def test_merge_equal():
parameters = [(35, 1), (1, 3), (15, 2), (40, 0.5)]
for n in test_sizes:
d = EmptyDataset(0)
s = GKArray(test_eps)
for params in parameters:
generator = Normal.from_params(params[0], params[1], n)
sketch = GKArray(test_eps)
for v in generator.data:
sketch.add(v)
d.add(v)
s.merge(sketch)
evaluate_sketch_accuracy(s, d, 2*s.eps)
def test_merge_mixed():
ntests = 20
datasets = [Normal, Exponential, Laplace, Bimodal]
for i in range(ntests):
d = EmptyDataset(0)
s = GKArray(test_eps)
for dataset in datasets:
generator = dataset(np.random.randint(0, 500))
sketch = GKArray(test_eps)
for v in generator.data:
sketch.add(v)
d.add(v)
s.merge(sketch)
evaluate_sketch_accuracy(s, d, 2*s.eps)
def test_merge_unequal():
ntests = 20
for i in range(ntests):
for n in test_sizes:
d = Lognormal(n)
s1 = GKArray(test_eps)
s2 = GKArray(test_eps)
for v in d.data:
if np.random.random() > 0.7:
s1.add(v)
else:
s2.add(v)
s1.merge(s2)
evaluate_sketch_accuracy(s1, d, 2*s1.eps)
def test_consistent_merge():
""" Test that merge() calls do not modify the argument sketch.
"""
s1 = GKArray(test_eps)
s2 = GKArray(test_eps)
d = Normal(100)
for v in d.data:
s1.add(v)
s1.merge(s2)
# s2 is still empty
np.testing.assert_equal(s2.num_values, 0)
d = Normal(50)
for v in d.data:
s2.add(v)
s2_summary = [s2.quantile(q) for q in test_quantiles]+ [s2.sum, s2.avg, s2.num_values]
s1.merge(s2)
d = Normal(10)
for v in d.data:
s1.add(v)
# changes to s1 does not affect s2 after merge
s2_summary = [s2.quantile(q) for q in test_quantiles] + [s2.sum, s2.avg, s2.num_values]
np.testing.assert_almost_equal([s2.quantile(q) for q in test_quantiles] + [s2.sum, s2.avg, s2.num_values],
s2_summary)
s3 = GKArray(test_eps)
s3.merge(s2)
# merging to an empty sketch does not change s2
np.testing.assert_almost_equal([s2.quantile(q) for q in test_quantiles] + [s2.sum, s2.avg, s2.num_values],
s2_summary)