-
Notifications
You must be signed in to change notification settings - Fork 6
/
RL4SRD.py
executable file
·324 lines (261 loc) · 13.6 KB
/
RL4SRD.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
"""
Created on 2016-12-11
class: RL4SRD
@author: Long Xia
@contact: xl.1988.life@gmail.com
"""
# !/usr/bin/python
# -*- coding:utf-8 -*-
import sys
import json
import yaml
import copy
import math
import random
import numpy as np
import os
class RL4SRD(object):
"""docstring for RL4SRD"""
def __init__(self, fileQueryPermutaion, fileQueryRepresentation, fileDocumentRepresentation, fileQueryDocumentSubtopics, folder):
super(RL4SRD, self).__init__()
with open(fileQueryPermutaion) as self.fileQueryPermutaion:
self.dictQueryPermutaion = json.load(self.fileQueryPermutaion)
with open(fileQueryRepresentation) as self.fileQueryRepresentation:
self.dictQueryRepresentation = json.load(self.fileQueryRepresentation)
for query in self.dictQueryRepresentation:
self.dictQueryRepresentation[query] = np.matrix([self.dictQueryRepresentation[query]], dtype=np.float)
self.dictQueryRepresentation[query] = np.transpose(self.dictQueryRepresentation[query])
with open(fileDocumentRepresentation) as self.fileDocumentRepresentation:
self.dictDocumentRepresentation = json.load(self.fileDocumentRepresentation)
for doc in self.dictDocumentRepresentation:
self.dictDocumentRepresentation[doc] = np.matrix([self.dictDocumentRepresentation[doc]], dtype=np.float)
self.dictDocumentRepresentation[doc] = np.transpose(self.dictDocumentRepresentation[doc])
with open(fileQueryDocumentSubtopics) as self.fileQueryDocumentSubtopics:
self.dictQueryDocumentSubtopics = json.load(self.fileQueryDocumentSubtopics)
self.folder = folder
with open(self.folder + '/config.yml') as self.confFile:
self.dictConf = yaml.load(self.confFile)
self.floatLearningRate = self.dictConf['learning_rate']
self.listTestSet = self.dictConf['test_set']
self.listValidationSet = self.dictConf['validation_set']
self.lenTrainPermutation = self.dictConf['length_train_permutation']
self.K = self.dictConf['K']
self.gamma = self.dictConf['gamma']
self.hidden_dim = self.dictConf['hidden_dim']
self.fileResult = open(self.folder + '/RL_result.dat', 'w')
self.fileValidation = open(self.folder + '/RL_validation.dat', 'w')
self.floatTestTime = 0.0
self.__RL_build__()
def __del__(self):
self.fileResult.close()
self.fileValidation.close()
def __RL_build__(self):
#self.U = np.random.uniform(-np.sqrt(1./self.hidden_dim), np.sqrt(1./self.hidden_dim), (100, self.hidden_dim))
#self.V_q = np.random.uniform(-np.sqrt(1./self.hidden_dim), np.sqrt(1./self.hidden_dim), (self.hidden_dim, 100))
#self.V = np.random.uniform(-np.sqrt(1./self.hidden_dim), np.sqrt(1./self.hidden_dim), (self.hidden_dim, 100))
#self.W = np.random.uniform(-np.sqrt(1./self.hidden_dim), np.sqrt(1./self.hidden_dim), (self.hidden_dim, self.hidden_dim))
self.U = np.random.uniform(-1./self.hidden_dim, 1./self.hidden_dim, (100, self.hidden_dim))
self.V_q = np.random.uniform(-1./self.hidden_dim, 1./self.hidden_dim, (self.hidden_dim, 100))
self.V = np.random.uniform(-1./self.hidden_dim, 1./self.hidden_dim, (self.hidden_dim, 100))
self.W = np.random.uniform(-1./self.hidden_dim, 1./self.hidden_dim, (self.hidden_dim, self.hidden_dim))
def __RL_derive_assign(self):
self.U_derive = np.zeros((100, self.hidden_dim))
self.V_q_derive = np.zeros((self.hidden_dim, 100))
self.V_derive = np.zeros((self.hidden_dim, 100))
self.W_derive = np.zeros((self.hidden_dim, self.hidden_dim))
def __RL_derive_list(self):
self.list_U_derive = []
self.list_V_derive = []
self.list_V_q_derive = []
self.list_W_derive = []
self.list_h = []
self.list_diag_h = []
self.list_G_t = []
def __sigmoid(self, arrayX):
return 1 / (1+np.exp(-arrayX))
def __sigmoid_derive(self, arrayX):
return arrayX * (1-arrayX)
def __sigmoid_diag(self, arrayX):
Ashape = arrayX.shape
tmp = np.zeros(Ashape[0])
for i in xrange(Ashape[0]):
tmp[i] = arrayX[i][0] * ( 1 - arrayX[i][0] )
return np.diag(tmp)
def preference(self, o_list):
tmp = random.random()
for i in xrange(len(o_list)):
if tmp < o_list[i]:
return i
def alphaDCG(self, alpha, query, docList, k):
DCG = 0.0
subtopics = []
for i in xrange(20):
subtopics.append(0)
for i in xrange(k):
G = 0.0
if docList[i] not in self.dictQueryDocumentSubtopics[query]:
continue
listDocSubtopics = self.dictQueryDocumentSubtopics[query][docList[i]]
if len(listDocSubtopics) == 0:
G = 0.0
else:
for subtopic in listDocSubtopics:
G += (1-alpha) ** subtopics[int(subtopic)-1]
subtopics[int(subtopic)-1] += 1
DCG += G/math.log(i+2, 2)
return DCG
def subtopic_recall(self):
pass
def Train(self):
listKeys = self.dictQueryPermutaion.keys()
random.shuffle(listKeys)
for query in listKeys:
#if (int(query) in self.listTestSet) or (int(query) in self.listValidationSet):
if int(query) in self.listTestSet:
continue
self.__RL_derive_assign()
self.__RL_derive_list()
q = self.dictQueryRepresentation[query]
self.h_t = self.__sigmoid(np.dot(self.V_q, q))
listPermutation = copy.deepcopy(self.dictQueryPermutaion[query]['permutation'])
listSelectedSet = []
for t in xrange(self.lenTrainPermutation):
#store h_t and diag(h_t * (1-h_t))
h_t_tmp = copy.deepcopy(self.h_t)
self.list_h.append(h_t_tmp)
self.list_diag_h.append(self.__sigmoid_diag(h_t_tmp))
Z_sum = 0.0
derive_U_sum = np.zeros((100, self.hidden_dim))
derive_V_q_sum = np.zeros((self.hidden_dim, 100))
derive_V_sum = np.zeros((self.hidden_dim, 100))
derive_W_sum = np.zeros((self.hidden_dim, self.hidden_dim))
s_t = np.dot(self.U, self.h_t)
x_list = []
x_prob = []
x_prob_sum = 0.0
x_prob_list = []
list_derive_f_U = []
list_derive_f_V_q = []
list_derive_f_V = []
list_derive_f_W = []
for j in xrange(len(listPermutation)):
x_score = np.exp(np.dot(np.transpose(self.dictDocumentRepresentation[listPermutation[j]]), s_t))
x_score = np.array(x_score)
x_score = x_score[0][0]
Z_sum += x_score
x_list.append(x_score)
derive_f_U_tmp = np.dot(self.dictDocumentRepresentation[listPermutation[j]], np.transpose(self.h_t))
list_derive_f_U.append(derive_f_U_tmp)
derive_U_sum += derive_f_U_tmp * x_score
derive_f_h_tmp = np.dot(np.transpose(self.U), self.dictDocumentRepresentation[listPermutation[j]])
if t == 0:
derive_f_V_q_tmp = np.dot(self.list_diag_h[t], np.dot(derive_f_h_tmp, np.transpose(q)))
list_derive_f_V_q.append(derive_f_V_q_tmp)
derive_V_q_sum += derive_f_V_q_tmp * x_score
else:
derive_f_V_tmp = np.dot(self.list_diag_h[t], np.dot(derive_f_h_tmp, np.transpose(self.dictDocumentRepresentation[listSelectedSet[t-1]])))
derive_f_W_tmp = np.dot(self.list_diag_h[t], np.dot(derive_f_h_tmp, np.transpose(self.list_h[t-1])))
for i in xrange(t-1, 1, -1):
derive_f_h_tmp = np.dot(np.dot(np.transpose(self.W),self.list_diag_h[i]), derive_f_h_tmp)
derive_f_V_tmp += np.dot(self.list_diag_h[i-1], np.dot(derive_f_h_tmp, np.transpose(self.dictDocumentRepresentation[listSelectedSet[i-2]])))
derive_f_W_tmp += np.dot(self.list_diag_h[i-1], np.dot(derive_f_h_tmp, np.transpose(self.list_h[i-1])))
list_derive_f_V.append(derive_f_V_tmp)
list_derive_f_W.append(derive_f_W_tmp)
derive_V_sum += derive_f_V_tmp * x_score
derive_W_sum += derive_f_W_tmp * x_score
#generage policy pi
for item in x_list:
x_one = item/Z_sum
x_prob.append(x_one)
x_prob_sum += x_one
x_prob_list.append(x_prob_sum)
#sample action
preference_j = self.preference(x_prob_list)
listSelectedSet.append(listPermutation[preference_j])
R = self.alphaDCG(0.5, query, listSelectedSet, t+1) - self.alphaDCG(0.5, query, listSelectedSet, t)
self.list_G_t.append(R)
x_t = self.dictDocumentRepresentation[listPermutation[preference_j]]
self.h_t = self.__sigmoid(np.dot(self.V, x_t) + np.dot(self.W, self.h_t))
self.list_U_derive.append(list_derive_f_U[preference_j] - derive_U_sum/Z_sum)
if t == 0:
self.list_V_q_derive.append(list_derive_f_V_q[preference_j] - derive_V_q_sum/Z_sum)
else:
self.list_V_derive.append(list_derive_f_V[preference_j] - derive_V_sum/Z_sum)
self.list_W_derive.append(list_derive_f_W[preference_j] - derive_W_sum/Z_sum)
#X = X/x_t
del listPermutation[preference_j]
for t in xrange(len(self.list_G_t)):
G = 0.0
for j in xrange(t, len(self.list_G_t)):
G += self.list_G_t[j] * (self.gamma ** j)
self.V_q_derive += (self.gamma ** t) * G * self.list_V_q_derive[0]
self.U_derive += (self.gamma ** t) * G * self.list_U_derive[t]
if t > 0:
self.V_derive += (self.gamma ** t) * G * self.list_V_derive[t-1]
self.W_derive += (self.gamma ** t) * G * self.list_W_derive[t-1]
self.U += self.floatLearningRate * self.U_derive
self.V += self.floatLearningRate * self.V_derive
self.V_q += self.floatLearningRate * self.V_q_derive
self.W += self.floatLearningRate * self.W_derive
def Prediction(self, listInput, boolTest):
if not os.path.exists(self.folder + '/ranking'):
os.makedirs(self.folder + '/ranking')
floatSumResultScore = 0.0
dictResult = {}
for query in listInput:
if boolTest:
fileRankingResult = open(self.folder + '/ranking/' + 'test' + str(query) + '.ranking', 'w')
else:
fileRankingResult = open(self.folder + '/ranking/' + 'val' + str(query) + '.ranking', 'w')
listSelectedSet = []
listTest = copy.deepcopy(self.dictQueryPermutaion[str(query)]['permutation'])
idealScore = self.alphaDCG(0.5, str(query), listTest, self.K)
if idealScore == 0:
continue
random.shuffle(listTest)
self.s_t = self.__sigmoid(np.dot(self.V_q, self.dictQueryRepresentation[str(query)]) + np.dot(self.W, np.zeros((self.hidden_dim,1))))
while len(listSelectedSet) < self.K:
bestScore = -10000000.0
bestDoc = ''
s_tmp = self.s_t
for doc in listTest:
o_tmp = np.dot(self.U, s_tmp)
doc_score = np.dot(np.transpose(o_tmp), self.dictDocumentRepresentation[doc])
if doc_score > bestScore and doc not in listSelectedSet:
bestScore = doc_score
bestDoc = doc
self.s_t = self.__sigmoid(np.dot(self.V, self.dictDocumentRepresentation[bestDoc]) + np.dot(self.W, s_tmp))
listSelectedSet.append(bestDoc)
fileRankingResult.write(bestDoc + '\n')
resultScore = self.alphaDCG(0.5, str(query), listSelectedSet, self.K)
dictResult[query] = resultScore/idealScore
floatSumResultScore += resultScore/idealScore
fileRankingResult.close()
print dictResult
return floatSumResultScore/len(dictResult.keys())
def main(self):
iteration = 1
#while (iteration < 151):
while True:
print 'iteration:' + str(iteration)
self.Train()
#validation = self.Prediction(self.listValidationSet, False)
#print round(validation, 4)
#print '\n'
#self.fileValidation.write(str(iteration) + ' ' + str(validation) + '\n')
#self.fileValidation.flush()
result = self.Prediction(self.listTestSet, True)
print 'ndcg:' + str(round(result, 4))
self.fileResult.write(str(iteration) + ' ' + str(result) + '\n')
self.fileResult.flush()
print '\n'
iteration += 1
if __name__ == '__main__':
if len(sys.argv) != 6:
print 'Error: params number is 5!'
print 'Need: query permutation file, query representation file, document representation file, query document subtopics file, and folder!'
sys.exit(-1)
carpe_diem = RL4SRD(sys.argv[1], sys.argv[2], sys.argv[3], sys.argv[4], sys.argv[5])
carpe_diem.main()
del carpe_diem
print 'Game over!'