-
Notifications
You must be signed in to change notification settings - Fork 21
/
pl_exhaustive_test.py
174 lines (156 loc) · 6.43 KB
/
pl_exhaustive_test.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
# Copyright 2024 DeepMind Technologies Limited
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for pl_exhaustive."""
from absl.testing import absltest
from absl.testing import parameterized
import jax
import jax.numpy as jnp
import numpy as np
import pl_exhaustive
class PlackettLuceExhaustiveTest(parameterized.TestCase):
def get_test_data_pl_likelihood(self):
"""Test data for PL likelihood functions.
Returns:
The dictionary for variables needed for testing the likelihood functions.
"""
theta_1 = jnp.array([0.3, 0.15, 0.4, 0.1, 0.05])
theta_2 = jnp.array([0.1, 0.3, 0.1, 0.15, 0.35])
phi = jnp.vstack((jnp.log(theta_1) + 1.0, jnp.log(theta_2) + 1.0))
num_readers = 2
batch_size, num_classes = phi.shape
rankings = jnp.zeros((batch_size, num_readers, num_classes), dtype=int)
# The first instance has two readers.
rankings = rankings.at[0, 0, :].set(jnp.array([3, 0, 1, 2, 4]))
rankings = rankings.at[0, 1, :].set(jnp.array([2, 1, 4, 3, 0]))
# The second instance has one reader.
rankings = rankings.at[1, 0, :].set(jnp.array([1, 4, 3, 0, 2]))
selectors_1_0 = [[3, 0], [1, 2]]
selectors_1_1 = [[2], [1], [4]]
selectors_1_2 = [[3, 0], [1, 2], [4]] # Equivalent to selectors_1_0.
selectors_2 = [[1, 4], [3]]
selectors = [[selectors_1_0, selectors_1_1, selectors_1_2], [selectors_2]]
likelihood_1_0 = ((0.1 * (0.3 / 0.9)) + (0.3 * (0.1 / 0.7))) * (
((0.15 / 0.6) * (0.4 / 0.45)) + ((0.4 / 0.6) * (0.15 / 0.2))
)
likelihood_1_1 = (0.4) * (0.15 / 0.6) * (0.05 / 0.45)
likelihood_1_2 = likelihood_1_0
likelihood_1 = likelihood_1_0 * likelihood_1_1 * likelihood_1_2
likelihood_2 = ((0.3 * (0.35 / 0.7)) + (0.35 * (0.3 / 0.65))) * (
0.15 / 0.35
)
correct_loglikelihoods = jnp.log(jnp.array([likelihood_1, likelihood_2]))
correct_reader_loglikelihoods = [
jnp.log(jnp.array([likelihood_1_0, likelihood_1_1, likelihood_1_2])),
jnp.log(jnp.array([likelihood_2])),
]
return { # pytype: disable=bad-return-type # jax-ndarray
"phi": phi,
"rankings": rankings,
"selectors": selectors,
"correct_loglikelihoods": correct_loglikelihoods,
"correct_reader_loglikelihoods": correct_reader_loglikelihoods,
}
def test_full_top_k_ordering_given_sum(self):
ordered_lam = jnp.array([3., 2., 4.])
sum_lam = 12.
ordered_phi = jnp.log(ordered_lam)
logsumexp_phi = jnp.log(sum_lam)
ll_expected = jnp.log((3 / 12) * (2 / 9) * (4 / 7))
var_full_top_k_ordering_given_sum = jax.jit(
pl_exhaustive.full_top_k_ordering_given_sum
)
ll_returned = var_full_top_k_ordering_given_sum(ordered_phi, logsumexp_phi)
self.assertAlmostEqual(ll_expected, ll_returned, places=5)
def test_pl_loglikelihood_group(self):
lam = jnp.array([1., 0.5, 3.])
sum_lam = 6.
logsumexp_phi = jnp.log(sum_lam)
selector = [1, 2]
phi = jnp.log(lam)
ll_expected = jnp.log((.5 / 6) * (3 / 5.5) + (3 / 6) * (.5 / 3))
var_pl_loglikelihood_group = jax.jit(pl_exhaustive._pl_loglikelihood_group)
ll_returned = var_pl_loglikelihood_group(phi, logsumexp_phi, selector)
self.assertAlmostEqual(ll_expected, ll_returned)
def test_pl_loglikelihood_reader(self):
test_data = self.get_test_data_pl_likelihood()
example_indices = [0, 1]
reader_indices = [[0, 1, 2], [0]]
var_pl_loglikelihood_reader = jax.jit(
pl_exhaustive._pl_loglikelihood_reader
)
for example_index in example_indices:
for reader_index in reader_indices[example_index]:
ll_returned = var_pl_loglikelihood_reader(
test_data["phi"][example_index],
test_data["selectors"][example_index][reader_index],
)
self.assertAlmostEqual(
test_data["correct_reader_loglikelihoods"][example_index]
[reader_index],
ll_returned,
places=5)
def test_pl_loglikelihood_reader_full_partial_ranking_equivalence(self):
phi = jnp.log(jnp.array([0.3, 0.15, 0.4, 0.1, 0.05])) + 1.0
selectors_1 = [[3, 0], [1, 2]]
selectors_2 = [[3, 0], [1, 2], [4]]
ll_1 = pl_exhaustive._pl_loglikelihood_reader(
phi,
selectors_1,
)
ll_2 = pl_exhaustive._pl_loglikelihood_reader(
phi,
selectors_2,
)
self.assertEqual(
ll_1,
ll_2,
)
def test_pl_loglikelihood_single(self):
test_data = self.get_test_data_pl_likelihood()
example_index = 0
var_pl_loglikelihood_single = jax.jit(pl_exhaustive.pl_loglikelihood_single)
ll_returned = var_pl_loglikelihood_single(
test_data["phi"][example_index],
test_data["selectors"][example_index])
self.assertAlmostEqual(
test_data["correct_loglikelihoods"][example_index],
ll_returned,
places=5)
# A test case specific to the exhaustive method. In the exhaustive method.
theta = jnp.array([0.15, 0.4, 0.05, 0.05, 0.35])
phi = jnp.log(theta) + 1.5
selector_1 = [[[4, 2], [3]]]
# Adding the last group should not change behavior.
selector_2 = [[[4, 2], [3], [0, 1]]]
ll_returned_1 = var_pl_loglikelihood_single(
phi, selector_1)
ll_returned_2 = var_pl_loglikelihood_single(
phi, selector_2)
self.assertAlmostEqual(ll_returned_1, ll_returned_2, places=5)
def test_pl_loglikelihood_batch(self):
test_data = self.get_test_data_pl_likelihood()
for var_pl_loglikelihood_batch in [
pl_exhaustive.pl_loglikelihood_batch_jit_per_instance,
pl_exhaustive.pl_loglikelihood_batch_jit_per_reader,
pl_exhaustive.pl_loglikelihood_batch,
jax.jit(pl_exhaustive.pl_loglikelihood_batch),
]:
ll_returned = var_pl_loglikelihood_batch(
test_data["phi"], test_data["selectors"])
np.testing.assert_array_almost_equal(
test_data["correct_loglikelihoods"], ll_returned
)
if __name__ == "__main__":
absltest.main()