-
Notifications
You must be signed in to change notification settings - Fork 0
/
ranker.py
345 lines (283 loc) · 14.8 KB
/
ranker.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
import math
import sys
import time
import metapy
import pytoml
import numpy as np
from scipy.stats import ttest_rel as tt
import random
from scipy.stats import wilcoxon
from matplotlib import pyplot as plt
def load_ranker(cfg_file,mu):
"""
Use this function to return the Ranker object to evaluate,
The parameter to this function, cfg_file, is the path to a
configuration file used to load the index.
"""
# return metapy.index.JelinekMercer(0.5)
return metapy.index.DirichletPrior(mu)
def score2(ranker,index,query,top_k,alpha):
print("Scoring")
# print("start")
results = ranker.score(index, query, 1000)
# print(results[:20])
for res in results[:20]:
doc_name = index.metadata(res[0]).get('doc_name')
# print(doc_name,res[1])
new_results = []
new_scores = []
updated_results = {}
alpha = alpha #all, 2500, 0.13, 0.666
for res in results:
doc_name = index.metadata(res[0]).get('doc_name')
# print(doc_name)
# print(float(index.metadata(res[0]).get('prior')))
updated_results[doc_name] = (1-alpha)*res[1]+alpha*float(index.metadata(res[0]).get('prior') )
new_scores.append(updated_results[doc_name])
# print (sorted(updated_results.items(),key=lambda k:k[1],reverse=True)[:])
new_idx = np.argsort(np.array(new_scores))[::-1][:top_k]
for idx in new_idx:
new_results.append((results[idx][0],new_scores[idx]))
return new_results,updated_results
def score1(ranker,index,query,top_k):
# print("Scoring")
# print(query)
results = ranker.score(index, query, top_k)
# print(results[:20])
for res in results[:20]:
doc_name = index.metadata(res[0]).get('doc_name')
# print(doc_name,res[1])
new_results = []
new_scores = []
updated_results = {}
for res in results:
doc_name = index.metadata(res[0]).get('doc_name')
# print(float(index.metadata(res[0]).get('prior')))
updated_results[doc_name] = res[1]+float(index.metadata(res[0]).get('prior') )
new_scores.append(updated_results[doc_name])
# print (sorted(updated_results.items(),key=lambda k:k[1],reverse=True)[:10])
new_idx = np.argsort(np.array(new_scores))[::-1][:10]
for idx in new_idx:
new_results.append((results[idx][0],new_scores[idx]))
return new_results
# return results[:top_k]
if __name__ == '__main__':
cfg = '/Users/bhavya/Documents/explanation/UI/para_idx_data/config.toml'
with open(cfg,'r') as f:
print(f.read())
print('Building or loading index...')
idx = metapy.index.make_inverted_index(cfg)
with open(cfg, 'r') as fin:
cfg_d = pytoml.load(fin)
query_cfg = cfg_d['query-runner']
if query_cfg is None:
print("query-runner table needed in {}".format(cfg))
sys.exit(1)
start_time = time.time()
query_path = query_cfg.get('query-path', 'slide-queries.txt')
query_start = query_cfg.get('query-id-start', 1)
ev = metapy.index.IREval(cfg)
query = metapy.index.Document()
query.content('WordNet ontology')
score2(ranker,idx,query)
# with open(query_path) as query_file:
# queries = [q.strip('\n') for q in query_file.readlines()]
# # random.shuffle(queries)
# qids = list(range(len(queries)))
# # random.shuffle(qids)
# folds_div_test = {}
# folds_div_train = {}
# for fold in range(1):
# st= int(0.0*(fold)*len(queries))
# folds_div_test[fold] = qids[st:int(0.0*(fold+1)*len(queries))]
# folds_div_train[fold] = []
# for qi in qids:
# if qi not in folds_div_test[fold]:
# folds_div_train[fold].append(qi)
# print('Running queries')
# overall_avg = {}
# fig, axs = plt.subplots(3, 3,sharex=True,sharey=True)
# cnt1 = 0
# cnt2 = 0
# fig.text(0.5, 0.04, 'alpha', ha='center')
# fig.text(0.055, 0.5, 'Recall@3', va='center', rotation='vertical')
# with open('garbage.txt','w',buffering=0) as f:
# for fold in range(1):
# train_queries = []
# test_queries = []
# train_qids = []
# test_qids = []
# for qidx,q in enumerate(queries):
# if qidx in folds_div_test[fold]:
# test_queries.append(q)
# test_qids.append(qidx)
# else:
# train_queries.append(q)
# train_qids.append(qidx)
# print(len(train_queries),len(test_queries))
# top_k = 3
# best_ndcg1 = 0
# best1_params = 0
# best_ndcg2 = 0
# best2_params = [0,0]
# for mu in np.arange(1000,5500,500):
# print(fold,top_k,mu)
# plt_ndcg1 = []
# plt_ndcg2 = []
# for alpha in np.arange(0,1.01,0.01):
# ndcg1 = 0.0
# ndcg2 = 0.0
# ndcg1_arr = []
# ndcg2_arr = []
# num_queries = 0
# for query_num, line in enumerate(train_queries):
# ranker = load_ranker(cfg,mu)
# # print(line)
# query.content(line.strip())
# results1 = score1(ranker,idx, query,top_k)
# results2 = score2(ranker,idx,query,top_k,alpha)
# # if ev.ndcg(results1, query_start + query_num, top_k)<ev.ndcg(results2, query_start + query_num, top_k):
# # print(query_num,results1[:top_k],results2[:top_k])
# # print( query.content(),ev.ndcg(results1, query_start + query_num, top_k),ev.ndcg(results2, query_start + query_num, top_k))
# # f.write(ev.ndcg(results2, query_start + query_num, top_k))
# tmp1 = ev.recall(results1, train_qids[query_num]+query_start, top_k)
# ndcg1 += tmp1
# ndcg1_arr.append(tmp1)
# tmp2 = ev.recall(results2, train_qids[query_num]+query_start, top_k)
# ndcg2 += tmp2
# ndcg2_arr.append(tmp2)
# num_queries+=1
# ndcg1= ndcg1 / num_queries
# ndcg2= ndcg2 / num_queries
# if ndcg1>best_ndcg1:
# best_ndcg1 = ndcg1
# best1_params = mu
# if ndcg2>best_ndcg2:
# best_ndcg2 = ndcg2
# best2_params = [mu,alpha]
# plt_ndcg2.append(ndcg2)
# plt_ndcg1.append(ndcg1)
# # f.write("Fold {} top {} {} {} {} {} \n".format(fold,top_k,ndcg1,ndcg2,mu,alpha))
# # f.write(str(tt(ndcg1_arr,ndcg2_arr)))
# print(ndcg1,ndcg2,best1_params,best2_params)
# axs[cnt1,cnt2].plot(list(np.arange(0,1.01,0.01)),plt_ndcg1,'k-.',linewidth=1,markersize=3)
# axs[cnt1,cnt2].plot(list(np.arange(0,1.01,0.01)),plt_ndcg2,'r-.',linewidth=1,markersize=3)
# axs[cnt1,cnt2].set_title('mu = {}'.format(mu))
# axs[cnt1,cnt2].grid()
# axs[cnt1,cnt2].set_ylim([0.69,0.73])
# # axs[cnt1,cnt2].legend(loc="upper right")
# cnt2 += 1
# if cnt2 == 3:
# cnt1+= 1
# cnt2 = 0
# print(best1_params,best2_params)
# plt.show()
# f.write("Fold {} top {}\n".format(fold,top_k))
# f.write("Train\n")
# f.write("NDCG1: {} \t{} \n".format(best_ndcg1,best1_params))
# f.write("NDCG2: {} \t{} \t{}\n".format(best_ndcg2,best2_params[0],best2_params[1]))
# f.write("Test:\n")
# print(best1_params,best2_params)
# best1_params = 3000
# best2_params = [2500,0.34]
# test_queries = queries
# test_qids = qids
# with open('complete.txt','w') as f:
# for top_k in [1,2,3]:
# print(top_k)
# ndcg1 = 0.0
# ndcg2 = 0.0
# avg_p1 = 0.0
# avg_p2 = 0.0
# avg_recall1 = 0.0
# avg_recall2 = 0.0
# ndcg1_arr = []
# ndcg2_arr = []
# avgp1_arr = []
# avgp2_arr = []
# avgr1_arr = []
# avgr2_arr = []
# num_queries = 0
# sgn = 0
# for query_num, line in enumerate(test_queries):
# ranker = load_ranker(cfg,best1_params)
# query.content(line.strip())
# results1 = score1(ranker,idx, query,top_k)
# tmp1 = ev.ndcg(results1, test_qids[query_num]+query_start, top_k)
# ndcg1 += tmp1
# ndcg1_arr.append(tmp1)
# ranker = load_ranker(cfg,best2_params[0])
# query.content(line.strip())
# results2 = score2(ranker,idx, query,top_k,best2_params[1])
# tmp2 = ev.ndcg(results2, test_qids[query_num]+query_start, top_k)
# ndcg2 += tmp2
# ndcg2_arr.append(tmp2)
# # if tmp1>tmp2:
# # print("1>2\n")
# # print(query.content(),tmp1,tmp2,results1,results2)
# sgn += (tmp2-tmp1)<0
# if tmp1<tmp2:
# print("1<2\n")
# print(query.content(),tmp1,tmp2,results1,results2)
# tmp1 = ev.recall(results1, test_qids[query_num]+query_start, top_k)
# avg_recall1 += tmp1
# avgr1_arr.append(tmp1)
# tmp1 = ev.avg_p(results1, test_qids[query_num]+query_start, top_k)
# avg_p1 += tmp1
# avgp1_arr.append(tmp1)
# tmp2 = ev.recall(results2, test_qids[query_num]+query_start, top_k)
# avg_recall2 += tmp2
# avgr2_arr.append(tmp2)
# tmp2 = ev.avg_p(results2, test_qids[query_num]+query_start, top_k)
# avg_p2 += tmp2
# avgp2_arr.append(tmp2)
# num_queries+=1
# ndcg1= ndcg1 / num_queries
# ndcg2= ndcg2 / num_queries
# avg_recall1= avg_recall1 / num_queries
# avg_recall2= avg_recall2 / num_queries
# avg_p1= avg_p1 / num_queries
# avg_p2= avg_p2 / num_queries
# sgn = sgn/float(num_queries)
# print(len(ndcg1_arr),len(ndcg2_arr),len(avgp2_arr),len(avgp1_arr),len(avgp1_arr),len(avgr2_arr))
# f.write("NDCG1: {} \n".format(ndcg1))
# f.write("NDCG2: {} \n".format(ndcg2))
# # print(np.std(np.array(ndcg1_arr)),np.std(np.array(ndcg2_arr)))
# f.write(str(ndcg1_arr))
# f.write('\n')
# f.write(str(ndcg2_arr))
# f.write(str(sgn))
# f.write('\n')
# f.write("pvalue: {} \n".format(wilcoxon(ndcg1_arr,ndcg2_arr)))
# f.write("AVG1: {} \n".format(avg_p1))
# f.write("AVG2: {} \n".format(avg_p2))
# f.write("pvalue: {} \n".format(wilcoxon(avgp1_arr,avgp2_arr)))
# f.write("rec1: {} \n".format(avg_recall1))
# f.write("rec2: {} \n".format(avg_recall2))
# f.write("pvalue: {} \n".format(wilcoxon(avgr1_arr,avgr2_arr)))
# # try:
# overall_avg['train_ndcg1_{}'.format(top_k)].append(best_ndcg1)
# overall_avg['test_ndcg1_{}'.format(top_k)].append(ndcg1)
# overall_avg['train_ndcg2_{}'.format(top_k)].append(best_ndcg2)
# overall_avg['test_ndcg2_{}'.format(top_k)].append(ndcg2)
# overall_avg['test_r1_{}'.format(top_k)].append(avg_recall1)
# overall_avg['test_r2_{}'.format(top_k)].append(avg_recall2)
# overall_avg['test_p1_{}'.format(top_k)].append(avg_p1)
# overall_avg['test_p2_{}'.format(top_k)].append(avg_p2)
# except:
# overall_avg['train_ndcg1_{}'.format(top_k)] = [best_ndcg1]
# overall_avg['test_ndcg1_{}'.format(top_k)] = [ndcg1]
# overall_avg['train_ndcg2_{}'.format(top_k)] = [best_ndcg2]
# overall_avg['test_ndcg2_{}'.format(top_k)] = [ndcg2]
# overall_avg['test_r1_{}'.format(top_k)] = [avg_recall1]
# overall_avg['test_r2_{}'.format(top_k)] = [avg_recall2]
# overall_avg['test_p1_{}'.format(top_k)] = [avg_p1]
# overall_avg['test_p2_{}'.format(top_k)] = [avg_p2]
# f.write('#'*20+'\n')
# for top_k in [1,3,5]:
# f.write("TOp k {}\n".format(top_k))
# f.write('Train avg ndcg1: {} \t {}\t Train avg ndcg2: {}\t {}\tTest avg ndcg1: {} \t {}\tTest avg ndcg2: {}\t{}\t \n\n'.format(np.average(np.array(overall_avg['train_ndcg1_{}'.format(top_k)])),1.96*np.std(np.array(overall_avg['train_ndcg1_{}'.format(top_k)])),np.average(np.array(overall_avg['train_ndcg2_{}'.format(top_k)])),1.96*np.std(np.array(overall_avg['train_ndcg2_{}'.format(top_k)])),np.average(np.array(overall_avg['test_ndcg1_{}'.format(top_k)])),1.96*np.std(np.array(overall_avg['test_ndcg1_{}'.format(top_k)])),np.average(np.array(overall_avg['test_ndcg2_{}'.format(top_k)])),1.96*np.std(np.array(overall_avg['test_ndcg2_{}'.format(top_k)]))))
# f.write('Test avg p1: {} \t {}\tTest avg p2: {}\t{}\t Test avg r1: {} \t {}\tTest avg r2: {}\t{}\t\n'.format(np.average(np.array(overall_avg['test_p1_{}'.format(top_k)])),1.96*np.std(np.array(overall_avg['test_p1_{}'.format(top_k)])),np.average(np.array(overall_avg['test_p2_{}'.format(top_k)])),1.96*np.std(np.array(overall_avg['test_p2_{}'.format(top_k)])),np.average(np.array(overall_avg['test_r1_{}'.format(top_k)])),1.96*np.std(np.array(overall_avg['test_r1_{}'.format(top_k)])),np.average(np.array(overall_avg['test_r2_{}'.format(top_k)])),1.96*np.std(np.array(overall_avg['test_r2_{}'.format(top_k)]))))
# f.write("{}\t{}\t{}\t{}\t{}\t{}\t{}\n".format(mu,alpha,top_k, ndcg1,ndcg2,(ndcg2-ndcg1)/ndcg1,tt(ndcg1_arr,ndcg2_arr)))
# print("Elapsed: {} seconds".format(round(time.time() - start_time, 4)))
# print(tt(ndcg1_arr,ndcg2_arr))