-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluation.py
426 lines (345 loc) · 15.8 KB
/
evaluation.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
413
414
415
416
417
418
419
420
421
422
423
424
425
426
import pdb
import torch
import torch.utils.data as torch_data
import numpy as np
import re
import nltk
nltk.download('stopwords')
from nltk.corpus import stopwords
from scipy import sparse
from sklearn.feature_extraction.text import TfidfVectorizer
from tqdm import tqdm
import time
class BM25Dataset(torch_data.Dataset):
def __init__(self, querys, datebase, vectorizer, avdl, b=0.75, k1=1.6):
super().__init__()
self.querys = querys
self.datebase = datebase
self.vectorizer = vectorizer
self.b = b
self.k1 = k1
self.avdl = avdl
self.X = None
self.len_X = None
def __len__(self):
return self.querys.shape[0]
def __getitem__(self, item):
one_q_vector = self.querys[item]
database_vectors = self.datebase
idf = self.vectorizer._tfidf.idf_[None, one_q_vector.indices] - 1.
b, k1, avdl = self.b, self.k1, self.avdl
device = torch.device("cpu")
if self.len_X is None:
self.len_X = database_vectors.sum(1).A1
len_X = self.len_X
with torch.no_grad():
database_vectors = torch.Tensor(database_vectors[:, one_q_vector.indices].toarray()).to(device)
len_X = torch.Tensor(len_X).to(device)
idf = torch.Tensor(idf).to(device)
denom = database_vectors + (k1 * (1 - b + b * len_X / avdl))[:, None]
numer = database_vectors.mul(idf.repeat(database_vectors.shape[0], 1)) * (k1 + 1)
bm25_score = (numer / denom).sum(1)
return bm25_score, item
class BM25(object):
def __init__(self, b=0.75, k1=1.6, stop_words=stopwords.words('english')):
self.vectorizer = TfidfVectorizer(norm=None, smooth_idf=False,
stop_words=stop_words)
self.b = b
self.k1 = k1
self.X = None
self.len_X = None
def fit(self, X):
""" Fit IDF to documents X """
self.vectorizer.fit(X)
y = super(TfidfVectorizer, self.vectorizer).transform(X)
self.avdl = y.sum(1).mean()
def vector_transform(self, one_q_vector, database_vectors, idf, device=torch.device("cpu")):
""" Calculate BM25 between query q and documents X
store_X: 是否存储 X 计算中间变量,加速
"""
b, k1, avdl = self.b, self.k1, self.avdl
if self.len_X is None:
self.len_X = database_vectors.sum(1).A1
len_X = self.len_X
# idf(t) = log [ n / df(t) ] + 1 in sklearn, so it need to be coneverted
# to idf(t) = log [ n / df(t) ] with minus 1
# idf = self.vectorizer._tfidf.idf_[None, one_q_vector.indices] - 1.
# idf = torch.Tensor([self.vectorizer._tfidf.idf_[q_vector.indices] - 1 for q_vector in q_vectors])
# ********************GPU begin*************************
with torch.no_grad():
database_vectors = torch.Tensor(database_vectors[:, one_q_vector.indices].toarray()).to(device)
len_X = torch.Tensor(len_X).to(device)
idf = torch.Tensor(idf).to(device)
denom = database_vectors + (k1 * (1 - b + b * len_X / avdl))[:, None]
numer = database_vectors.mul(idf.repeat(database_vectors.shape[0], 1)) * (k1 + 1)
bm25_score = (numer / denom).sum(1)
# ********************GPU end*************************
return bm25_score
def bm25_sim(self, query_list: list, datebase: list, device=torch.device("cpu"), num_workers=20):
datebase = super(TfidfVectorizer, self.vectorizer).transform(datebase)
querys = super(TfidfVectorizer, self.vectorizer).transform(query_list)
assert sparse.isspmatrix_csr(querys[0])
dataset = BM25Dataset(querys, datebase, self.vectorizer, self.avdl)
dataloader = torch_data.dataloader.DataLoader(
dataset=dataset, batch_size=1, shuffle=False, num_workers=num_workers)
score = torch.zeros((querys.shape[0], datebase.shape[0]))
for doc_scores, item in tqdm(dataloader):
score[item] = doc_scores
return score
def bm25_sim_with_time(self, query_list: list, datebase: list):
datebase = super(TfidfVectorizer, self.vectorizer).transform(datebase)
querys = super(TfidfVectorizer, self.vectorizer).transform(query_list)
assert sparse.isspmatrix_csr(querys[0])
dataset = BM25Dataset(querys, datebase, self.vectorizer, self.avdl)
dataloader = torch_data.dataloader.DataLoader(dataset=dataset, batch_size=1, shuffle=False, num_workers=16)
score = torch.zeros((querys.shape[0], datebase.shape[0]))
start_time = None
for doc_scores, item in tqdm(dataloader):
if start_time is None:
start_time = time.time()
score[item] = doc_scores
end_time = time.time()
return score, end_time-start_time
def bm25_sim_old(self, query_list: list, datebase: list, device=torch.device("cpu")):
score = None
doc_scores_list = []
datebase = super(TfidfVectorizer, self.vectorizer).transform(datebase)
querys = super(TfidfVectorizer, self.vectorizer).transform(query_list)
assert sparse.isspmatrix_csr(querys[0])
with tqdm(total=querys.shape[0]) as pbar:
for i, query in enumerate(querys):
pbar.update(1)
# tokenized_query = query.split(" ")
doc_scores = self.vector_transform(query, datebase, device=device).unsqueeze(0)
doc_scores_list.append(doc_scores)
if (i+1) % 200 == 0:
temp = torch.cat(doc_scores_list, dim=0).cpu()
doc_scores_list = []
if score is None:
score = temp
else:
score = torch.cat((score, temp), dim=0)
if len(doc_scores_list) > 0:
score = torch.cat((score, torch.cat(doc_scores_list, dim=0).cpu()), dim=0)
return score
def vector_transform_old(self, one_q_vector, database_vectors, device=torch.device("cpu")):
""" Calculate BM25 between query q and documents X
store_X: 是否存储 X 计算中间变量,加速
"""
b, k1, avdl = self.b, self.k1, self.avdl
if self.len_X is None:
self.len_X = database_vectors.sum(1).A1
len_X = self.len_X
# idf(t) = log [ n / df(t) ] + 1 in sklearn, so it need to be coneverted
# to idf(t) = log [ n / df(t) ] with minus 1
idf = self.vectorizer._tfidf.idf_[None, one_q_vector.indices] - 1.
# convert to csc for better column slicing "https://zhuanlan.zhihu.com/p/36122299"
# X = X.tocsc()[:, q.indices]
# denom = X + (k1 * (1 - b + b * len_X / avdl))[:, None]
# numer = X.multiply(np.broadcast_to(idf, X.shape)) * (k1 + 1)
# bm25_score = (numer / denom).sum(1).A1
# ********************GPU begin*************************
with torch.no_grad():
database_vectors = torch.Tensor(database_vectors[:, one_q_vector.indices].toarray()).to(device)
len_X = torch.Tensor(len_X).to(device)
idf = torch.Tensor(idf).to(device)
denom = database_vectors + (k1 * (1 - b + b * len_X / avdl))[:, None]
numer = database_vectors.mul(idf.repeat(database_vectors.shape[0], 1)) * (k1 + 1)
bm25_score = (numer / denom).sum(1)
# ********************GPU end*************************
return bm25_score
def transform_old(self, q, X, store_X=True, device=torch.device("cpu")):
""" Calculate BM25 between query q and documents X
store_X: 是否存储 X 计算中间变量,加速
"""
b, k1, avdl = self.b, self.k1, self.avdl
# apply CountVectorizer
if store_X:
if self.X is None:
self.X = super(TfidfVectorizer, self.vectorizer).transform(X)
X = self.X
else:
X = self.X
else:
X = super(TfidfVectorizer, self.vectorizer).transform(X)
len_X = X.sum(1).A1
q, = super(TfidfVectorizer, self.vectorizer).transform([q])
assert sparse.isspmatrix_csr(q)
# idf(t) = log [ n / df(t) ] + 1 in sklearn, so it need to be coneverted
# to idf(t) = log [ n / df(t) ] with minus 1
idf = self.vectorizer._tfidf.idf_[None, q.indices] - 1.
# convert to csc for better column slicing "https://zhuanlan.zhihu.com/p/36122299"
# X = X.tocsc()[:, q.indices]
# denom = X + (k1 * (1 - b + b * len_X / avdl))[:, None]
# numer = X.multiply(np.broadcast_to(idf, X.shape)) * (k1 + 1)
# bm25_score = (numer / denom).sum(1).A1
# ********************GPU begin*************************
with torch.no_grad():
X = torch.Tensor(X[:, q.indices].toarray()).to(device)
len_X = torch.Tensor(len_X).to(device)
idf = torch.Tensor(idf).to(device)
denom = X + (k1 * (1 - b + b * len_X / avdl))[:, None]
numer = X.mul(idf.repeat(X.shape[0], 1)) * (k1 + 1)
bm25_score = (numer / denom).sum(1)
# ********************GPU end*************************
return bm25_score
def jaccard_sim(query, retrieval_base, device=torch.device("cpu")):
with torch.no_grad():
query, retrieval_base = torch.Tensor(query), torch.Tensor(retrieval_base)
query = (query > 0).float()
retrieval_base = (retrieval_base > 0).float().to(device)
eps = 1e-8
score = None
for each_query in tqdm(torch.chunk(query, len(query)//48 + 1, dim=0)):
# ***************GPU begin******************
each_query = each_query.to(device)
score_temp_list = []
for each_retrieval_base in torch.chunk(retrieval_base, len(retrieval_base)//2048 + 1, dim=0):
base_num = each_retrieval_base.size(0)
each_query1 = each_query.unsqueeze(1).repeat(1, base_num, 1)
intersection = torch.min(each_query1, each_retrieval_base).sum(-1)
union = torch.max(each_query1, each_retrieval_base).sum(-1) + eps
score_temp1 = (intersection / union)
score_temp_list.append(score_temp1)
score_temp = torch.cat(score_temp_list, dim=1)
# ***************GPU end******************
score_temp = score_temp.cpu()
if score is None:
score = score_temp
else:
score = torch.cat((score, score_temp), dim=0)
return score
def jaccard_sim1(query, retrieval_base, device=torch.device("cpu")):
query, retrieval_base = torch.Tensor(query), torch.Tensor(retrieval_base)
query = (query > 0).float()
retrieval_base = (retrieval_base > 0).float()
eps = 1e-8
score = None
base_num = retrieval_base.size(0)
for each in query:
each = each.unsqueeze(0).repeat(base_num, 1)
intersection = torch.min(each, retrieval_base).sum(-1)
union = torch.max(each, retrieval_base).sum(-1) + eps
score_temp = (intersection / union).unsqueeze(0)
if score is None:
score = score_temp
else:
score = torch.cat((score, score_temp), dim=0)
return score
def l2norm(X, eps=1e-13, dim=-1):
"""L2-normalize columns of X
"""
norm = torch.pow(X, 2).sum(dim=dim, keepdim=True).sqrt() + eps + 1e-14
X = torch.div(X, norm)
return X
def l1norm(X, eps=1e-13, dim=-1):
"""L2-normalize columns of X
"""
norm = torch.abs(X).sum(dim=dim, keepdim=True) + eps + 1e-14
X = torch.div(X, norm)
return X
def z_norm(a, eps=1e-12, dim=-1, type="z"):
# Z-score standardization
mean_a = torch.mean(a)
std_a = torch.std(a)
n1 = (a - mean_a) / std_a
# print(n1, torch.mean(n1), torch.std(n1))
if type == 'Min-Max':
# Min-Max scaling
min_a = torch.min(a, dim=dim, keepdim=True).values
max_a = torch.max(a, dim=dim, keepdim=True).values
n = (a - min_a) / (max_a+eps)
else:
# Z-score standardization
mean_a = torch.mean(a, dim=dim, keepdim=True)
std_a = torch.std(a, dim=dim, keepdim=True)
n = (a - mean_a) / (std_a+eps)
return n
def cosine_sim(query, retrio, device=torch.device("cpu")):
"""Cosine similarity between all the query and retrio pairs
"""
query, retrio = l2norm(query), l2norm(retrio) # m*d, n*d
score = None
retrio = retrio.to(device)
for each_query in tqdm(torch.chunk(query, len(query)//256 + 1, dim=0)):
# ************GPU begin**************
with torch.no_grad():
each_query = each_query.to(device) # 48*d
score_temp = each_query.mm(retrio.t()) # 48*n
# ************GPU end**************
score_temp = score_temp.cpu()
score = score_temp if score is None else torch.cat((score, score_temp), 0)
return score
def evaluate(label_matrix):
label_matrix = label_matrix.astype(int)
ranks = np.zeros(label_matrix.shape[0])
aps = np.zeros(label_matrix.shape[0])
for index in range(len(ranks)):
rank = np.where(label_matrix[index] == 1)[0] + 1
ranks[index] = rank[0]
aps[index] = np.mean([(i + 1.) / rank[i] for i in range(len(rank))])
r1, r5, r10, r100, r1000, r5000 = [100.0 * np.mean([x <= k for x in ranks]) for k in [1, 5, 10, 100, 1000, 5000]]
medr = np.floor(np.median(ranks))
meanr = ranks.mean()
mir = (1.0 / ranks).mean()
mAP = aps.mean()
return (r1, r5, r10, r100, r1000, r5000, medr, meanr, mir, mAP)
def recall_eval(scores, code_ids, query_ids, num_workers=16, new_query_ids=None):
print("Get Recall")
# scores, code_ids, query_ids, new_query_ids = graph_code_bert1.score, graph_code_bert1.code_ids, graph_code_bert1.query_ids, graph_code_bert1.new_query_ids,
scores = np.array(scores)
code_ids = code_ids.copy()
query_ids = query_ids.copy()
# 转换成字符型
for idx in range(len(code_ids)):
code_ids[idx] = str(code_ids[idx])
for idx in range(len(query_ids)):
query_ids[idx] = str(query_ids[idx])
if new_query_ids is not None:
for each in new_query_ids[idx][1:]:
code_ids[each] = str(new_query_ids[idx][0])
# 开始计算
inds = np.argsort(scores, axis=1)
label_matrix = np.zeros(inds.shape)
# for index in tqdm(torch_data.dataloader.DataLoader(
# dataset=list(range(inds.shape[0])), batch_size=1, shuffle=False, num_workers=num_workers)):
for index in list(range(inds.shape[0])):
index = int(index)
ind = inds[index][::-1]
gt_index = np.where(np.array(code_ids)[ind] == query_ids[index].split('#')[0])[0]
label_matrix[index][gt_index] = 1
(r1, r5, r10, r100, r1000, r5000, medr, meanr, mir, mAP) = evaluate(label_matrix)
output = {
'text': "%.5f \t%.5f \t%.5f \t%.5f \t%.5f" % (r1, r5, r10, r100, r1000),
'tuple': (r1, r5, r10, r100, r1000, medr, mAP)
}
return output
def get_mrr(scores, code_ids, query_ids, device=torch.device("cpu"), new_query_ids=None):
scores = scores.to(device)
mrr = 0
mrr_list = []
print("Get MRR")
for query_idx in tqdm(range(len(scores))):
rank = torch.argsort(-scores[query_idx]).tolist()
if new_query_ids is not None:
codeId2Rank = {}
for each in range(len(rank)):
codeId2Rank[rank[each]] = each
try:
items = [1 / (codeId2Rank[int(each)] + 1) for each in new_query_ids[query_idx]]
except:
pdb.set_trace()
item = max(items)
else:
try:
item = 1 / (rank[0:3000].index(int(query_ids[query_idx])) + 1)
except:
item = 0
mrr += item
mrr_list.append(item)
mrr = mrr / len(scores)
return_dict = {
"mrr": mrr,
"mrr_list": mrr_list
}
return return_dict