class topK_max:
'''
Given a stream of (key,value,ridealong) tuples, remember the k largest values.
If a key is added repeatedly, use the largest value.
'''
def __init__(self, size):
self.size = size
self.element = namedtuple('topK_max_element', ['value', 'ridealong'])
self.d = ItemSortedDict(getvaluevalue)
def add(self, key, value, ridealong):
if key in self.d:
if value >= self.d[key].value:
self.d[key] = self.element(value, ridealong)
elif len(self.d) < self.size:
self.d[key] = self.element(value, ridealong)
elif value > self.d.peekitem()[1].value:
self.d.popitem()
self.d[key] = self.element(value, ridealong)
def readout(self):
return [(k, list(v)) for k, v in self.d.iteritems()]
class topK_sum:
'''
Space-saving heavy hitters.
Given a stream of (key, value) tuples, rememgber the k items
with the largest sum of values.
http://www.cse.ust.hk/~raywong/comp5331/References/EfficientComputationOfFrequentAndTop-kElementsInDataStreams.pdf
'''
def __init__(self, size):
self.size = size
self.element = namedtuple('topK_max_element',
['value', 'ridealong', 'fake'])
self.d = ItemSortedDict(getvaluevalue)
def add(self, key, value, ridealong):
if key in self.d:
self.d[key] = self.element(self.d[key].value + value, ridealong,
self.d[key].fake)
elif len(self.d) < self.size:
self.d[key] = self.element(value, ridealong, 0)
elif value >= self.d.peekitem()[1].value:
self.d.popitem()
self.d[key] = self.element(value, ridealong, 0)
else:
evicted = self.d.popitem()
oldvalue = evicted[1].value
newvalue = max(value, oldvalue + 1)
fake = max(newvalue - value, 0)
self.d[key] = self.element(newvalue, ridealong, fake)
def readout(self):
ret = []
for i in self.d.iteritems():
if i[1].value > 2 * i[1].fake:
ret.append((i[0], [i[1].value, i[1].ridealong]))
return ret