def test_sample_string_for_tree():
t = nx.DiGraph([(0, 1), (1, 2), (2, 3)])
nomi_samples = [
immutably(sample_string_for_tree)(t, [2]) for _ in range(1000)
]
mi_samples = [ # MI 1
immutably(sample_string_for_tree)(t, [0, 1]) for _ in range(1000)
]
# When self-information is high, samples become sparser, so more samples
# are needed to get a distribution with the right MI.
mi_samples2 = [ # MI 1
immutably(sample_string_for_tree)(t, [2, 1]) for _ in range(10000)
]
mi_samples3 = [ # MI 2
immutably(sample_string_for_tree)(t, [0, 2]) for _ in range(1000)
]
assert len(set(mi_samples)) < len(set(nomi_samples))
def is_close(x, y, tol):
return abs(x - y) < tol
import rfutils.entropy
nomi_mi = rfutils.entropy.mutual_information(
Counter(rfutils.flatmap(lambda s: rfutils.sliding(s, 2),
nomi_samples)))
assert is_close(nomi_mi, 0, 0.01)
mi_mi = rfutils.entropy.mutual_information(
Counter(rfutils.flatmap(lambda s: rfutils.sliding(s, 2), mi_samples)))
assert is_close(mi_mi, 1, 0.01)
mi2_mi = rfutils.entropy.mutual_information(
Counter(rfutils.flatmap(lambda s: rfutils.sliding(s, 2), mi_samples2)))
assert is_close(mi2_mi, 1, 0.01)
mi3_mi = rfutils.entropy.mutual_information(
Counter(rfutils.flatmap(lambda s: rfutils.sliding(s, 2), mi_samples3)))
assert is_close(mi3_mi, 2, 0.1)
def is_monotonic(cmp, xs):
try:
return all(cmp(x, y) for x, y in sliding(xs, 2))
except StopIteration: # TODO fix sliding so this doesn't need to be special
return True
def skipgrams(xs, k):
for gram in sliding(xs, k + 2):
yield gram[0], gram[-1]
def is_monotonic(cmp, xs):
try:
return all(cmp(x, y) for x, y in sliding(xs, 2))
except StopIteration: # TODO fix sliding so this doesn't need to be special
return True