forked from davidlizeng/cs224w-project
-
Notifications
You must be signed in to change notification settings - Fork 0
/
comtagcombine.py
260 lines (233 loc) · 9.02 KB
/
comtagcombine.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
import codecs
import loadData as ld
import json
import nbtext
import tagaffinity
import similarity
import wordvectors
import usergraph
posts_body_file = 'data/posts-body.csv'
posts_bodies = codecs.open(posts_body_file, 'r', 'utf-8')
# Get top tags. Diction of <int, int>: tagID to count in first 50,000 questions
# Only includes tags which appeared more than 50 times.
tcount_infile = open('tcount-50000.txt', 'r')
tempTopTags = json.load(tcount_infile)
tcount_infile.close()
topTags = {}
for idStr, count in tempTopTags.iteritems():
topTags[int(idStr)] = count
# Dict of question ID (int) to set of tag IDs (ints) which include synonyms
questionTagsSyn = {}
# Dict of question ID (int) to size (int) of question tag list only including popular tags
questionTagsPopOnly = {}
# Multinomial Naive Bayes Model:
mnbd = {}
multiLabelDCache = {}
multiLabelD = {}
# Tag affinity model: precomputed for each fold iteration
tagaff_ttas = {}
tagTermDCache = {}
tagTermD = {}
# Similarity model: precomputed for each fold iteration
sim_model = {}
similarityDCache = {}
similarityD = {}
# Community model: based on user to tag network
usertagnetwork = {}
communityDCache = {}
def setQuestionModelModifications(allQuestions):
global questionTagsSyn
global questionTagsPopOnly
print 'Setting question modifications'
infile = open('synonyms.txt', 'r')
tagSyns = {}
while True:
s = infile.readline()
if not s:
break
data = set([int(d) for d in s.split()])
for tid in data:
assert tid in topTags
tagSyns[tid] = data
for qid, question in allQuestions.iteritems():
currentTags = question.tags
newTags = set()
count = 0
for tid in currentTags:
newTags.add(tid)
if tid in tagSyns:
newTags |= tagSyns[tid]
if tid in topTags:
count += 1
questionTagsSyn[qid] = newTags
questionTagsPopOnly[qid] = count
print 'Done with modifications'
def resetModels():
global tagaff_ttas
global tagTermDCache
global mnbd
global multiLabelDCache
global sim_model
global similarityDCache
global usertagnetwork
global communityDCache
tagaff_ttas = {}
tagTermDCache = {}
mnbd = {}
multiLabelDCache = {}
sim_model = {}
similarityDCache = {}
usertagnetwork = {}
communityDCache = {}
def computeComTagCombineD(alpha, beta, gamma, delta, question, trainQuestions):
comTagCombineD = {}
# Tag Term. Stores stuff in cache so we don't have to recompute more than once per question.
if question.id in tagTermDCache:
multiLabelD = multiLabelDCache[question.id]
tagTermD = tagTermDCache[question.id]
similarityD = similarityDCache[question.id]
communityD = communityDCache[question.id]
else:
posts_bodies.seek(question.bodyByte)
body = posts_bodies.readline()
multiLabelD = nbtext.getProbForQuestion(question.id, mnbd, topTags, wordVecs)
multiLabelDCache[question.id] = multiLabelD
tagTermD = tagaffinity.getTagTermBasedRankingScores(body, tagaff_ttas, topTags)
tagTermDCache[question.id] = tagTermD
similarityD = similarity.getSimilarityRankingScores(question.id, trainQuestions, wordVecs, sim_model, topTags)
similarityDCache[question.id] = similarityD
communityD = usergraph.getTagPredictions(question.id, question, usertagnetwork, topTags)
communityDCache[question.id] = communityD
for t in topTags:
comTagCombineD[t] = (alpha * multiLabelD.get(t, 0.0)) + (beta * similarityD.get(t, 0.0)) +\
(gamma * tagTermD.get(t, 0.0)) + (delta * communityD.get(t, 0.0))
return comTagCombineD
# Returns tuple of (scoreAllTags, scorePopularTagsOnly)
def recall_k(k, topTags, actualTags, qid, denominator):
recall_score = 0.0
num = min(len(topTags),k)
for i in xrange(num):
if topTags[i] in actualTags:
recall_score += 1.0
r = recall_score / denominator
if questionTagsPopOnly[qid] == 0:
rPop = 0
else:
rPop = recall_score / questionTagsPopOnly[qid]
return (r, rPop)
def updateParameters(alpha, beta, gamma, delta, bestParams):
bestParams[0] = alpha
bestParams[1] = beta
bestParams[2] = gamma
bestParams[3] = delta
"""
Takes in a training set of questions and trains alpha, beta, gamma, delta for
recall@5 and recall@10. Returns the best params for recall@5 and recall@10 and
the corresponding parameters
"""
def comTagCombineModelTrain(trainQuestions):
global tagaff_ttas
global mnbd
global sim_model
global usertagnetwork
print "Begin Naive Bayes Training"
mnbd = nbtext.getTagNaiveBayesScores(trainQuestions, topTags, wordToIndex, wordVecs)
print "Naive Bayes Training Complete"
print "Begin Tag Affinity Training"
tagaff_ttas = tagaffinity.getTagTermAffinityScores(trainQuestions, includeCounts=False)
print "Tag Affinity Training Complete"
sim_model = similarity.similarityModel(trainQuestions, wordVecs)
print "Similarity Training Complete"
usertagnetwork = usergraph.createUserGraph(trainQuestions)
print "Community Training Complete"
"""
Evaluates the recall@5 and recall@10 scores using the input parameters for each
on the input test data set. Returns the recall@5 score and the recall@10 score.
"""
def comTagCombineModelTest(trainQuestions, testQuestions, outfile=None):
best_r5_avg, best_r10_avg = 0.0, 0.0
bestParams5 = [-1.0, -1.0, -1.0, -1.0]
bestParams10 = [-1.0, -1.0, -1.0, -1.0]
for alpha in xrange(6):
alpha = alpha * 0.2
for beta in xrange(6):
beta = beta * 0.2
for gamma in xrange(6):
gamma = gamma * 0.2
for delta in xrange(6):
delta = delta * 0.2
(r5_sum, r5pop_sum) = (0.0, 0.0)
(r10_sum, r10pop_sum) = (0.0, 0.0)
(r5syn_sum, r5synpop_sum) = (0.0, 0.0)
(r10syn_sum, r10synpop_sum) = (0.0, 0.0)
for (qid, question) in testQuestions.iteritems():
comTagCombineD = computeComTagCombineD(alpha, beta, gamma, delta, question, trainQuestions)
sortedTags = sorted(comTagCombineD, key=lambda x: comTagCombineD[x], reverse=True)
num = min(10, len(sortedTags))
topTags = sortedTags[0:num]
(r5, r5pop) = recall_k(5, topTags, question.tags, qid, len(question.tags))
(r10, r10pop) = recall_k(10, topTags, question.tags, qid, len(question.tags))
(r5syn, r5synpop) = recall_k(5, topTags, questionTagsSyn[qid], qid, len(question.tags))
(r10syn, r10synpop) = recall_k(10, topTags, questionTagsSyn[qid], qid, len(question.tags))
r5_sum += r5
r5pop_sum += r5pop
r5syn_sum += r5syn
r5synpop_sum += r5synpop
r10_sum += r10
r10pop_sum += r10pop
r10syn_sum += r10syn
r10synpop_sum += r10synpop
r5_avg = r5_sum / len(testQuestions)
r5pop_avg = r5pop_sum / len(testQuestions)
r5syn_avg = r5syn_sum / len(testQuestions)
r5synpop_avg = r5synpop_sum / len(testQuestions)
r10_avg = r10_sum / len(testQuestions)
r10pop_avg = r10pop_sum / len(testQuestions)
r10syn_avg = r10syn_sum / len(testQuestions)
r10synpop_avg = r10synpop_sum / len(testQuestions)
if r5_avg > best_r5_avg:
best_r5_avg = r5_avg
updateParameters(alpha, beta, gamma, delta, bestParams5)
if r10_avg > best_r10_avg:
best_r10_avg = r10_avg
updateParameters(alpha, beta, gamma, delta, bestParams10)
r5_tuple = (r5_avg, r5pop_avg, r5syn_avg, r5synpop_avg)
r10_tuple = (r10_avg, r10pop_avg, r10syn_avg, r10synpop_avg)
print '(%f, %f, %f, %f): r5 = %s, r10 = %s' % (alpha, beta, gamma, delta, r5_tuple, r10_tuple)
if outfile:
outfile.write('%f,%f,%f,%f,%s,%s\n' % (alpha, beta, gamma, delta, r5_tuple, r10_tuple))
return bestParams5, best_r5_avg, bestParams10, best_r10_avg
## Comment out this entire block if not running from Python shell
ld.loadData(True)
# This function must be run. Be careful if this is commented out.
setQuestionModelModifications(ld.questions)
folds = ld.getCVFolds()
print 'Generating word vectors'
frequentWords, wordToIndex = wordvectors.getFrequentWords(ld.questions)
wordVecs = wordvectors.getWordVectors(ld.questions, wordToIndex)
## End block
counter = 0
recall_test_scores = [0.0, 0.0]
for fold in folds[0:5]:
resetModels()
counter += 1
print 'Starting Fold %d' % counter
trainQuestions = fold[0]
print 'Fold size %d' % len(fold[0])
comTagCombineModelTrain(trainQuestions)
print 'Train complete. Beginning test.'
testQuestions = fold[1]
outfile = open('temp/ctc-out_%d.csv' % counter, 'w+')
params5, score5, params10, score10 = comTagCombineModelTest(trainQuestions, testQuestions, outfile)
print "Test Values for Run #%d:" % counter
print "recall@5 params: " + str(params5)
print "recall@5 score: " + str(score5)
print "recall@10 params: " + str(params10)
print "recall@10 score: " + str(score10)
recall_test_scores[0] += score5
recall_test_scores[1] += score10
outfile.close()
print "Average Test Scores:"
print "recall@5 score: " + str(recall_test_scores[0]/10)
print "recall@10 score: " + str(recall_test_scores[1]/10)
posts_bodies.close()