def test_jaccard_integers(data):
set1 = data.draw(st.sets(elements=st.integers(), average_size=100, min_size=1))
overlap_size = data.draw(st.integers(min_value=0, max_value=len(set1)))
if set1:
set2 = set(random.sample(set1, overlap_size))
else:
set2 = set([])
set2 = set2.union(data.draw(st.sets(elements=st.integers(), average_size=50, min_size=1)))
score = len(set1.intersection(set2))/len(set1.union(set2))*1.0
note("{}, {}".format(embed.jaccard(set1, set2), score))
assert(embed.jaccard(set1, set2) == score)
assert(score <= 1)
assert(score >= 0)
def test_pairs_list_similarity(context, label, data, metric):
list1 = data.draw(st.lists(elements=st.integers(), unique=True, average_size=100))
# check score with myself is 1
if list1:
_, (_, self_score) = embed.pairs_list_similarity(context, label, list1, list1, how=metric)
note("{} score of list with itself: {}".format(metric, self_score))
note(list1)
assert(self_score == 1)
# check score is always between 0 and 1
overlap_size = data.draw(st.integers(min_value=0, max_value=len(list1)))
if list1 and overlap_size:
list2 = random.sample(list1, overlap_size)
min_value = max(list1) + 1
else:
list2 = []
min_value = None
list2 = list2 + data.draw(st.lists(elements=st.integers(min_value=min_value), unique=True, average_size=50))
ret_context, (ret_label, score) = embed.pairs_list_similarity(context, label, list1, list2, how=metric)
assert(label == ret_label)
assert(context == ret_context)
assert(score <= 1)
assert(score >= 0)
# check that we computed the correct score
if metric == 'jaccard':
assert(score == embed.jaccard(set(list1), set(list2)))
elif metric == 'adapted-ktau':
assert(score == embed.adaptedKendallTau(list1, list2))
def test_akt_vs_jaccard(data):
list1 = data.draw(st.lists(elements=st.integers(), unique=True, average_size=100))
# top elements from list 1 are the ones that that overlap with list 2, akt >=jac
# only overlap on up to %90 of the elements
overlap_size = data.draw(st.integers(min_value=0, max_value=int(len(list1)*.8)))
if list1 and overlap_size:
list2 = list1[:overlap_size]
random.shuffle(list2)
min_value = max(list1) + 1
elif not list1:
list2 = []
min_value = None
else:
list2 = []
min_value = max(list1) + 1
list2 = list2 + data.draw(st.lists(elements=st.integers(min_value=min_value), unique=True, average_size=50))
note((list1, list2, list1==list2))
jac_score = embed.jaccard(set(list1), set(list2))
akt_score = embed.adaptedKendallTau(list1, list2)
if (overlap_size > 0):
assert(akt_score >= jac_score)
else:
assert(akt_score <= jac_score)
# check that shuffling elements in the same list results in an akt_score of <= 1 and jac score of 1
list3 = list2
random.shuffle(list3)
jac_score = embed.jaccard(set(list3), set(list2))
akt_score = embed.adaptedKendallTau(list3, list2)
assert(akt_score <= jac_score)
if list2:
assert(jac_score == 1)
else:
assert(jac_score == 0)