/
middleware.py
185 lines (156 loc) · 7.58 KB
/
middleware.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
from tqdm import tqdm
from sentence_splitter import SentenceSplitter
import spacy
import re
import torch
# from itertools import chain
NLP = spacy.blank("en")
NLP.add_pipe(NLP.create_pipe("sentencizer"))
"""
THIS FILE MAINLY CONTAINS FUNCTIONS THAT COMMUNICATE BETWEEN MAIN AND RETRIEVER/WORKER
"""
def getDocumentContentFromDict(docid, COLLECTION_DICT):
return COLLECTION_DICT[docid]
def getBatchDocumentContentFromDict(docids, COLLECTION_DICT):
result = []
for docid in docids:
result.append(COLLECTION_DICT[docid].lower())
return result
def getBatchSplittedDocumentSentences(docids, COLLECTION_DICT):
result = []
for docid in docids:
splitter = SentenceSplitter(language='en')
document = COLLECTION_DICT[docid]
sentences = splitter.split(text=document)
for sentence in sentences:
result.append((docid, sentence))
return result
def getNLPbatchSlidingWindows(docids, COLLECTION_DICT):
windows = []
ids = []
for docid in docids:
doc_text = COLLECTION_DICT[docid]
doc = NLP(doc_text[:1000000])
sentences = [sent.string.strip() for sent in doc.sents]
for i in range(0, len(sentences), 5):
segment = ' '.join(sentences[i:i + 10])
segment = re.sub(r'^#*', '', segment)
segment += ' </s>'
windows.append(segment)
ids.append(docid)
if i + 10 >= len(sentences):
break
return windows, ids
def rerankDocuments(RANKED_FILE_CONTENT, COLLECTION_DICT, CONF, SCONF, QUERY, topK, worker, workerNum):
# Construct the sorted re-ranked list
resPath = SCONF["RESULT"]
for query in tqdm(RANKED_FILE_CONTENT, desc="Process Query With Worker " + str(workerNum)):
queryCollection = []
if topK > len(query):
temp = len(query)
else:
temp = topK
for i in tqdm(range(temp), desc="Process Document With Worker " + str(workerNum)):
document = query[i]
docid = document[1]
qid = document[0]
rank = document[2]
docContents = getDocumentContentFromDict(docid, COLLECTION_DICT)
queryContents = QUERY[qid]
score = worker.predict(docContents, queryContents, SCONF)
queryCollection.append([qid, docid, rank, score])
sortedQueryCollection = sorted(queryCollection, key=lambda l: l[3], reverse=True)
lines = []
for inde, document in enumerate(sortedQueryCollection):
line = document[0] + "\t" + document[1] + "\t" + str(inde + 1) + "\t" + str(document[3]) + "\n"
lines.append(line)
with open("{}{}/{}_rerank_{}-{}.res".format(resPath, CONF["MODEL_NAME"], CONF["MODEL_NAME"], topK, workerNum), "a+") as f:
f.writelines(lines)
def batchRerankDocuments(RANKED_FILE_CONTENT, COLLECTION_DICT, CONF, SCONF, QUERY, topK, worker, workerNum):
# Construct the sorted re-ranked list
resPath = SCONF["RESULT"]
for query in tqdm(RANKED_FILE_CONTENT, desc="Process Query With Worker " + str(workerNum)):
qid = query[0][0]
queryContents = QUERY[qid]
batchSize = 64
encoded_decoder_inputs = worker.tokenize([queryContents] * batchSize)
if topK > len(query):
temp = len(query)
else:
temp = topK
if CONF["METHOD"] == "PASS":
queryCollection = []
numIter = temp // batchSize + 1
for iter in tqdm(range(numIter), desc="Process Document With Worker " + str(workerNum)):
start = iter * batchSize
end = (iter + 1) * batchSize
if end > temp:
end = temp
encoded_decoder_inputs = worker.tokenize([queryContents] * (end - start))
docids = query[start:end, 1]
ranks = query[start:end, 2]
batchDocContents = getBatchDocumentContentFromDict(docids, COLLECTION_DICT)
scores = worker.batchPredict(batchDocContents, encoded_decoder_inputs, SCONF)
for i in range(len(scores)):
queryCollection.append([qid, docids[i], ranks[i], scores[i]])
sortedQueryCollection = sorted(queryCollection, key=lambda l: l[3], reverse=True)
lines = []
for inde, document in enumerate(sortedQueryCollection):
line = document[0] + "\t" + document[1] + "\t" + str(inde + 1) + "\t" + str(document[3]) + "\n"
lines.append(line)
with open("{}{}/{}_rerank_JelinekMercer_{}-{}_lower.res".format(resPath, CONF["MODEL_NAME"], CONF["MODEL_NAME"], topK, workerNum), "a+") as f:
f.writelines(lines)
else:
docids = query[:temp, 1]
windows, ids = getNLPbatchSlidingWindows(docids, COLLECTION_DICT)
num_sentences = len(ids)
numIter = num_sentences // batchSize + 1
if num_sentences % batchSize == 0:
numIter -= 1
temp_socres = torch.Tensor([0] * num_sentences)
for iter in tqdm(range(numIter), desc="Process Document With Worker " + str(workerNum)):
start = iter * batchSize
end = (iter + 1) * batchSize
if end > num_sentences:
end = num_sentences
encoded_decoder_inputs = worker.tokenize([queryContents] * (end-start))
scores = worker.batchPredict(windows[start:end], encoded_decoder_inputs, SCONF)
temp_socres[start:end] = scores[:]
temp_socres = temp_socres.tolist()
with open("result/doc_rerank/t5/sentence_scoring/t5_doc_sentence_scoring.txt", "a+") as f:
for i in range(len(ids)):
f.write("{}\t{}\t{}\n".format(qid, ids[i], temp_socres[i]))
# fullSentenceList = list(chain.from_iterable(documentInSentences))
# CHUNK_SIZE = SCONF["CHUNK_SIZE"]
#
# chunkedSentenceList = [fullSentenceList[i:i + CHUNK_SIZE] for i in range(0, len(fullSentenceList), CHUNK_SIZE)]
# positionList = [{} for _ in range(len(chunkedSentenceList))]
#
# index = 0
#
# for ind, sentences in enumerate(documentInSentences):
# total = sum(positionList[index].values())
# if len(sentences) + total < CHUNK_SIZE:
# positionList[index][docids[ind]] = len(sentences)
# elif len(sentences) + total == CHUNK_SIZE:
# positionList[index][docids[ind]] = len(sentences)
# index += 1
# else:
# need = CHUNK_SIZE - total
# rest = total + len(sentences) - CHUNK_SIZE
# positionList[index][docids[ind]] = need
# index += 1
# positionList[index][docids[ind]] = rest
#
# for ind, sentences in enumerate(chunkedSentenceList):
# count = 0
# scores = worker.batchPredict(sentences, queryContents, SCONF)
# innerCount = 0
# positionItem = positionList[ind]
# positionKeys = positionItem.keys()
# positionValues = positionItem.values()
# for i in range(positionValues[innerCount]):
# with open("result/doc_rerank/t5/sentence_scoring/t5_doc_sentence_scoring.txt", "a+") as f:
# f.write("{}\t{}_{}\t{}\n".format(qid, positionKeys[innerCount], i, scores[count]))
# count += 1
# innerCount += 1