def test_sgt_dist(self):
"""
Test for the Simple Good-Turing distribution.
"""
# Only run the test if the numpy and scipy libraries (imported
# by the function) are installed; an ImportError will not be a
# test failure as we decided that they should not be dependencies.
try:
# The first distribution does not have enough data for SGT,
# assert that an assertion is raised.
with pytest.raises(RuntimeWarning):
smooth_dist(self.observ1, "sgt")
# The second distribution also does not have enough data for
# a confident results, but it does have enough to stress-test the
# method, so we are not going to allow it to fail (i.e., raise
# exceptions)
seen_p05, unseen_p05 = smooth_dist(self.observ2,
"sgt",
allow_fail=False)
assert seen_p05['x'] == seen_p05['y']
assert seen_p05['x'] < seen_p05['z']
assert seen_p05['0'] > unseen_p05
except ImportError: # pragma: no cover
pass
def test_laplace_dist(self):
"""
Test for the Laplace distribution.
"""
seen, unseen = smooth_dist(self.observ1, "laplace")
assert seen['A'] < seen['B']
assert seen['D'] == seen['E']
assert seen['A'] > unseen
seen, unseen = smooth_dist(self.observ2, "laplace")
assert seen['x'] == seen['y']
assert seen['x'] < seen['z']
assert seen['0'] > unseen
def test_ele_dist(self):
"""
Test for the Expected-Likelihood estimation distribution.
"""
seen, unseen = smooth_dist(self.observ1, "ele")
assert seen['A'] < seen['B']
assert seen['D'] == seen['E']
assert seen['A'] > unseen
seen, unseen = smooth_dist(self.observ2, "ele")
assert seen['x'] == seen['y']
assert seen['x'] < seen['z']
assert seen['0'] > unseen
def test_uniform_dist(self):
"""
Test for the uniform distribution.
"""
# Easiest distribution to test: everything must be equal, unobserved
# must be less than any observed one.
seen, unseen = smooth_dist(self.observ1, "uniform")
assert seen['A'] == seen['E']
assert seen['A'] > unseen
seen, unseen = smooth_dist(self.observ2, "uniform")
assert seen['0'] == seen['~']
assert seen['x'] == seen['z']
assert seen['0'] > unseen
def test_mle_dist(self):
"""
Test for the Maximum-Likelihood Estimation distribution.
"""
# This is easy to test as the results of the MLE are the ones we intuitively
# expect/compute.
seen, unseen = smooth_dist(self.observ1, "mle")
assert seen['A'] < seen['B']
assert seen['D'] == seen['E']
assert seen['A'] > unseen
seen, unseen = smooth_dist(self.observ2, "mle")
assert seen['x'] == seen['y']
assert seen['x'] < seen['z']
assert seen['0'] > unseen
def test_wittenbell_dist(self):
"""
Test for the Witten-Bell distribution.
"""
seen10, unseen10 = smooth_dist(self.observ1, "wittenbell", bins=10)
seen99, unseen99 = smooth_dist(self.observ1, "wittenbell", bins=99)
assert seen10['A'] < seen10['B']
assert seen10['D'] == seen10['E']
assert seen10['A'] == unseen10
assert seen10['A'] == seen99['A']
assert unseen99 < unseen10
seen, unseen = smooth_dist(self.observ2, "wittenbell")
assert seen['x'] == seen['y']
assert seen['x'] < seen['z']
assert seen['0'] > unseen
def test_random_dist(self):
"""
Test for the random distribution.
"""
# Here, we can essentially only test if there are no execution bugs,
# besides the unobserved probability being less than any other. Nonetheless,
# we set a seed for comparing the results across platforms.
seen, unseen = smooth_dist(self.observ1, "random", seed=1305)
assert seen['A'] < seen['E']
assert seen['B'] < seen['D']
assert seen['A'] < seen['C']
assert seen['A'] > unseen
seen, unseen = smooth_dist(self.observ2, "random", seed=1305)
assert seen['B1'] < seen['4']
assert seen['F5'] < seen['C8']
assert seen['0'] > unseen
def test_certaintydegree_dist(self):
"""
Test for the Degree of Certainty distribution.
"""
seen, unseen = smooth_dist(self.observ1, 'certaintydegree')
seen99, unseen99 = smooth_dist(self.observ1, 'certaintydegree', bins=99)
assert seen['A'] < seen['B']
assert seen['D'] == seen['E']
assert seen['A'] < unseen
assert seen['B'] > unseen
assert seen['A'] < seen99['A']
assert unseen > unseen99
seen, unseen = smooth_dist(self.observ2, "certaintydegree")
assert seen['x'] == seen['y']
assert seen['x'] < seen['z']
assert seen['0'] > unseen