def testMaintainsOrder(self):
b = reservoir._ReservoirBucket(100)
for i in xrange(10000):
b.AddItem(i)
items = b.Items()
prev = -1
for item in items:
self.assertTrue(item > prev)
prev = item
def testMaintainsOrder(self):
b = reservoir._ReservoirBucket(100)
for i in xrange(10000):
b.AddItem(i)
items = b.Items()
prev = -1
for item in items:
self.assertTrue(item > prev)
prev = item
def testRemovesItems(self):
b = reservoir._ReservoirBucket(100)
for i in xrange(10):
b.AddItem(i)
self.assertEqual(len(b.Items()), 10)
self.assertEqual(b._num_items_seen, 10)
self.assertEqual(b.FilterItems(lambda x: x <= 7), 2)
self.assertEqual(len(b.Items()), 8)
self.assertEqual(b._num_items_seen, 8)
def testRemovesItems(self):
b = reservoir._ReservoirBucket(100)
for i in xrange(10):
b.AddItem(i)
self.assertEqual(len(b.Items()), 10)
self.assertEqual(b._num_items_seen, 10)
self.assertEqual(b.FilterItems(lambda x: x <= 7), 2)
self.assertEqual(len(b.Items()), 8)
self.assertEqual(b._num_items_seen, 8)
def testRemovesItemsWhenItemsAreReplaced(self):
b = reservoir._ReservoirBucket(100)
for i in xrange(10000):
b.AddItem(i)
self.assertEqual(b._num_items_seen, 10000)
# Remove items
num_removed = b.FilterItems(lambda x: x <= 7)
self.assertGreater(num_removed, 92)
self.assertEqual([], [item for item in b.Items() if item > 7])
self.assertEqual(b._num_items_seen,
int(round(10000 * (1 - float(num_removed) / 100))))
def testRemovesItemsWhenItemsAreReplaced(self):
b = reservoir._ReservoirBucket(100)
for i in xrange(10000):
b.AddItem(i)
self.assertEqual(b._num_items_seen, 10000)
# Remove items
num_removed = b.FilterItems(lambda x: x <= 7)
self.assertGreater(num_removed, 92)
self.assertEqual([], [item for item in b.Items() if item > 7])
self.assertEqual(b._num_items_seen,
int(round(10000 * (1 - float(num_removed) / 100))))
def testLazyFunctionEvaluationAndAlwaysKeepLast(self):
class FakeRandom(object):
def randint(self, a, b): # pylint:disable=unused-argument
return 999
class Incrementer(object):
def __init__(self):
self.n = 0
def increment_and_double(self, x):
self.n += 1
return x * 2
# We've mocked the randomness generator, so that once it is full, the last
# item will never get durable reservoir inclusion. Since always_keep_last is
# false, the function should only get invoked 100 times while filling up
# the reservoir. This laziness property is an essential performance
# optimization.
b = reservoir._ReservoirBucket(100,
FakeRandom(),
always_keep_last=False)
incrementer = Incrementer()
for i in range(1000):
b.AddItem(i, incrementer.increment_and_double)
self.assertEqual(incrementer.n, 100)
self.assertEqual(b.Items(), [x * 2 for x in range(100)])
# This time, we will always keep the last item, meaning that the function
# should get invoked once for every item we add.
b = reservoir._ReservoirBucket(100,
FakeRandom(),
always_keep_last=True)
incrementer = Incrementer()
for i in range(1000):
b.AddItem(i, incrementer.increment_and_double)
self.assertEqual(incrementer.n, 1000)
self.assertEqual(b.Items(), [x * 2 for x in range(99)] + [999 * 2])
def testLazyFunctionEvaluationAndAlwaysKeepLast(self):
class FakeRandom(object):
def randint(self, a, b): # pylint:disable=unused-argument
return 999
class Incrementer(object):
def __init__(self):
self.n = 0
def increment_and_double(self, x):
self.n += 1
return x * 2
# We've mocked the randomness generator, so that once it is full, the last
# item will never get durable reservoir inclusion. Since always_keep_last is
# false, the function should only get invoked 100 times while filling up
# the reservoir. This laziness property is an essential performance
# optimization.
b = reservoir._ReservoirBucket(100, FakeRandom(), always_keep_last=False)
incrementer = Incrementer()
for i in xrange(1000):
b.AddItem(i, incrementer.increment_and_double)
self.assertEqual(incrementer.n, 100)
self.assertEqual(b.Items(), [x * 2 for x in xrange(100)])
# This time, we will always keep the last item, meaning that the function
# should get invoked once for every item we add.
b = reservoir._ReservoirBucket(100, FakeRandom(), always_keep_last=True)
incrementer = Incrementer()
for i in xrange(1000):
b.AddItem(i, incrementer.increment_and_double)
self.assertEqual(incrementer.n, 1000)
self.assertEqual(b.Items(), [x * 2 for x in xrange(99)] + [999 * 2])
def testBucketReservoirSamplingViaStatisticalProperties(self):
# Not related to a 'ReservoirBucket', but instead number of buckets we put
# samples into for testing the shape of the distribution
b = reservoir._ReservoirBucket(_max_size=self.samples)
# add one extra item because we always keep the most recent item, which
# would skew the distribution; we can just slice it off the end instead.
for i in xrange(self.total + 1):
b.AddItem(i)
divbins = [0] * self.n_buckets
modbins = [0] * self.n_buckets
# Slice off the last item when we iterate.
for item in b.Items()[0:-1]:
divbins[item // self.total_per_bucket] += 1
modbins[item % self.n_buckets] += 1
for bucket_index in xrange(self.n_buckets):
divbin = divbins[bucket_index]
modbin = modbins[bucket_index]
self.AssertBinomialQuantity(divbin)
self.AssertBinomialQuantity(modbin)
def testBucketReservoirSamplingViaStatisticalProperties(self):
# Not related to a 'ReservoirBucket', but instead number of buckets we put
# samples into for testing the shape of the distribution
b = reservoir._ReservoirBucket(_max_size=self.samples)
# add one extra item because we always keep the most recent item, which
# would skew the distribution; we can just slice it off the end instead.
for i in xrange(self.total + 1):
b.AddItem(i)
divbins = [0] * self.n_buckets
modbins = [0] * self.n_buckets
# Slice off the last item when we iterate.
for item in b.Items()[0:-1]:
divbins[item // self.total_per_bucket] += 1
modbins[item % self.n_buckets] += 1
for bucket_index in xrange(self.n_buckets):
divbin = divbins[bucket_index]
modbin = modbins[bucket_index]
self.AssertBinomialQuantity(divbin)
self.AssertBinomialQuantity(modbin)
def testEmptyBucket(self): b = reservoir._ReservoirBucket(1) self.assertFalse(b.Items())
def testDoesntOverfill(self):
b = reservoir._ReservoirBucket(10)
for i in xrange(1000):
b.AddItem(i)
self.assertEqual(len(b.Items()), 10)
self.assertEqual(b._num_items_seen, 1000)
def testSizeRequirement(self):
with self.assertRaises(ValueError):
reservoir._ReservoirBucket(-1)
with self.assertRaises(ValueError):
reservoir._ReservoirBucket(10.3)
def testSizeZeroBucket(self):
b = reservoir._ReservoirBucket(0)
for i in xrange(20):
b.AddItem(i)
self.assertEqual(b.Items(), list(range(i + 1)))
self.assertEqual(b._num_items_seen, 20)
def testSizeOneBucket(self):
b = reservoir._ReservoirBucket(1)
for i in xrange(20):
b.AddItem(i)
self.assertEqual(b.Items(), [i])
self.assertEqual(b._num_items_seen, 20)
def testKeepsLatestItem(self):
b = reservoir._ReservoirBucket(5)
for i in xrange(100):
b.AddItem(i)
last = b.Items()[-1]
self.assertEqual(last, i)
def testKeepsLatestItem(self):
b = reservoir._ReservoirBucket(5)
for i in xrange(100):
b.AddItem(i)
last = b.Items()[-1]
self.assertEqual(last, i)
def testDoesntOverfill(self):
b = reservoir._ReservoirBucket(10)
for i in xrange(1000):
b.AddItem(i)
self.assertEqual(len(b.Items()), 10)
self.assertEqual(b._num_items_seen, 1000)
def testFillToSize(self):
b = reservoir._ReservoirBucket(100)
for i in xrange(100):
b.AddItem(i)
self.assertEqual(b.Items(), list(xrange(100)))
self.assertEqual(b._num_items_seen, 100)
def testEmptyBucket(self):
b = reservoir._ReservoirBucket(1)
self.assertFalse(b.Items())
def testSizeOneBucket(self):
b = reservoir._ReservoirBucket(1)
for i in xrange(20):
b.AddItem(i)
self.assertEqual(b.Items(), [i])
self.assertEqual(b._num_items_seen, 20)
def testFillToSize(self):
b = reservoir._ReservoirBucket(100)
for i in xrange(100):
b.AddItem(i)
self.assertEqual(b.Items(), list(xrange(100)))
self.assertEqual(b._num_items_seen, 100)
def testSizeZeroBucket(self):
b = reservoir._ReservoirBucket(0)
for i in xrange(20):
b.AddItem(i)
self.assertEqual(b.Items(), list(range(i + 1)))
self.assertEqual(b._num_items_seen, 20)
def testSizeRequirement(self):
with self.assertRaises(ValueError):
reservoir._ReservoirBucket(-1)
with self.assertRaises(ValueError):
reservoir._ReservoirBucket(10.3)