-
Notifications
You must be signed in to change notification settings - Fork 4
/
_NBPO.py
436 lines (417 loc) · 16 KB
/
_NBPO.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
427
428
429
430
431
432
433
434
435
## Noise robust Bayesian Point-wise Optimization (NBPO)
## Wenhui Yu 2020.04.26
## author @Wenhui Yu, yuwh16@mails.tsinghua.edu.cn
## Parameters setting
dataset = 0 # datasets, 0 for Amazon, 1 for movielens
eta = [0.05, 0.02][dataset] # learning rate
lambda_r = [1, 0.5][dataset] # coefficient of regularization term
lambda_phi = [5, 1][dataset] # coefficient of regularization term
K0 = 50 # length of latent factors
K1 = 50 # length of latent factors for probability
vali_test = 0 # 0 for validation set, 1 for test set
sample_rate = 4 # sample rate, the number of negative samples foreach positive one
batch_size = 5000 # batch size
epoch = 200 # number of epochs
top_k = [2, 5, 10, 20, 50, 100] # top k items to recommend
Model = 'NBPO'
## Paths to save and read
# list for datasets
dataset_list = ['amazon', 'movielens']
path_train = 'dataset\\' + dataset_list[dataset] + '\\train_data.json'
path_validation = 'dataset\\' + dataset_list[dataset] + '\\validation_data.json'
path_test = 'dataset\\' + dataset_list[dataset] + '\\test_data.json'
import numpy as np
from numpy import *
import xlwt
import time
from Library import readdata
from Library import evaluation_F1
from Library import evaluation_NDCG
from Library import save_parameters
from Library import save_df
import xlrd, xlwt
from xlutils.copy import copy as xl_copy
from openpyxl import load_workbook
from openpyxl import Workbook
import pandas as pd
def d(x):
# sigmoid(x)
if x > 10:
return 1
if x < -10:
return 0
if x >= -10 and x <= 10:
return 1.0 / (1.0 + exp(-x))
def test_Model(U, V):
# test the precision
k_num = len(top_k)
# k_num-length list to record F1 and NDCG
F1 = np.zeros(k_num)
NDCG = np.zeros(k_num)
# test all test samples
user_num = 0
for u in range(M):
# the data in test set is [[i, i, i, i],...,[i, i]]
test_item = Test[u]
if len(test_item) > 0:
user_num += 1
# score all items
score = np.dot(V, U[u])
# order
b = zip(score, range(N))
b.sort(key=lambda x: x[0])
order = [x[1] for x in b]
order.reverse()
# remove the training samples from the recommendations
train_positive = train_data_aux[u]
for item in train_positive:
order.remove(item)
# for each k, calculate top_k
for i in range(len(top_k)):
F1[i] += evaluation_F1(order, top_k[i], test_item)
NDCG[i] += evaluation_NDCG(order, top_k[i], test_item)
# calculate the average
F1 = (F1 / user_num).tolist()
NDCG = (NDCG / user_num).tolist()
return F1, NDCG
def train_Model(eta):
# training the model
# initialization
U = np.array([np.array([(random.random() / math.sqrt(K0)) for j in range(K0)]) for i in range(M)])
V = np.array([np.array([(random.random() / math.sqrt(K0)) for j in range(K0)]) for i in range(N)])
P = 5 * np.array([np.array([(random.random() / math.sqrt(K1)) for j in range(K1)]) for i in range(M)])
Q = 5 * np.array([np.array([(-random.random() / math.sqrt(K1)) for j in range(K1)]) for i in range(N)])
e = 10 ** 10
# output a result without training
print 'iteration ', 0,
[F1, NDCG] = test_Model(U, V)
Fmax = 0
if F1[0] > Fmax:
Fmax = F1[0]
print Fmax, 'F1: ', F1, ' ', 'NDCG: ', NDCG
# save in .xls file
F1_df = pd.DataFrame(columns=top_k)
NDCG_df = pd.DataFrame(columns=top_k)
F1_df.loc[0] = F1
NDCG_df.loc[0] = NDCG
save_df([[F1_df, 'F1'], [NDCG_df, 'NDCG']], path_excel, first_sheet=False) # @x
# get the numer of training samples
Re = len(train_data)
# split the training samples with batch_size
bs = range(0, Re, batch_size)
bs.append(Re)
# begin iterating
for ep in range(epoch):
print 'iteration ', ep + 1,
eta = eta * 0.99
# for each iterating, we user all training samples to train
for ii in range(len(bs) - 1):
if abs(U.sum()) < e:
# input the samples as batches
# initialize dU and dC to record the gradient
dU = np.zeros((M, K0))
dV = np.zeros((N, K0))
dP = np.zeros((M, K1))
dQ = np.zeros((N, K1))
for re in range(bs[ii], bs[ii + 1]):
# iterate each batch
# train data, [u, i]
u = train_data[re][0]
i = train_data[re][1]
Ri = np.dot(U[u], V[i])
gammai = np.dot(P[u], Q[i])
# select negative samples randomly
num = 0
while num < sample_rate:
j = int(random.uniform(0, N))
# check if the current sample is positive sample
if not (j in train_data_aux[u]):
num += 1
Rj = np.dot(U[u], V[j])
gammaj = np.dot(P[u], Q[j])
D = d(-gammai) * d(-Ri)
dU[u] += D * V[i]
dV[i] += D * U[u]
D = -d(gammai) * d(Ri)
dP[u] += D * Q[i]
dQ[i] += D * P[u]
D = d(gammaj) * d(-Rj) - d(Rj)
dU[u] += D * V[j]
dV[j] += D * U[u]
D = d(-gammaj) * d(Rj)
dP[u] += D * Q[j]
dQ[j] += D * P[u]
# update the matrices
U += eta * (dU - lambda_r * U)
V += eta * (dV - lambda_r * V)
P += eta * (dP - lambda_phi * P)
Q += eta * (dQ - lambda_phi * Q)
if abs(U.sum()) < e:
[F1, NDCG] = test_Model(U, V)
if F1[0] > Fmax:
Fmax = F1[0]
F1_df.loc[ep + 1] = F1
NDCG_df.loc[ep + 1] = NDCG
print 'F1: ', F1, ' ', 'NDCG: ', NDCG
save_df([[F1_df, 'F1'], [NDCG_df, 'NDCG']], path_excel, first_sheet=False) # @x
else:
break
return F1_df, NDCG_df
def print_parameter():
# print all parameters
print 'model:', Model
print 'dataset:', dataset_list[dataset]
print 'eta:', eta
print 'lambda_r:', lambda_r, 'lambda_phi:', lambda_phi
print 'K0:', K0, 'K1:', K1
print 'vali_test:', ['validation', 'test'][vali_test]
print 'sample_rate:', sample_rate
print 'batch_size:', batch_size
print 'epoch:', epoch
print 'top_k:', top_k
print
'''**************************main_function***************************'''
'''**************************main_function***************************'''
# load the data
[train_data, train_data_aux, M, N] = readdata(path_train)
validation_data = readdata(path_validation)[1]
test_data = readdata(path_test)[1]
# choose validation or test set
if vali_test == 0:
Test = validation_data
else:
Test = test_data
data = [
["Model", [Model]],
["dataset", [dataset_list[dataset]]],
["eta", [eta]],
["lambda_r", [lambda_r]],
["lambda_phi", [lambda_phi]],
["K0", [K0]],
["K1", [K1]],
["vali_test", [['validation', 'test'][vali_test]]],
["sample_rate", [sample_rate]],
["batch_size", [batch_size]],
['epoch', [epoch]],
["top_k", top_k]
]
path_excel = 'experiment_result\\' + dataset_list[dataset] + '_' + Model + '_' + str(int(time.time())) + str(int(random.uniform(100, 900))) + '.xlsx'
print_parameter()
save_parameters(data, path_excel)
F1_df, NDCG_df = train_Model(eta)
# ## Noise robust Bayesian Point-wise Optimization (NBPO)
# ## Wenhui Yu 2020.04.26
# ## author @Wenhui Yu, yuwh16@mails.tsinghua.edu.cn
#
# ## Parameters setting
# dataset = 0 # datasets, 0 for Amazon, 1 for movielens
# eta = [0.05, 0.02][dataset] # learning rate
# lambda_r = [1, 0.5][dataset] # coefficient of regularization term
# lambda_phi = [5, 1][dataset] # coefficient of regularization term
# K0 = 50 # length of latent factors
# K1 = 50 # length of latent factors for probability
# vali_test = 0 # 0 for validation set, 1 for test set
# sample_rate = 4 # sample rate, the number of negative samples foreach positive one
# batch_size = 5000 # batch size
# epoch = 200 # number of epochs
# top_k = [2, 5, 10, 20, 50, 100] # top k items to recommend
# Model = 'NBPO'
#
# ## Paths to save and read
# # list for datasets
# dataset_list = ['amazon', 'movielens']
# path_train = 'dataset\\' + dataset_list[dataset] + '\\train_data.json'
# path_validation = 'dataset\\' + dataset_list[dataset] + '\\validation_data.json'
# path_test = 'dataset\\' + dataset_list[dataset] + '\\test_data.json'
#
# import numpy as np
# from numpy import *
# import xlwt
# import time
# from Library import readdata
# from Library import evaluation_F1
# from Library import evaluation_NDCG
# from Library import save_parameters
# from Library import save_df
#
# import xlrd, xlwt
# from xlutils.copy import copy as xl_copy
# from openpyxl import load_workbook
# from openpyxl import Workbook
# import pandas as pd
#
# def d(x):
# # sigmoid(x)
# if x > 10:
# return 1
# if x < -10:
# return 0
# if x >= -10 and x <= 10:
# return 1.0 / (1.0 + exp(-x))
#
# def test_Model(U, V):
# # test the precision
# k_num = len(top_k)
# # k_num-length list to record F1 and NDCG
# F1 = np.zeros(k_num)
# NDCG = np.zeros(k_num)
#
# # test all test samples
# user_num = 0
# for u in range(M):
# # the data in test set is [[i, i, i, i],...,[i, i]]
# test_item = Test[u]
# if len(test_item) > 0:
# user_num += 1
# # score all items
# score = np.dot(V, U[u])
# # order
# b = zip(score, range(N))
# b.sort(key=lambda x: x[0])
# order = [x[1] for x in b]
# order.reverse()
# # remove the training samples from the recommendations
# train_positive = train_data_aux[u]
# for item in train_positive:
# order.remove(item)
# # for each k, calculate top_k
# for i in range(len(top_k)):
# F1[i] += evaluation_F1(order, top_k[i], test_item)
# NDCG[i] += evaluation_NDCG(order, top_k[i], test_item)
# # calculate the average
# F1 = (F1 / user_num).tolist()
# NDCG = (NDCG / user_num).tolist()
#
# return F1, NDCG
#
# def train_Model(eta):
# # training the model
# # initialization
# U = np.array([np.array([(random.random() / math.sqrt(K0)) for j in range(K0)]) for i in range(M)])
# V = np.array([np.array([(random.random() / math.sqrt(K0)) for j in range(K0)]) for i in range(N)])
# P = 5 * np.array([np.array([(random.random() / math.sqrt(K1)) for j in range(K1)]) for i in range(M)])
# Q = 5 * np.array([np.array([(-random.random() / math.sqrt(K1)) for j in range(K1)]) for i in range(N)])
# e = 10 ** 10
# # output a result without training
# print 'iteration ', 0,
# [F1, NDCG] = test_Model(U, V)
# Fmax = 0
# if F1[0] > Fmax:
# Fmax = F1[0]
# print Fmax, 'F1: ', F1, ' ', 'NDCG: ', NDCG
# # save in .xls file
# F1_df = pd.DataFrame(columns=top_k)
# NDCG_df = pd.DataFrame(columns=top_k)
# F1_df.loc[0] = F1
# NDCG_df.loc[0] = NDCG
# save_df([[F1_df, 'F1'], [NDCG_df, 'NDCG']], path_excel, first_sheet=False) # @x
# # get the numer of training samples
# Re = len(train_data)
# # split the training samples with batch_size
# bs = range(0, Re, batch_size)
# bs.append(Re)
# # begin iterating
# for ep in range(epoch):
# print 'iteration ', ep + 1,
# eta = eta * 0.99
# # for each iterating, we user all training samples to train
# for ii in range(len(bs) - 1):
# if abs(U.sum()) < e:
# # input the samples as batches
# # initialize dU and dC to record the gradient
# dU = np.zeros((M, K0))
# dV = np.zeros((N, K0))
# dP = np.zeros((M, K1))
# dQ = np.zeros((N, K1))
# for re in range(bs[ii], bs[ii + 1]):
# # iterate each batch
# # train data, [u, i]
# u = train_data[re][0]
# i = train_data[re][1]
# Ri = d(np.dot(U[u], V[i]))
# gammai = d(np.dot(P[u], Q[i]))
# # select negative samples randomly
# num = 0
# while num < sample_rate:
# j = int(random.uniform(0, N))
# # check if the current sample is positive sample
# if not (j in train_data_aux[u]):
# num += 1
# Rj = d(np.dot(U[u], V[j]))
# gammaj = d(np.dot(P[u], Q[j]))
# D = (1 - Ri) * (1 - gammai)
# dU[u] += D * V[i]
# dV[i] += D * U[u]
# D = -Ri * gammai
# dP[u] += D * Q[i]
# dQ[i] += D * P[u]
#
# D = gammaj * (1 - Rj) - Rj
# dU[u] += D * V[j]
# dV[j] += D * U[u]
# D = (1 - gammaj) * Rj
# dP[u] += D * Q[j]
# dQ[j] += D * P[u]
#
# # update the matrices
# U += eta * (dU - lambda_r * U)
# V += eta * (dV - lambda_r * V)
# P += eta * (dP - lambda_phi * P)
# Q += eta * (dQ - lambda_phi * Q)
# if abs(U.sum()) < e:
# [F1, NDCG] = test_Model(U, V)
# if F1[0] > Fmax:
# Fmax = F1[0]
# F1_df.loc[ep + 1] = F1
# NDCG_df.loc[ep + 1] = NDCG
# print 'F1: ', F1, ' ', 'NDCG: ', NDCG
# save_df([[F1_df, 'F1'], [NDCG_df, 'NDCG']], path_excel, first_sheet=False) # @x
# else:
# break
# return F1_df, NDCG_df
#
# def print_parameter():
# # print all parameters
# print 'model:', Model
# print 'dataset:', dataset_list[dataset]
# print 'eta:', eta
# print 'lambda_r:', lambda_r, 'lambda_phi:', lambda_phi
# print 'K0:', K0, 'K1:', K1
# print 'vali_test:', ['validation', 'test'][vali_test]
# print 'sample_rate:', sample_rate
# print 'batch_size:', batch_size
# print 'epoch:', epoch
# print 'top_k:', top_k
# print
#
# '''**************************main_function***************************'''
# '''**************************main_function***************************'''
#
# # load the data
# [train_data, train_data_aux, M, N] = readdata(path_train)
# validation_data = readdata(path_validation)[1]
# test_data = readdata(path_test)[1]
# # choose validation or test set
# if vali_test == 0:
# Test = validation_data
# else:
# Test = test_data
#
# data = [
# ["Model", [Model]],
# ["dataset", [dataset_list[dataset]]],
# ["eta", [eta]],
# ["lambda_r", [lambda_r]],
# ["lambda_phi", [lambda_phi]],
# ["K0", [K0]],
# ["K1", [K1]],
# ["vali_test", [['validation', 'test'][vali_test]]],
# ["sample_rate", [sample_rate]],
# ["batch_size", [batch_size]],
# ['epoch', [epoch]],
# ["top_k", top_k]
# ]
# path_excel = 'experiment_result\\' + dataset_list[dataset] + '_' + Model + '_' + str(int(time.time())) + str(int(random.uniform(100, 900))) + '.xlsx'
# print_parameter()
# save_parameters(data, path_excel)
# F1_df, NDCG_df = train_Model(eta)