def test_correct_despite_outliers(self):
SAMPLE_COUNT = 10
fine = [[o] for o in self.row_ids]
coarse = [[o for o in self.row_ids]]
groupings = [fine, coarse] + [self.grouping] * (SAMPLE_COUNT - 2)
grouping = find_consensus_grouping(groupings)
self._assert_correct(grouping)
def test_simple(self):
groupings = self.sample_groupings()
grouping = find_consensus_grouping(groupings, debug=True)
assert isinstance(grouping, list)
for row in grouping:
assert isinstance(row, loom.group.Row), row
row_ids = set(row.row_id for row in grouping)
assert len(row_ids) == len(grouping), 'grouping had duplicate rows'
assert_set_equal(set(self.row_ids), row_ids)
group_ids = sorted(list(set(row.group_id for row in grouping)))
assert_equal(
group_ids,
range(len(group_ids)),
'group ids were not a contiguous range of integers')
def test_sorting(self):
for i in xrange(10):
groupings = self.sample_groupings()
grouping = find_consensus_grouping(groupings, debug=True)
assert_equal(
grouping,
sorted(
grouping,
key=lambda x: (x.group_id, -x.confidence, x.row_id)))
group_ids = sorted(set(row.group_id for row in grouping))
counts = [
sum(1 for row in grouping if row.group_id == gid)
for gid in group_ids
]
assert_equal(counts, sorted(counts, reverse=True))
def test_correct_on_noisy_data(self):
SAMPLE_COUNT = 10
GROUP_COUNT = len(self.grouping)
object_index = {
o: g
for g, group in enumerate(self.grouping)
for o in group
}
# each object is in the wrong place in one grouping
groupings = []
for g in range(SAMPLE_COUNT):
groups = self.grouping
for o in self.row_ids[g::SAMPLE_COUNT]:
t = object_index[o]
f = (t + 1) % GROUP_COUNT
groups[t].remove(o)
groups[f].append(o)
groups = filter(len, groups)
groupings.append(groups)
grouping = find_consensus_grouping(groupings)
self._assert_correct(grouping)
def test_correct_on_perfect_data(self):
for sample_count in range(1, 11):
groupings = [self.grouping] * sample_count
grouping = find_consensus_grouping(groupings)
self._assert_correct(grouping, confidence=1.0)
