_NAME].append(
united_new_cluster)
self.task_distance_pairs[constants.SUPERTYPE_NAME].remove(cluster_a)
self.task_distance_pairs[constants.SUPERTYPE_NAME].remove(cluster_b)
self.unite_recursively()
def _cluster_contribution_url(self, cluster):
evaluator = TaskGraphEvaluator(self.graph)
task_names = {l for l in cluster}
aspects = (self._aspects_with_task_name(task_name) for task_name in task_names)
try:
urls = {aspect[0]['url'] for aspect in aspects}
except IndexError:
urls = set()
result = (cluster, evaluator.contribution_with_cluster(cluster), urls)
return result
def _select_subtype_tasks(self, pairs, subtype):
"""
SUBTYPEがあるときのみ使う
pairs => {('ハウスクリーニング_を_利用する', 25),
('新聞_も_分別する', -33), ('ブランド_を_教える', -18)}
"""
try:
task_names = {pair[0] for pair in pairs}
except TypeError:
pdb.set_trace()
evaluator = TaskGraphEvaluator(self.graph)
result = (task_names, evaluator.contribution_with_cluster(task_names), subtype)
return result
def _select_subtype_tasks(self, pairs, subtype):
"""
SUBTYPEがあるときのみ使う
pairs => {('ハウスクリーニング_を_利用する', 25),
('新聞_も_分別する', -33), ('ブランド_を_教える', -18)}
"""
try:
task_names = {pair[0] for pair in pairs}
except TypeError:
pdb.set_trace()
evaluator = TaskGraphEvaluator(self.graph)
result = (task_names, evaluator.contribution_with_cluster(task_names),
subtype)
return result
def _cluster_contribution_url_intersection(self, cluster):
task_names = {l for l in cluster}
url_set = set()
for task_name in task_names:
aspects = self._aspects_with_task_name(task_name)
urls = {aspect['url'] for aspect in aspects}
if not url_set:
url_set = urls
else:
url_set = url_set.intersection(urls)
evaluator = TaskGraphEvaluator(self.graph)
# cluster => TaskCluster({'', '', ...})
result = (cluster, evaluator.contribution_with_cluster(cluster), url_set)
return result
def _cluster_contribution_url_intersection(self, cluster):
task_names = {l for l in cluster}
url_set = set()
for task_name in task_names:
aspects = self._aspects_with_task_name(task_name)
urls = {aspect['url'] for aspect in aspects}
if not url_set:
url_set = urls
else:
url_set = url_set.intersection(urls)
evaluator = TaskGraphEvaluator(self.graph)
# cluster => TaskCluster({'', '', ...})
result = (cluster, evaluator.contribution_with_cluster(cluster),
url_set)
return result
def unite_recursively(self):
threshold = constants.THRESHOLD_FOR_REMOVING_FROM_PART_OF # 0.2くらい
try:
cluster_a, cluster_b = self._find_clusters_to_be_united()
except ValueError: # []つまり統合すべきタスクなし
return
united_new_cluster = cluster_a.union(cluster_b)
not_shared_tasks = cluster_a.difference(cluster_b).union(cluster_b.difference(cluster_a))
evaluator = TaskGraphEvaluator(self.graph)
for task_name in not_shared_tasks:
contribution = float(evaluator.contribution_with_task_name(task_name)[0])
average_of_contribution = evaluator.contribution_with_cluster(united_new_cluster) / len(united_new_cluster)
if contribution < average_of_contribution * threshold:
print('%sは別ルートの可能性が高いので排除しました' % task_name)
united_new_cluster.remove(task_name)
self.task_distance_pairs[constants.SUPERTYPE_NAME].append(united_new_cluster)
self.task_distance_pairs[constants.SUPERTYPE_NAME].remove(cluster_a)
self.task_distance_pairs[constants.SUPERTYPE_NAME].remove(cluster_b)
self.unite_recursively()