-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdata.py
193 lines (164 loc) · 10.6 KB
/
data.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
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
import numpy as np
class RelevanceJudgments:
def __init__(self, labels_list, labels_range):
'''
:param label_list: [L, M, N] array, L: #queries, M: #documents per query, N: #assessors
:param label_range: a int list, eg: 4-grade [0, 1, 2, 3]
'''
self.labels_list_raw = np.array(labels_list, dtype=np.int)
self.labels_range = labels_range
self.labels_range_dict = {}
for i, label in enumerate(self.labels_range):
self.labels_range_dict[label] = i
self.L, self.M, self.N = self.labels_list_raw.shape
self.G = len(labels_range) # grade scale
self.labels_list = np.array([[[self.labels_range_dict[self.labels_list_raw[l, m, n]]
for n in range(self.N)]
for m in range(self.M)]
for l in range(self.L)])
class PreferenceJudgments:
def __init__(self, judgments, relevancejudgments: RelevanceJudgments, SMOOTH_PARAM=1e-6):
'''
:param judgments: [H, 4] array,
first col: the query id (range from 0 to L - 1)
second col: the document id of left result (range from 0 to M - 1)
third col: the document id of right result (range from 0 to M - 1)
fourth col: preference result, < 0 / = 0 / > 0 represents left result is better / tied / worse than the right one
:param SMOOTH_PARAM: param to exhibit zero-sample condition
'''
self.judgments = judgments
self.relevancejudgments = relevancejudgments
self.SMOOTH_PARAM = SMOOTH_PARAM
self.gradelevel_preference_matrix_individual = None
self.gradelevel_preference_matrix_aggregate = None
self.documentlevel_preference_matrix_individual = None
self.documentlevel_preference_matrix_aggregate = None
def get_gradelevel_preference_matrix(self, mode):
# mode: individual or aggregate
if mode not in ['individual', 'aggregate']:
assert 'invalid grade-level preference matrix mode'
if mode == 'individual':
if self.gradelevel_preference_matrix_individual is not None:
return self.gradelevel_preference_matrix_individual
elif mode == 'aggregate':
if self.gradelevel_preference_matrix_aggregate is not None:
return self.gradelevel_preference_matrix_aggregate
grade_preference_counts_matrix = np.zeros([self.relevancejudgments.G, self.relevancejudgments.G], dtype=np.float)
for judgment in self.judgments:
qid, uidA, uidB, pscore = judgment
scoresA, scoresB = self.relevancejudgments.labels_list[qid, uidA, :], \
self.relevancejudgments.labels_list[qid, uidB, :]
if mode == 'individual':
for i in range(len(scoresA)):
scoreA, scoreB = scoresA[i], scoresB[i]
if pscore < 0:
grade_preference_counts_matrix[scoreA, scoreB] += 1.
elif pscore == 0:
grade_preference_counts_matrix[scoreA, scoreB] += 0.5
grade_preference_counts_matrix[scoreB, scoreA] += 0.5
else:
grade_preference_counts_matrix[scoreB, scoreA] += 1.
elif mode == 'aggregate':
scoresA_sorted, scoresB_sorted = sorted(scoresA), sorted(scoresB)
if len(scoresA_sorted) % 2 == 1:
idx_median = len(scoresA_sorted) // 2
scoreA, scoreB = scoresA_sorted[idx_median], scoresB_sorted[idx_median]
if pscore < 0:
grade_preference_counts_matrix[scoreA, scoreB] += 1.
elif pscore == 0:
grade_preference_counts_matrix[scoreA, scoreB] += 0.5
grade_preference_counts_matrix[scoreB, scoreA] += 0.5
else:
grade_preference_counts_matrix[scoreB, scoreA] += 1.
else:
idx_medianR = len(scoresA_sorted) // 2
idx_medianL = idx_medianR - 1
scoreAL, scoreAR = scoresA_sorted[idx_medianL], scoresA_sorted[idx_medianR]
scoreBL, scoreBR = scoresB_sorted[idx_medianL], scoresB_sorted[idx_medianR]
if pscore < 0:
grade_preference_counts_matrix[scoreAL, scoreBL] += 0.25
grade_preference_counts_matrix[scoreAL, scoreBR] += 0.25
grade_preference_counts_matrix[scoreAR, scoreBL] += 0.25
grade_preference_counts_matrix[scoreAR, scoreBR] += 0.25
elif pscore == 0:
grade_preference_counts_matrix[scoreAL, scoreBL] += 0.125
grade_preference_counts_matrix[scoreAL, scoreBR] += 0.125
grade_preference_counts_matrix[scoreAR, scoreBL] += 0.125
grade_preference_counts_matrix[scoreAR, scoreBR] += 0.125
grade_preference_counts_matrix[scoreBL, scoreAL] += 0.125
grade_preference_counts_matrix[scoreBL, scoreAR] += 0.125
grade_preference_counts_matrix[scoreBR, scoreAL] += 0.125
grade_preference_counts_matrix[scoreBR, scoreAR] += 0.125
else:
grade_preference_counts_matrix[scoreBL, scoreAL] += 0.25
grade_preference_counts_matrix[scoreBL, scoreAR] += 0.25
grade_preference_counts_matrix[scoreBR, scoreAL] += 0.25
grade_preference_counts_matrix[scoreBR, scoreAR] += 0.25
grade_preference_matrix = np.zeros([self.relevancejudgments.G, self.relevancejudgments.G],
dtype=np.float)
for i in range(self.relevancejudgments.G):
for j in range(self.relevancejudgments.G):
grade_preference_matrix[i, j] = grade_preference_counts_matrix[i, j] / (grade_preference_counts_matrix[i, j]
+ grade_preference_counts_matrix[j, i])
if mode == 'individual':
self.gradelevel_preference_matrix_individual = grade_preference_matrix
elif mode == 'aggregate':
self.gradelevel_preference_matrix_aggregate = grade_preference_matrix
return grade_preference_matrix
def get_documentlevel_preference_matrix(self, mode):
# mode: individual or aggregate
if mode not in ['individual', 'aggregate']:
assert 'invalid grade-level preference matrix mode'
if mode == 'individual':
if self.documentlevel_preference_matrix_individual is not None:
return self.documentlevel_preference_matrix_individual
elif mode == 'aggregate':
if self.documentlevel_preference_matrix_aggregate is not None:
return self.documentlevel_preference_matrix_aggregate
if mode == 'individual' and self.gradelevel_preference_matrix_individual is None:
self.gradelevel_preference_matrix_individual = self.get_gradelevel_preference_matrix('individual')
elif mode == 'aggregate' and self.gradelevel_preference_matrix_aggregate is None:
self.gradelevel_preference_matrix_aggregate = self.get_gradelevel_preference_matrix('aggregate')
document_preference_matrix = np.zeros([self.relevancejudgments.L, self.relevancejudgments.M, self.relevancejudgments.M], dtype=np.float)
for i in range(self.relevancejudgments.L):
for j in range(self.relevancejudgments.M):
document_preference_matrix[i, j, j] = 0.5
scoresJ = self.relevancejudgments.labels_list[i, j, :]
for k in range(j):
scoresK = self.relevancejudgments.labels_list[i, k, :]
if mode == 'individual':
_prob = 0.
for o in range(len(scoresJ)):
_prob += self.gradelevel_preference_matrix_individual[scoresJ[o], scoresK[o]]
_prob /= float(len(scoresJ))
document_preference_matrix[i, j, k] = _prob
document_preference_matrix[i, k, j] = 1. - _prob
elif mode == 'aggregate':
scoresJ_sorted, scoresK_sorted = sorted(scoresJ), sorted(scoresK)
if len(scoresJ_sorted) % 2 == 1:
idx_median = len(scoresJ_sorted) // 2
document_preference_matrix[i, j, k] = self.gradelevel_preference_matrix_aggregate[scoresJ_sorted[idx_median], scoresK_sorted[idx_median]]
document_preference_matrix[i, k, j] = self.gradelevel_preference_matrix_aggregate[
scoresK_sorted[idx_median], scoresJ_sorted[idx_median]]
else:
idx_medianR = len(scoresJ_sorted) // 2
idx_medianL = idx_medianR - 1
document_preference_matrix[i, j, k] = (self.gradelevel_preference_matrix_aggregate[scoresJ_sorted[idx_medianL], scoresK_sorted[idx_medianL]] +
self.gradelevel_preference_matrix_aggregate[scoresJ_sorted[idx_medianL], scoresK_sorted[idx_medianR]] +
self.gradelevel_preference_matrix_aggregate[scoresJ_sorted[idx_medianR], scoresK_sorted[idx_medianL]]+
self.gradelevel_preference_matrix_aggregate[scoresJ_sorted[idx_medianR], scoresK_sorted[idx_medianR]]) / 4.
document_preference_matrix[i, k, j] = 1. - document_preference_matrix[i, j, k]
if mode == 'individual':
self.documentlevel_preference_matrix_individual = document_preference_matrix
elif mode == 'aggregate':
self.documentlevel_preference_matrix_aggregate = document_preference_matrix
return document_preference_matrix
def reset(self):
self.gradelevel_preference_matrix_individual = None
self.gradelevel_preference_matrix_aggregate = None
self.documentlevel_preference_matrix_individual = None
self.documentlevel_preference_matrix_aggregate = None
if __name__ == '__main__':
data_4grade = np.array([[[3, 3, 3], [2, 3, 3], [2, 2, 2], [3, 3, 3], [3, 3, 3]]])
rj = RelevanceJudgments(data_4grade, [3, 2, 1, 0])
print(rj.labels_list)