-
Notifications
You must be signed in to change notification settings - Fork 0
/
ndcg.py
315 lines (255 loc) · 10.4 KB
/
ndcg.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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
#!/home/workboots/VirtualEnvs/aiml/bin/python3
# -*- encoding: utf-8 -*-
# Birth: 2022-07-05 14:56:27.996067960 +0530
# Modify: 2022-07-27 19:18:02.467252970 +0530
"""Compute graded relevance"""
import argparse
import json
import math
import os
from collections import defaultdict
__author__ = "Upal Bhattacharya"
__license__ = ""
__copyright__ = ""
__version__ = "1.0"
__email__ = "upal.bhattacharya@gmail.com"
def create_adv_scores(charges: list, charge_adv_win_ratios: dict,
weights: dict = None, strategy: str = 'equal',):
"""Generate scores of advocates for a given combination of charges
depending on the weightage strategy
Parameters
----------
charges: list
List of charges.
charge_adv_win_ratios: dict
Dictionary of win ratios of each advocate for each charge.
weight: dict, default None
Weights to be used for combining win-ratios for resultant score.
Used when 'strategy' is not 'equal'.
strategy: str
Weightage strategy to use.
'equal': Equal weightage to each charge.
'case_fraction': Weigh charges based on fraction of cases.
Returns
-------
scores: dict
Scores of advocates.
"""
assert strategy in ['equal', 'case_fraction'], "Invalid strategy"
if strategy == 'case_fraction':
assert weights is not None, ("Weights need to be specified when "
"using 'case_fraction' weightage.")
scores = defaultdict(float)
if strategy == "equal":
weights = {charge: 1./len(charges) for charge in charges}
for charge in charges:
for adv, score in charge_adv_win_ratios[charge].items():
scores[adv] += weights[charge] * score
else:
print(weights)
for charge in charges:
for adv, score in charge_adv_win_ratios[charge].items():
scores[adv] += weights[charge] * score
scores = {k: v for k, v in sorted(
scores.items(), key=lambda x: x[1], reverse=True)}
return scores
def create_targets(targets_dict, adv_index, case,
case_charges=None, adv_charges=None, threshold=None):
"""Create targets from a dictionary of targets and advocate ordering.
Parameters
----------
targets_dict : dict
Dictionary with the targets of each case.
adv_list : list
List of advocates to consider.
cases : list
Ordered list cases.
Returns
-------
result : numpy.array
Stacked target mult-hot vectors.
"""
actual = []
lenient = []
# Lenient
if all(ele is not None
for ele in [case_charges, adv_charges, threshold]):
lenient = [adv
for adv in list(adv_index.keys())
if adv not in targets_dict[case] and
len(set(adv_charges[adv]).intersection(
set(case_charges[case]))) * 1./len(
case_charges[case]) >= threshold]
actual = [adv
for adv in list(adv_index.keys())
if adv in targets_dict[case]]
return actual, lenient
def dcg(relevance, predicted):
score = sum([relevance[name] * 1./math.log(i+2, 2)
for i, name in enumerate(predicted)])
return score
def idcg(relevance, predicted):
relevance = {k: v for k, v in sorted(
relevance.items(), key=lambda x: x[1],
reverse=True)[:len(predicted)]}
score = sum([val * 1./math.log(i+2, 2)
for i, val in enumerate(relevance.values())])
return score
# def relevance_rank(actual, lenient, full):
# relevance = {}
# for adv in full:
# if adv in actual:
# relevance[adv] = 3 if lenient is not [] else 1
# elif adv in lenient[:int(len(lenient)/2) + 1]:
# relevance[adv] = 2
# elif adv in lenient[int(len(lenient)/2) + 1:]:
# relevance[adv] = 1
# else:
# relevance[adv] = 0
# relevance = {k: v for k, v in sorted(
# relevance.items(),
# key=lambda x: x[1], reverse=True)}
# return relevance
def integer_relevance_rank(actual, lenient, full, relevance_scores):
advs = [*actual, *lenient]
num = len(advs)
relevance = {k: v for k, v in zip([a for a in sorted(
advs, key=lambda x: relevance_scores[x],
reverse=True)], range(num, 0, -1))}
relevance.update({k: 0.0 for k in full if k not in advs})
print(relevance)
return relevance
def cont_relevance_rank(actual, lenient, full, relevance_scores):
relevance = {}
print(relevance_scores)
for adv in full:
if adv in actual or adv in lenient:
relevance[adv] = relevance_scores[adv]
else:
relevance[adv] = 0
return relevance
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--charge_adv_win_ratios",
help="Win ratios of advocates for each charge.")
parser.add_argument("--case_charges",
help="Charges/offences of each case.")
parser.add_argument("--charge_cases",
help="Cases of each charge.")
parser.add_argument("--advocate_charges",
help="Charges of each advocate.")
parser.add_argument("--charge_targets",
help="Target charges to consider.")
parser.add_argument("--targets",
help="Target advocates of each case.")
parser.add_argument("--relevant_advocates",
help="Advocates to consider.")
parser.add_argument("--relevant_cases", type=str, default=None,
help="Cases to consider in evaluation.")
parser.add_argument("--scores",
help="Ranked scores of advocates.")
parser.add_argument("--strategy", type=str, default='equal',
help="Score combination strategy.")
parser.add_argument("--output_path",
help="Path to save generated ndcg scores.")
parser.add_argument("--threshold", type=float, default=None,
help="Threshold for lenient targets.")
args = parser.parse_args()
if not os.path.exists(args.output_path):
os.makedirs(args.output_path)
# Charge advocate win ratios
with open(args.charge_adv_win_ratios, 'r') as f:
charge_adv_win_ratios = json.load(f)
# Case offences
with open(args.case_charges, 'r') as f:
case_charges = json.load(f)
# Charge cases
with open(args.charge_cases, 'r') as f:
charge_cases = json.load(f)
# Advocate charges
with open(args.advocate_charges, 'r') as f:
adv_charges = json.load(f)
# Charge targets
with open(args.charge_targets, 'r') as f:
charge_targets = f.readlines()
charge_targets = list(filter(None, map(lambda x: x.strip("\n"),
charge_targets)))
# Case advocates
with open(args.targets, 'r') as f:
case_advs = json.load(f)
# Relevant advocates
with open(args.relevant_advocates, 'r') as f:
advs = json.load(f)
advs = {k: i for i, k in advs.items()}
# Relevant cases
if args.relevant_cases is not None:
with open(args.relevant_cases, 'r') as f:
rel_cases = f.readlines()
rel_cases = list(filter(None, map(lambda x: x.strip("\n"),
rel_cases)))
else:
rel_cases = None
# Scores
with open(args.scores, 'r') as f:
scores = json.load(f)
# Getting scores of charges
result_scores = {}
ndcg = {}
if args.strategy == 'equal':
weights = None
else:
total_cases = set([value
for values in charge_cases.values()
for value in values])
weights = {}
for charge, cases in charge_cases.items():
weights[charge] = len(cases) * 1./len(total_cases)
print(weights)
for case, ranks in scores.items():
if rel_cases is not None:
if case not in rel_cases:
continue
if case_advs.get(case, -1) == -1:
print(f"{case} not found.")
continue
if set(case_advs[case]).intersection(set(advs)) == set():
print(f"{case} not found.")
continue
rel_charges = set(case_charges[case]).intersection(set(charge_targets))
if rel_charges == set():
continue
result_scores[case] = create_adv_scores(rel_charges,
charge_adv_win_ratios,
strategy=args.strategy,
weights=weights)
actual, lenient = create_targets(targets_dict=case_advs,
adv_index=advs,
case_charges=case_charges,
adv_charges=adv_charges,
threshold=args.threshold,
case=case)
# Continuous Scale Relevance Ranking
# relevance = cont_relevance_rank(actual=actual,
# lenient=lenient,
# full=advs,
# relevance_scores=result_scores[case])
relevance = integer_relevance_rank(
actual=actual,
lenient=[],
full=advs,
relevance_scores=result_scores[case])
predicted = list(ranks.keys())
print(predicted)
ndcg_score = dcg(relevance, predicted) * 1./idcg(relevance, predicted)
ndcg[case] = ndcg_score
print(len(ndcg.keys()))
avg = sum(ndcg.values()) * 1./len(ndcg.keys())
# Saving
with open(os.path.join(args.output_path, "ndcg.json"), 'w') as f:
json.dump(ndcg, f, indent=4)
with open(os.path.join(args.output_path, "avg_ndcg.txt"), 'w') as f:
print(avg, file=f)
with open(os.path.join(args.output_path, "result_scores.json"), 'w') as f:
json.dump(result_scores, f, indent=4)
if __name__ == "__main__":
main()