def process(self, new_features):
for feature in new_features:
self.insert_feature(feature)
# update meta-object
meta = Meta(feature.name, self.db)
meta.learn_from_attributes(feature)
meta.save()
features = self.db.features.find({"name": feature.name})
if(features.count() % meta.get_merge_frequency() == 0):
logging.debug("Summarizing %s" % feature.name)
# insert new features as "characteristics"
self.update_characteristics(feature.name, meta.get_merge_frequency())
# merge characteristics
while True:
changed = self.merge_characteristics(feature.name, meta.get_merge_threshold())
if not changed: break
return True
def merge_characteristics(self, name, merge_threshold): # returns if one or more characteristics were merged (boolean)
logging.debug("Merging characterics")
characteristics = self.db.characteristics.find({"name": name})
meta = Meta(name, self.db)
chars = list(characteristics)
# find "best fit" (other characteristic with minimal distance) for each characteristic
best_fits = [None]*len(chars) # list of tuples: (index of best fit, distance)
best_distance = 1.
for first_i, first in enumerate(chars):
for second_i, second in enumerate(chars):
if (first['_id'] is not second['_id']) and (idents_disjoint(first['ident'], second['ident'])):
distance = Feature.from_db(first).distance_to(Feature.from_db(second), meta.get_attr_ranges())
if (not best_fits[first_i]) or (distance < best_fits[first_i][2]):
best_fits[first_i] = (first_i, second_i, distance)
if distance < best_distance: best_distance = distance
if best_distance > (1-merge_threshold): return False # signalize that no characteristic needed to be merged
for bf in best_fits:
if not bf: continue # continue if first has been merged before
if not best_fits[bf[1]]: continue # continue if second has been merged before
first = Feature.from_db(chars[bf[0]])
second = Feature.from_db(chars[bf[1]])
first.merge(second)
self.db.characteristics.save(first.db_entry())
self.db.characteristics.remove({"_id": second._id})
best_fits[bf[0]] = None
best_fits[bf[1]] = None
# TODO recalculate best_fits with same bf[0]
return True # signalize that one or more characteristics were merged