def testHistProbabilisticInterleave(self):
r1 = ProbabilisticRankingFunction(3, self.weights_1)
r2 = ProbabilisticRankingFunction(3, self.weights_2)
hpi = HistProbabilisticInterleave(None)
a = ([0, 1, 1, 0], r1, r2)
self.assertEqual(hpi.infer_outcome([0, 1, 2, 3], a, [0, 0, 0, 0],
r1, r2, self.query), 0, "No clicks, outcome should be 0.")
o = hpi.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r1, r2,
self.query)
self.assertAlmostEquals(o, -0.0486, 4, "Same target as original "
"rankers. Ranker 1 should win (o = %.4f)." % o)
o = hpi.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r2, r1,
self.query)
self.assertAlmostEquals(o, 0.0486, 4, "Target rankers switched. "
"Ranker 2 should win (o = %.4f)." % o)
test_r1 = ProbabilisticRankingFunction(3, self.weights_1)
a = ([0, 1, 1, 0], r1, test_r1)
o = hpi.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r2, r1,
self.query)
self.assertAlmostEquals(o, 0.1542, 4, "Same original ranker. "
"Ranker 2 should win (o = %.4f)." % o)
def testHistProbabilisticInterleaveWithoutMarginalization(self):
r1 = ProbabilisticRankingFunction(3, self.weights_1)
r2 = ProbabilisticRankingFunction(3, self.weights_2)
hpiIs = HistProbabilisticInterleave("--biased False "
"--marginalize False")
# test get_probability_of_list_and_assignment
p = hpiIs._get_probability_of_list_and_assignment([1, 3, 2, 0],
[0, 0, 0, 0], r1, r2,
self.query)
self.assertAlmostEqual(p, 0.026261, 6, "Most likely list for ranker 1."
" p = %e" % p)
# test overall outcomes
a = ([0, 1, 1, 0], r1, r2)
self.assertEqual(
hpiIs.infer_outcome([0, 1, 2, 3], a, [0, 0, 0, 0], r1, r2,
self.query), 0,
"No clicks, outcome should be 0.")
o = hpiIs.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r1, r2,
self.query)
self.assertEquals(
o, -1, "Same original and target pair. "
"Ranker 1 should win (o = %d)." % o)
test_r1 = ProbabilisticRankingFunction(3, self.weights_1)
a = ([0, 1, 1, 0], r1, test_r1)
o = hpiIs.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r2, r1,
self.query)
self.assertAlmostEquals(
o, -0.1250, 4, "Different original pair. "
"Ranker 1 should win (o = %.4f)." % o)
def testHistProbabilisticInterleave(self):
r1 = ProbabilisticRankingFunction(3, self.weights_1)
r2 = ProbabilisticRankingFunction(3, self.weights_2)
hpi = HistProbabilisticInterleave(None)
a = ([0, 1, 1, 0], r1, r2)
self.assertEqual(
hpi.infer_outcome([0, 1, 2, 3], a, [0, 0, 0, 0], r1, r2,
self.query), 0,
"No clicks, outcome should be 0.")
o = hpi.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r1, r2,
self.query)
self.assertAlmostEquals(
o, -0.0486, 4, "Same target as original "
"rankers. Ranker 1 should win (o = %.4f)." % o)
o = hpi.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r2, r1,
self.query)
self.assertAlmostEquals(
o, 0.0486, 4, "Target rankers switched. "
"Ranker 2 should win (o = %.4f)." % o)
test_r1 = ProbabilisticRankingFunction(3, self.weights_1)
a = ([0, 1, 1, 0], r1, test_r1)
o = hpi.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r2, r1,
self.query)
self.assertAlmostEquals(
o, 0.1542, 4, "Same original ranker. "
"Ranker 2 should win (o = %.4f)." % o)
def testHistProbabilisticInterleaveWithoutMarginalization(self):
r1 = ProbabilisticRankingFunction(3, self.weights_1)
r2 = ProbabilisticRankingFunction(3, self.weights_2)
hpiIs = HistProbabilisticInterleave("--biased False "
"--marginalize False")
# test get_probability_of_list_and_assignment
p = hpiIs._get_probability_of_list_and_assignment([1, 3, 2, 0],
[0, 0, 0, 0], r1, r2, self.query)
self.assertAlmostEqual(p, 0.026261, 6, "Most likely list for ranker 1."
" p = %e" % p)
# test overall outcomes
a = ([0, 1, 1, 0], r1, r2)
self.assertEqual(hpiIs.infer_outcome([0, 1, 2, 3], a, [0, 0, 0, 0],
r1, r2, self.query), 0, "No clicks, outcome should be 0.")
o = hpiIs.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r1, r2,
self.query)
self.assertEquals(o, -1, "Same original and target pair. "
"Ranker 1 should win (o = %d)." % o)
test_r1 = ProbabilisticRankingFunction(3, self.weights_1)
a = ([0, 1, 1, 0], r1, test_r1)
o = hpiIs.infer_outcome([1, 0, 3, 2], a, [1, 0, 0, 0], r2, r1,
self.query)
self.assertAlmostEquals(o, -0.1250, 4, "Different original pair. "
"Ranker 1 should win (o = %.4f)." % o)
