-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathvote_rigging.py
412 lines (298 loc) · 17.9 KB
/
vote_rigging.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
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
import os
import copy
import numpy as np
import pandas as pd
pd.options.display.float_format = '{:.2f}'.format
from utils import preety_print_model_ratings, get_rank, get_battle_pair, preprocess_data, compute_mle_elo_dict
import argparse
import os
import json
parser = argparse.ArgumentParser()
parser.add_argument('--rigging_mode', type=str, default='omni_bt_diff')
parser.add_argument('--beta', type=float, default=1.0)
parser.add_argument('--vote_num', type=int, default=20000)
parser.add_argument('--seed', type=int, default=2025)
parser.add_argument('--classifier_acc', type=float, default=1.0)
parser.add_argument('--model_name_list', nargs='+', default=['phi-3-mini-4k-instruct-june-2024'])
args = parser.parse_args()
K = 4
BASE = 10
SCALE = 400
# --------------------------------------------Initialize the rigging environment--------------------------------------------
# Load the initial data
X_initial, Y_initial, win_matrix_initial, sample_weights_ori = preprocess_data('data/data_x.npy', 'data/data_y.npy','data/vh_win_matrix.csv')
model_name_sorted = []
for model_name in win_matrix_initial.index:
model_name_sorted.append(model_name) if model_name not in model_name_sorted else None
print('Calculate Initial Rating')
elo_ratings, _ = compute_mle_elo_dict([], X=X_initial, Y=Y_initial, ptbl_win=win_matrix_initial, sample_weights=sample_weights_ori)
initial_ranking = preety_print_model_ratings(elo_ratings)
print('---------------initial ranking---------------')
print(initial_ranking)
print('---------------------------------------------')
initial_rating = {}
for idx, key in enumerate(initial_ranking['Model'].keys()):
initial_rating[initial_ranking.loc[key, 'Model']] = initial_ranking.loc[key, 'Elo rating']
if args.rigging_mode == 'omni_bt_diff':
diff_weight = 1
else:
diff_weight = 0
win_dict = {}
for model in model_name_sorted:
win_dict[model] = {}
# Calculate the historical win rate between each pair of models
for model_a in model_name_sorted:
for model_b in model_name_sorted:
if model_a != model_b:
win_dict[model_a][model_b] = {'win': 0, 'lose': 0, 'tie': 0}
with open(f'data/vh.json') as f:
vh_dict = json.load(f)
print(len(vh_dict.keys()))
for key_idx in vh_dict:
model_a = vh_dict[key_idx]['model_a']
model_b = vh_dict[key_idx]['model_b']
winner = vh_dict[key_idx]['winner']
if winner == 'model_a':
win_dict[model_a][model_b]['win'] += 1
win_dict[model_b][model_a]['lose'] += 1
elif winner == 'model_b':
win_dict[model_a][model_b]['lose'] += 1
win_dict[model_b][model_a]['win'] += 1
elif 'tie' in winner:
win_dict[model_a][model_b]['tie'] += 1
win_dict[model_b][model_a]['tie'] += 1
# --------------------------------------------Rigging simulation--------------------------------------------
for target_model in args.model_name_list:
np.random.seed(args.seed)
battle_dict = {}
model_name_sorted_prob = {}
rank_list = [get_rank(initial_ranking, target_model)]
# Generate the sampling distribution for each model, \beta controls the marginal probability of sampling the target model
for model in model_name_sorted:
if model == target_model:
model_weight_tmp = 1 * args.beta
else:
model_weight_tmp = 1 + (1 - args.beta)/(len(model_name_sorted) - 1)
model_name_sorted_prob[model] = model_weight_tmp
print(target_model)
sample_weights_tmp, sample_weights_tmp_noise = copy.deepcopy(sample_weights_ori), copy.deepcopy(sample_weights_ori)
final_ranking, final_ranking_noise = copy.deepcopy(initial_ranking), copy.deepcopy(initial_ranking)
for idx in range(args.vote_num):
model_a, model_b = get_battle_pair(model_name_sorted, model_name_sorted_prob)
current_battle = {'model_a':model_a, 'model_b':model_b, 'winner':None}
# For a fair comparison, we ensure the sampled battles are the same for each rigging strategy
vote_var = np.random.uniform()
acc_var = np.random.uniform()
decision_var = np.random.uniform()
pred_model_a_var = np.random.choice(len(model_name_sorted)-1)
pred_model_b_var = np.random.choice(len(model_name_sorted)-1)
# Rigging strategies
if args.rigging_mode == 't_random' or args.rigging_mode == 't_tie' or args.rigging_mode == 't_normal':
if (model_b == target_model or model_a == target_model) and acc_var <= args.classifier_acc:
decision = 'model_b' if model_b == target_model else 'model_a'
else:
if args.rigging_mode == 't_random':
if vote_var < 1/4:
decision = 'model_a'
elif vote_var < 2/4:
decision = 'model_b'
elif vote_var < 3/4:
decision = 'tie'
elif vote_var < 1:
decision = 'remove'
elif args.rigging_mode == 't_tie':
decision = 'tie'
elif args.rigging_mode == 't_normal':
if acc_var > args.classifier_acc:
tmp_model_name_sorted = copy.deepcopy(model_name_sorted)
tmp_model_name_sorted.pop(tmp_model_name_sorted.index(model_a))
tmp_model_name_sorted.pop(tmp_model_name_sorted.index(model_b))
# pred_model_a = np.random.choice(tmp_model_name_sorted)
# pred_model_b = np.random.choice(tmp_model_name_sorted)
pred_model_a = tmp_model_name_sorted[pred_model_a_var]
pred_model_b = tmp_model_name_sorted[pred_model_b_var]
else:
pred_model_a, pred_model_b = model_a, model_b
win_rate_base = win_dict[pred_model_a][pred_model_b]['win'] + win_dict[pred_model_a][pred_model_b]['lose'] + win_dict[pred_model_a][pred_model_b]['tie']
if win_rate_base != 0 :
win_rate = win_dict[pred_model_a][pred_model_b]['win']/win_rate_base
lose_rate = win_dict[pred_model_a][pred_model_b]['lose']/win_rate_base
tie_rate = win_dict[pred_model_a][pred_model_b]['tie']/win_rate_base
# decision = np.random.choice(['model_a', 'model_b', 'tie'], p=[win_rate, lose_rate, tie_rate])
if decision_var < win_rate:
decision = 'model_a'
elif decision_var < win_rate + lose_rate:
decision = 'model_b'
else:
decision = 'tie'
else:
win_prob_a = 1 / (1 + BASE ** ((initial_rating[pred_model_b] - initial_rating[pred_model_a]) / SCALE))
if vote_var > win_prob_a:
decision = 'model_b'
else:
decision = 'model_a'
current_battle['winner'] = decision
final_ranking, sample_weights_tmp = compute_mle_elo_dict([current_battle], X=X_initial, Y=Y_initial, ptbl_win=win_matrix_initial, sample_weights=copy.deepcopy(sample_weights_tmp))
final_ranking = preety_print_model_ratings(final_ranking)
elif args.rigging_mode == 't_abstain':
if (model_b == target_model or model_a == target_model) and acc_var <= args.classifier_acc:
decision = 'model_b' if model_b == target_model else 'model_a'
current_battle['winner'] = decision
final_ranking, sample_weights_tmp = compute_mle_elo_dict([current_battle], X=X_initial, Y=Y_initial, ptbl_win=win_matrix_initial, sample_weights=copy.deepcopy(sample_weights_tmp))
final_ranking = preety_print_model_ratings(final_ranking)
else:
decision = 'remove'
elif args.rigging_mode == 'omni_on':
pred_model_a = model_a
pred_model_b = model_b
if acc_var > args.classifier_acc:
tmp_model_name_sorted = copy.deepcopy(model_name_sorted)
tmp_model_name_sorted.pop(tmp_model_name_sorted.index(model_a))
tmp_model_name_sorted.pop(tmp_model_name_sorted.index(model_b))
# pred_model_a = np.random.choice(tmp_model_name_sorted)
# pred_model_b = np.random.choice(tmp_model_name_sorted)
pred_model_a = tmp_model_name_sorted[pred_model_a_var]
pred_model_b = tmp_model_name_sorted[pred_model_b_var]
if pred_model_a == pred_model_b:
rank_list.append(rank_list[-1])
battle_dict[f'idx_{idx}'] = {'model_a': model_a, 'model_b': model_b, 'winner': 'remove'}
continue
reward_list = []
ra = initial_rating[pred_model_a]
rb = initial_rating[pred_model_b]
rt = initial_rating[target_model]
ea = 1 / (1 + BASE ** ((rb - ra) / SCALE))
eb = 1 / (1 + BASE ** ((ra - rb) / SCALE))
# case 1
ra1 = ra + K * eb
rb1 = rb - K * eb
reward = 1 / (1 + BASE ** ((ra1 - rt) / SCALE)) + 1 / (1 + BASE ** ((rb1 - rt) / SCALE))
reward_list.append(reward)
# case 2
ra1 = ra - K * ea
rb1 = rb + K * ea
reward = 1 / (1 + BASE ** ((ra1 - rt) / SCALE)) + 1 / (1 + BASE ** ((rb1 - rt) / SCALE))
reward_list.append(reward)
# case 3
ra1 = ra - 0.5 * K * (ea - eb)
rb1 = rb + 0.5 * K * (ea - eb)
reward = 1 / (1 + BASE ** ((ra1 - rt) / SCALE)) + 1 / (1 + BASE ** ((rb1 - rt) / SCALE))
reward_list.append(reward)
# case 4
ra1 = ra
rb1 = rb
reward = 1 / (1 + BASE ** ((ra1 - rt) / SCALE)) + 1 / (1 + BASE ** ((rb1 - rt) / SCALE))
reward_list.append(reward)
if reward_list.index(max(reward_list)) == 0:
decision = 'model_a'
elif reward_list.index(max(reward_list)) == 1:
decision = 'model_b'
elif reward_list.index(max(reward_list)) == 2:
decision = 'tie'
elif reward_list.index(max(reward_list)) == 3:
decision = 'remove'
if pred_model_a == target_model:
decision = 'model_a'
if pred_model_b == target_model:
decision = 'model_b'
current_battle['winner'] = decision
final_ratings, sample_weights_tmp = compute_mle_elo_dict([current_battle], X=X_initial, Y=Y_initial, ptbl_win=win_matrix_initial, sample_weights=copy.deepcopy(sample_weights_tmp))
final_ranking = preety_print_model_ratings(final_ratings)
elif args.rigging_mode == 'omni_bt_diff' or args.rigging_mode == 'omni_bt_abs':
reward_list = []
tmp_ranking_list = []
tmp_weights_list = []
pred_model_a = model_a
pred_model_b = model_b
if acc_var > args.classifier_acc:
tmp_model_name_sorted = copy.deepcopy(model_name_sorted)
tmp_model_name_sorted.pop(tmp_model_name_sorted.index(model_a))
tmp_model_name_sorted.pop(tmp_model_name_sorted.index(model_b))
# pred_model_a = np.random.choice(tmp_model_name_sorted)
# pred_model_b = np.random.choice(tmp_model_name_sorted)
pred_model_a = tmp_model_name_sorted[pred_model_a_var]
pred_model_b = tmp_model_name_sorted[pred_model_b_var]
if pred_model_a == pred_model_b:
rank_list.append(rank_list[-1])
battle_dict[f'idx_{idx}'] = {'model_a': model_a, 'model_b': model_b, 'winner': 'remove'}
continue
for key in final_ranking_noise['Model'].keys():
if final_ranking_noise.loc[key, 'Model'] == target_model:
if key != 1:
tmp_anchor_model = final_ranking_noise.loc[key-1, 'Model']
tmp_anchor_model_rating = final_ranking_noise.loc[key-1, 'Elo rating']
else:
tmp_anchor_model = final_ranking_noise.loc[key+1, 'Model']
tmp_anchor_model_rating = final_ranking_noise.loc[key+1, 'Elo rating']
remove_rating = final_ranking_noise.loc[key, 'Elo rating'] - tmp_anchor_model_rating * diff_weight
else:
for key in final_ranking['Model'].keys():
if final_ranking.loc[key, 'Model'] == target_model:
if key != 1:
tmp_anchor_model = final_ranking.loc[key-1, 'Model']
tmp_anchor_model_rating = final_ranking.loc[key-1, 'Elo rating']
else:
tmp_anchor_model = final_ranking.loc[key+1, 'Model']
tmp_anchor_model_rating = final_ranking.loc[key+1, 'Elo rating']
remove_rating = final_ranking.loc[key, 'Elo rating'] - tmp_anchor_model_rating * diff_weight
for tmp_vote in ['model_a', 'model_b', 'tie']:
tmp_battle = {'model_a':pred_model_a, 'model_b':pred_model_b, 'winner': tmp_vote}
if args.classifier_acc < 1:
tmp_ratings, tmp_weights = compute_mle_elo_dict([tmp_battle], X=X_initial, Y=Y_initial, ptbl_win=win_matrix_initial, sample_weights=copy.deepcopy(sample_weights_tmp_noise))
tmp_ranking = preety_print_model_ratings(tmp_ratings)
else:
tmp_ratings, tmp_weights = compute_mle_elo_dict([tmp_battle], X=X_initial, Y=Y_initial, ptbl_win=win_matrix_initial, sample_weights=copy.deepcopy(sample_weights_tmp))
tmp_ranking = preety_print_model_ratings(tmp_ratings)
tmp_ranking_list.append(copy.deepcopy(tmp_ranking))
tmp_weights_list.append(copy.deepcopy(tmp_weights))
for _, key in enumerate(tmp_ranking['Model'].keys()):
if tmp_ranking.loc[key, 'Model'] == tmp_anchor_model:
tmp_anchor_model_rating = tmp_ranking.loc[key, 'Elo rating']
if tmp_ranking.loc[key, 'Model'] == target_model:
tmp_target_model_rating = tmp_ranking.loc[key, 'Elo rating']
reward_list.append(tmp_target_model_rating - tmp_anchor_model_rating * diff_weight)
reward_list.append(remove_rating)
if reward_list.index(max(reward_list)) == 0:
decision = 'model_a'
elif reward_list.index(max(reward_list)) == 1:
decision = 'model_b'
elif reward_list.index(max(reward_list)) == 2:
decision = 'tie'
elif reward_list.index(max(reward_list)) == 3:
decision = 'remove'
if args.classifier_acc < 1:
current_battle['winner'] = decision
final_ratings, sample_weights_tmp = compute_mle_elo_dict([current_battle], X=X_initial, Y=Y_initial, ptbl_win=win_matrix_initial, sample_weights=copy.deepcopy(sample_weights_tmp))
final_ranking = preety_print_model_ratings(final_ratings)
if decision == 'model_a':
final_ranking_noise = tmp_ranking_list[0]
sample_weights_tmp_noise = tmp_weights_list[0]
elif decision == 'model_b':
final_ranking_noise = tmp_ranking_list[1]
sample_weights_tmp_noise = tmp_weights_list[1]
elif decision == 'tie':
final_ranking_noise = tmp_ranking_list[2]
sample_weights_tmp_noise = tmp_weights_list[2]
else:
if decision == 'model_a':
final_ranking = tmp_ranking_list[0]
sample_weights_tmp = tmp_weights_list[0]
elif decision == 'model_b':
final_ranking = tmp_ranking_list[1]
sample_weights_tmp = tmp_weights_list[1]
elif decision == 'tie':
final_ranking = tmp_ranking_list[2]
sample_weights_tmp = tmp_weights_list[2]
# assert reward_list[0] != reward_list[1]
del tmp_ranking_list, tmp_ranking, tmp_weights_list
battle_dict[f'idx_{idx}'] = {'model_a': model_a, 'model_b': model_b, 'winner': decision}
rank_list.append(get_rank(final_ranking, target_model))
if model_a == target_model or model_b == target_model:
print(f'Battle idx: {idx} | Mode: {args.rigging_mode} | decision: {decision} | Rank: {get_rank(final_ranking, target_model)} | {model_a} vs {model_b} (target)')
else:
print(f'Battle idx: {idx} | Mode: {args.rigging_mode} | decision: {decision} | Rank: {get_rank(final_ranking, target_model)} | {model_a} vs {model_b}')
os.makedirs('ranking_output/', exist_ok=True)
os.makedirs('voting_output/', exist_ok=True)
np.save(f'ranking_output/{target_model}_{args.rigging_mode}_acc_{args.classifier_acc}_prob_dec_{args.beta}.npy', np.array(rank_list))
with open(f'voting_output/{target_model}_{args.rigging_mode}_acc_{args.classifier_acc}_prob_dec_{args.beta}.json', 'w') as f:
json.dump(battle_dict, f, indent=4)