def test_cascade_compute_grads_last_stage():
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
weights = np.array([1., 0.5, 0.25])
grads = cascade.compute_grads(1, cascade.last(), weights)
expected = [1., 0.5, 0.25]
np.testing.assert_almost_equal(expected, grads)
def test_cascade_first_stage_has_no_score_mask():
X, _, qid = fixtures.train_data()
cascade = factories.cascade(num_stages=1, cutoffs=[5])
ranker = cascade.rankers[0]
cascade.predict(X, qid)
assert Cascade.SCORE_MASK not in ranker.predict
def test_cascade_second_stage_score_unkown_type():
cascade = factories.cascade(num_stages=2, cutoffs=[2, 4])
ranker_one = cascade.rankers[0]
ranker_two = cascade.rankers[1]
with pytest.raises(RuntimeError):
cascade.score_type = 'unkown-foo-bar-baz'
prev_ranker = ranker_one
cascade.ranker_score(ranker_two, prev_ranker)
def test_cascade_last_ranker_indicator_is_zero():
cascade = factories.cascade(num_stages=1, cutoffs=[1])
ranker = cascade.last()
ranker.predict = np.array([1., 2., 3.])
ranker.kappa = np.array([1., 1., 1.])
cascade.ranker_indicator()
assert 0. == ranker.indicator_score.all()
assert 0. == ranker.indicator_derivative.all()
def test_cascade_first_stage_score_any_type():
cascade = factories.cascade(num_stages=1, cutoffs=[4])
for name, member in ScoreType.__members__.items():
if member.name != name: # skip alias names
continue
cascade.set_score_type(name)
ranker_one = cascade.rankers[0]
cascade.ranker_score(ranker_one)
assert ranker_one.predict is ranker_one.estimate
def test_cascade_first_stage_applies_cutoff():
X, _, qid = fixtures.train_data()
cascade = factories.cascade(num_stages=1, cutoffs=[2])
ranker = cascade.rankers[0]
ranker.booster.update()
offsets = group_offsets(qid)
a, b = next(offsets)
cascade.predict(X, qid)
expected = (b - a) * [0.01948363]
np.testing.assert_almost_equal(ranker.kappa[a:b], expected)
def test_cascade_score_mask_does_not_appear_in_first_stage():
X, _, qid = fixtures.train_data()
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
cascade.update()
ranker_one = cascade.rankers[0]
ranker_two = cascade.rankers[1]
offsets = group_offsets(qid)
a, b = next(offsets)
cascade.predict(X, qid, is_train=True)
assert Cascade.SCORE_MASK not in ranker_one.predict
def test_cascade_second_stage_applies_cutoff():
X, _, qid = fixtures.train_data()
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
cascade.update()
ranker_two = cascade.rankers[1]
offsets = group_offsets(qid)
a, b = next(offsets)
cascade.predict(X, qid)
topk = sorted(ranker_two.predict[a:b], reverse=True)
expected = (b - a) * [topk[1]]
np.testing.assert_almost_equal(ranker_two.kappa[a:b], expected)
def test_cascade_collect_weights_independent_scoring():
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
cascade.set_score_type('independent')
ranker_one = cascade.rankers[0]
ranker_one.weights = np.array([1., 2., 3.])
ranker_two = cascade.rankers[1]
ranker_two.weights = np.array([3., 2., 1.])
result = cascade.collect_weights(0, ranker_one)
expected = [1., 2., 3.]
np.testing.assert_almost_equal(expected, result)
result = cascade.collect_weights(1, ranker_two)
expected = [3., 2., 1.]
np.testing.assert_almost_equal(expected, result)
def test_cascade_second_stage_score_weak_type():
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
cascade.set_score_type('weak')
ranker_one = cascade.rankers[0]
ranker_one.mask = np.array([1, 1, 1, 1, 0])
ranker_one.estimate = np.array([4., 3., 2., 1., 0.])
ranker_two = cascade.rankers[1]
ranker_two.predict = np.array([3., 5., 1., 5., 5.])
prev_ranker = ranker_one
cascade.ranker_score(ranker_two, prev_ranker)
assert ranker_two.predict is not ranker_two.estimate
np.testing.assert_almost_equal(ranker_two.estimate,
np.array([4., 5., 2., 5., 0.]))
def test_cascade_second_stage_applies_mask():
X, _, qid = fixtures.train_data()
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
cascade.update()
ranker_one = cascade.rankers[0]
ranker_two = cascade.rankers[1]
offsets = group_offsets(qid)
a, b = next(offsets)
cascade.predict(X, qid)
expected = [1, 0, 1, 1, 1]
np.testing.assert_almost_equal(ranker_one.mask[a:b], expected)
expected = [0, 0, 1, 1, 0]
np.testing.assert_almost_equal(ranker_two.mask[a:b], expected)
def test_cascade_collect_weights_full_scoring():
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
cascade.set_score_type('full')
ranker_one = cascade.rankers[0]
ranker_one.weights = np.array([0.75, 0.5, 0.2])
ranker_two = cascade.rankers[1]
ranker_two.weights = np.array([0.75, 0.3, 0.1])
result = cascade.collect_weights(0, ranker_one)
expected = [1.5, 0.8, 0.3]
np.testing.assert_almost_equal(expected, result)
result = cascade.collect_weights(1, ranker_two)
expected = [0.75, 0.3, 0.1]
np.testing.assert_almost_equal(expected, result)
def test_cascade_uses_score_mask():
"""As per previous implementation, always use the SCORE_MASK during predict
regardless of whether we are doing training or inference.
"""
X, _, qid = fixtures.train_data()
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
cascade.update()
ranker_one = cascade.rankers[0]
ranker_two = cascade.rankers[1]
offsets = group_offsets(qid)
a, b = next(offsets)
for is_train in [True, False]:
cascade.predict(X, qid, is_train=is_train)
assert Cascade.SCORE_MASK in ranker_two.predict
def test_cascade_compute_grads():
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
ranker_one = cascade.rankers[0]
ranker_one.weights = np.array([0.5, 0.5, 0.5])
ranker_one.estimate = np.array([-0.5, 0.2, 1.])
ranker_one.indicator_score = np.array([0.5, 0.5, 0.5])
ranker_one.indicator_derivative = np.array([0.25, 0., 0.5])
ranker_two = cascade.rankers[1]
ranker_two.weights = np.array([0.7, 0.7, 0.7])
ranker_two.estimate = np.array([0.25, 0.6, 0.3])
ranker_two.indicator_score = np.array([0.5, 0.5, 0.5])
ranker_two.indicator_derivative = np.array([0., 0., 0.])
weights = cascade.compute_grads(0, ranker_one, ranker_one.weights)
expected = [Cascade.EPSILON, 0.5, 1.0]
np.testing.assert_almost_equal(expected, weights, decimal=4)
def test_cascade_collect_weights_weak_scoring():
cascade = factories.cascade(num_stages=2, cutoffs=[4, 2])
cascade.set_score_type('weak')
ranker_one = cascade.rankers[0]
ranker_one.weights = np.array([0.7, 0.5, 0.2])
ranker_one.predict = np.array([0.1, 0.2, 0.5])
ranker_one.estimate = np.array([0.1, 0.2, 0.5])
ranker_two = cascade.rankers[1]
ranker_two.weights = np.array([0.3, 0.3, 0.1])
ranker_two.predict = np.array([-0.1, 0.3, 0.7])
ranker_two.estimate = np.array([0.1, 0.3, 0.7])
result = cascade.collect_weights(0, ranker_one)
expected = [1.0, 0.5, 0.2]
np.testing.assert_almost_equal(expected, result)
result = cascade.collect_weights(1, ranker_two)
expected = [0.0, 0.3, 0.1]
np.testing.assert_almost_equal(expected, result)
def test_cascade_set_unkown_score_type():
cascade = factories.cascade(num_stages=1, cutoffs=[4])
with pytest.raises(KeyError):
cascade.set_score_type('unkown-foo-bar-baz')