-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_process_reconsume.py
490 lines (382 loc) · 22.9 KB
/
train_process_reconsume.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
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
# -*- coding: utf-8 -*-
# @Time : 2020/10/21 10:17
# @Author : zxl
# @FileName: train_process_reconsume.py
import time
import random
import traceback
import numpy as np
import tensorflow.compat.v1 as tf
from DataHandle.get_origin_data_amazon_music import Get_amazon_data_music
from DataHandle.get_origin_data_taobao import Get_taobaoapp_data
from Embedding.Behavior_embedding_time_aware_attention import Behavior_embedding_time_aware_attention
from Model.BPRMF import BPRMF
from Model.hybird_baseline_models import NARM, NARM_time_att, NARM_time_att_time_rnn, LSTUR, LSTUR_time_rnn, STAMP
from util.model_log import create_log
from DataHandle.get_input_data import DataInput
from Prepare.prepare_data_base import prepare_data_base
from DataHandle.get_origin_data_yoochoose import Get_yoochoose_data
from DataHandle.get_origin_data_movielen import Get_movie_data
from DataHandle.get_origin_data_tmall import Get_tmall_data
from DataHandle.get_origin_data_amazon_movie_tv import Get_amazon_data_movie_tv
from DataHandle.get_origin_data_amazon_elec import Get_amazon_data_elec
from DataHandle.get_origin_data_brightkite import Get_BrightKite_data
from DataHandle.get_origin_data_order import Get_Order_data
from config.model_parameter import model_parameter
from Model.PISTRec_model import Time_Aware_self_Attention_model
from Model.attention_baseline_models import Self_Attention_Model, Time_Aware_Self_Attention_Model, \
Ti_Self_Attention_Model
from Model.RNN_baesline_models import Gru4Rec, Vallina_Gru4Rec
from Model.RNN_baesline_models import Gru4Rec,T_SeqRec
from Model.MTAMRec_model import MTAM, MTAM_via_T_GRU, MTAM_no_time_aware_rnn, \
MTAM_no_time_aware_att, MTAM_hybird, MTAM_only_time_aware_RNN, MTAM_via_rnn, MTAM_with_T_SeqRec
from Model.MSRRP import MSRRP
import os
random.seed(1234)
np.random.seed(1234)
tf.set_random_seed(1234)
class Train_main_process:
def __init__(self):
start_time = time.time()
model_parameter_ins = model_parameter()
experiment_name = model_parameter_ins.flags.FLAGS.experiment_name
self.FLAGS = model_parameter_ins.get_parameter(experiment_name).FLAGS
log_ins = create_log(type = self.FLAGS.type, experiment_type = self.FLAGS.experiment_type,
version=self.FLAGS.version)
self.logger = log_ins.logger
self.logger.info("hello world the experiment begin")
# logger.info("The model parameter is :" + str(self.FLAGS._parse_flags()))
if self.FLAGS.type == "yoochoose":
get_origin_data_ins = Get_yoochoose_data(FLAGS=self.FLAGS)
get_origin_data_ins.getDataStatistics()
elif self.FLAGS.type == "movielen":
get_origin_data_ins = Get_movie_data(FLAGS=self.FLAGS)
get_origin_data_ins.getDataStatistics()
if self.FLAGS.type == "tmall":
get_origin_data_ins = Get_tmall_data(FLAGS=self.FLAGS)
elif self.FLAGS.type == "movie_tv":
get_origin_data_ins = Get_amazon_data_movie_tv(FLAGS=self.FLAGS)
get_origin_data_ins.getDataStatistics()
elif self.FLAGS.type == "elec":
get_origin_data_ins = Get_amazon_data_elec(FLAGS=self.FLAGS)
get_origin_data_ins.getDataStatistics()
elif self.FLAGS.type == "music":
get_origin_data_ins = Get_amazon_data_music(FLAGS=self.FLAGS)
get_origin_data_ins.getDataStatistics()
elif self.FLAGS.type == 'taobaoapp':
get_origin_data_ins = Get_taobaoapp_data(FLAGS=self.FLAGS)
get_origin_data_ins.getDataStatistics()
elif self.FLAGS.type == 'brightkite':
get_origin_data_ins = Get_BrightKite_data(FLAGS=self.FLAGS)
get_origin_data_ins.getDataStatistics()
elif self.FLAGS.type == "order":
get_origin_data_ins = Get_Order_data(FLAGS=self.FLAGS)
get_origin_data_ins.getDataStatistics()
#get_train_test_ins = Get_train_test(FLAGS=self.FLAGS,origin_data=get_origin_data_ins.origin_data)
prepare_data_behavior_ins = prepare_data_base(self.FLAGS,get_origin_data_ins.origin_data)
self.train_set, self.test_set = prepare_data_behavior_ins.get_train_test()
#fetch part of test_data
#if len(self.train_set) > 2000000:
#self.test_set = random.sample(self.train_set,2000000)
#self.test_set = self.test_set.sample(3500)
self.logger.info('DataHandle Process.\tCost time: %.2fs' % (time.time() - start_time))
start_time = time.time()
self.emb = Behavior_embedding_time_aware_attention(is_training = self.FLAGS.is_training,
user_count = prepare_data_behavior_ins.user_count,
item_count = prepare_data_behavior_ins.item_count,
category_count= prepare_data_behavior_ins.category_count,
max_length_seq = self.FLAGS.length_of_user_history
)
self.logger.info('Get Train Test Data Process.\tCost time: %.2fs' % (time.time() - start_time))
self.item_category_dic = prepare_data_behavior_ins.item_category_dic
self.global_step = 0
self.one_epoch_step = 0
self.now_epoch = 0
'''
def _eval_auc(self, test_set):
auc_input = []
auc_input = np.reshape(auc_input,(-1,2))
for _, uij in DataInputTest(test_set, FLAGS.test_batch_size):
#auc_sum += model.eval(sess, uij) * len(uij[0])
auc_input = np.concatenate((auc_input,self.model.eval_test(self.sess,uij)))
#test_auc = auc_sum / len(test_set)
test_auc = roc_auc_score(auc_input[:,1],auc_input[:,0])
self.model.eval_writer.add_summary(
summary=tf.Summary(
value=[tf.Summary.Value(tag='New Eval AUC', simple_value=test_auc)]),
global_step=self.model.global_step.eval())
return test_auc
'''
def train(self):
start_time = time.time()
# Config GPU options
if self.FLAGS.per_process_gpu_memory_fraction == 0.0:
gpu_options = tf.GPUOptions(allow_growth=True)
elif self.FLAGS.per_process_gpu_memory_fraction == 1.0:
gpu_options = tf.GPUOptions()
else:
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction=self.FLAGS.per_process_gpu_memory_fraction,allow_growth=True)
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ['CUDA_VISIBLE_DEVICES'] = self.FLAGS.cuda_visible_devices
self.sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
if not tf.test.gpu_device_name():
self.logger.warning("No GPU is found")
else: self.logger.info(tf.test.gpu_device_name())
global_step_lr = tf.Variable(0, trainable=False)
lr1 = tf.train.exponential_decay(
learning_rate=self.FLAGS.learning_rate, global_step=global_step_lr, decay_steps=1000, decay_rate=0.995, staircase=True)
lr2 = tf.train.exponential_decay(
learning_rate=0.001, global_step=global_step_lr, decay_steps=1000, decay_rate=0.995,
staircase=True)
with self.sess.as_default():
if self.FLAGS.experiment_type == "MSRRP":
self.model = MSRRP(self.FLAGS,self.emb, self.sess)
self.logger.info('Init finish.\tCost time: %.2fs' % (time.time() - start_time))
#AUC暂时不看
# test_auc = self.model.metrics(sess=self.sess,
# batch_data=self.test_set,
# global_step=self.global_step,
# name='test auc')
# Eval init AUC
# self.logger.info('Init AUC: %.4f' % test_auc)
test_start = time.time()
self.hr_1, self.ndcg_1, self.hr_5, self.ndcg_5, self.hr_10, self.ndcg_10, self.hr_30, self.ndcg_30, self.hr_50, self.ndcg_50 = \
0,0,0,0,0,0,0,0,0,0
self.precision = 0.0
self.recall = 0.0
self.best_result_hr = []
self.best_result_ndcg = []
def eval_classification():
max_step = 0.
precision = 0.
recall = 0.
for step_i, batch_data in DataInput(self.test_set, self.FLAGS.test_batch_size):
max_step = 1 + max_step
step_precision, step_recall = self.model.metrics_classification(sess=self.sess,
batch_data=batch_data,
global_step=self.global_step,
topk=self.FLAGS.top_k)
precision += step_precision[0]
recall += step_recall[0]
precision_val = precision/max_step
recall_val = recall/max_step
if precision_val > self.precision and recall_val > self.recall:
self.precision = precision_val
self.recall = recall_val
print('----test precision : %.5f, recall : %.5f-----' % (precision_val,recall_val))
print('----MAX precision: %.5f, MAX recall : %.5f-----' % (self.precision,self.recall))
def eval_topk():
sum_hr_1, sum_ndcg_1, sum_hr_5, sum_ndcg_5, sum_hr_10, sum_ndcg_10, sum_hr_30, sum_ndcg_30, sum_hr_50, sum_ndcg_50 = 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
result_list_hr_all = []
result_list_ndcg_all = []
max_step=0
for step_i, batch_data in DataInput(self.test_set, self.FLAGS.test_batch_size):
max_step =1+max_step
if self.FLAGS.experiment_type == "NARM" or \
self.FLAGS.experiment_type == "NARM+" or \
self.FLAGS.experiment_type == "NARM++": # or \
#self.FLAGS.experiment_type == "MTAM" :
hr_1, ndcg_1, hr_5, ndcg_5, hr_10, ndcg_10, hr_30, ndcg_30, hr_50, ndcg_50, \
result_list_hr, result_list_ndcg= \
self.model.metrics_topK_concat(sess=self.sess,
batch_data=batch_data,
global_step=self.global_step,
topk=self.FLAGS.top_k)
else:
hr_1, ndcg_1, hr_5, ndcg_5, hr_10, ndcg_10, hr_30, ndcg_30, hr_50, ndcg_50, \
result_list_hr, result_list_ndcg = \
self.model.metrics_topK(sess=self.sess,
batch_data=batch_data,
global_step=self.global_step,
topk=self.FLAGS.top_k)
sum_hr_1 = sum_hr_1+hr_1
sum_ndcg_1 = sum_ndcg_1 + ndcg_1
sum_hr_5 = sum_hr_5 + hr_5
sum_ndcg_5 = sum_ndcg_5 + ndcg_5
sum_hr_10 = sum_hr_10 + hr_10
sum_ndcg_10 = sum_ndcg_10 + ndcg_10
sum_hr_30 = sum_hr_30 + hr_30
sum_ndcg_30 = sum_ndcg_30 + ndcg_30
sum_hr_50 = sum_hr_50 + hr_50
sum_ndcg_50 = sum_ndcg_50 + ndcg_50
result_list_hr_all= result_list_hr_all+result_list_hr
result_list_ndcg_all= result_list_ndcg_all+result_list_ndcg
sum_hr_1 /= max_step
sum_ndcg_1/= max_step
sum_hr_5 /= max_step
sum_ndcg_5 /= max_step
sum_hr_10 /= max_step
sum_ndcg_10 /= max_step
sum_hr_30/= max_step
sum_ndcg_30/= max_step
sum_hr_50 /= max_step
sum_ndcg_50 /= max_step
if sum_hr_1> self.hr_1 and sum_ndcg_1 > self.ndcg_1:
self.hr_1, self.ndcg_1 = sum_hr_1, sum_ndcg_1
if sum_hr_5> self.hr_5 and sum_ndcg_5 > self.ndcg_5:
self.hr_5, self.ndcg_5 = sum_hr_5, sum_ndcg_5
if sum_hr_10> self.hr_10 and sum_ndcg_10 > self.ndcg_10:
self.hr_10, self.ndcg_10 = sum_hr_10 , sum_ndcg_10
self.best_result_hr = result_list_hr_all
self.best_result_ndcg = result_list_ndcg_all
if sum_hr_30> self.hr_30 and sum_ndcg_30 > self.ndcg_30:
self.hr_30, self.ndcg_30 = sum_hr_30, sum_ndcg_30
if sum_hr_50> self.hr_50 and sum_ndcg_50 > self.ndcg_50:
self.hr_50, self.ndcg_50 = sum_hr_50, sum_ndcg_50
def summery(k,hr,ndcg):
tag_recall = 'recall@'+str(k)
tag_ndcg = 'ndgc@'+str(k)
summary_recall_rate = tf.Summary(value=[tf.Summary.Value(tag=tag_recall, simple_value=hr)])
self.model.train_writer.add_summary(summary_recall_rate, global_step=self.global_step)
summary_avg_ndcg = tf.Summary(value=[tf.Summary.Value(tag=tag_ndcg, simple_value=ndcg)])
self.model.train_writer.add_summary(summary_avg_ndcg, global_step=self.global_step)
self.logger.info('Test recall rate @ %d : %.4f ndcg @ %d: %.4f' % (k , hr, k, ndcg))
summery(1,sum_hr_1,sum_ndcg_1)
summery(5, sum_hr_5, sum_ndcg_5)
summery(10, sum_hr_10, sum_ndcg_10)
summery(30, sum_hr_30, sum_ndcg_30)
summery(50, sum_hr_50, sum_ndcg_50)
eval_topk()
eval_classification()
self.logger.info('End test. \tTest Cost time: %.2fs' % (time.time() - test_start))
# Start training
self.logger.info('Training....\tmax_epochs:%d\tepoch_size:%d' % (self.FLAGS.max_epochs,self.FLAGS.train_batch_size))
start_time, avg_loss, self.best_auc,self.best_recall,self.best_ndcg = time.time(), 0.0,0.0,0.0,0.0
loss_origin = []
loss_reconsume = []
for epoch in range(self.FLAGS.max_epochs):
#if epoch > 2:
#lr = lr/1.5
random.shuffle(self.train_set)
self.logger.info('tain_set:%d'%len(self.train_set))
epoch_start_time = time.time()
learning_rate = self.FLAGS.learning_rate
for step_i, train_batch_data in DataInput(self.train_set, self.FLAGS.train_batch_size):
try:
#print(self.sess.run(global_step_lr))
if learning_rate > 0.001:
learning_rate = self.sess.run(lr1,feed_dict={global_step_lr: self.global_step})
else:
learning_rate = self.sess.run(lr2, feed_dict={global_step_lr: self.global_step})
#print(learning_rate)
add_summary = bool(self.global_step % self.FLAGS.display_freq == 0)
step_loss,step_loss_origin,step_loss_reconsume, merge = self.model.train(self.sess,train_batch_data,learning_rate,
add_summary,self.global_step,epoch)
self.sess.graph.finalize()
self.model.train_writer.add_summary(merge,self.global_step)
avg_loss = avg_loss + step_loss
loss_origin.extend(step_loss_origin)
loss_reconsume.extend(step_loss_reconsume)
self.global_step = self.global_step + 1
self.one_epoch_step = self.one_epoch_step + 1
#evaluate for eval steps
if self.global_step % self.FLAGS.eval_freq == 0:
print(learning_rate)
self.logger.info("Epoch step is " + str(self.one_epoch_step))
self.logger.info("Global step is " + str(self.global_step))
self.logger.info("Train_loss is " + str(avg_loss / self.FLAGS.eval_freq))
self.logger.info("Cross Entropy Loss is "+str(np.mean(np.array(loss_origin))))
self.logger.info("Reconsume Loss is "+ str(np.mean(np.array(loss_reconsume))))
# train_auc = self.model.metrics(sess=self.sess, batch_data=train_batch_data,
# global_step=self.global_step,name='train auc')
# self.logger.info('Batch Train AUC: %.4f' % train_auc)
# self.test_auc = self.model.metrics(sess=self.sess, batch_data=self.test_set,
# global_step=self.global_step,name='test auc')
# self.logger.info('Test AUC: %.4f' % self.test_auc)
eval_topk()
eval_classification()
avg_loss = 0
loss_origin = []
loss_time = []
loss_reconsume = []
self.save_model()
if self.FLAGS.draw_pic == True:
self.save_fig()
except Exception as e:
self.logger.info("Error！！！！！！！！！！！！")
self.logger.info(e)
traceback.print_exc()
self.logger.info('one epoch Cost time: %.2f' %(time.time() - epoch_start_time))
self.logger.info("Epoch step is " + str(self.one_epoch_step))
self.logger.info("Global step is " + str(self.global_step))
self.logger.info("Train_loss is " + str(step_loss))
self.logger.info("Cross Entropy Loss is " + str(np.mean(np.array(loss_origin))))
self.logger.info("Reconsume Loss is " + str(np.mean(np.array(loss_reconsume))))
eval_topk()
eval_classification()
with open('best_result_hr_'+self.FLAGS.version, 'w+') as f:
f.write(str(self.best_result_hr))
with open('best_result_ndcg'+self.FLAGS.version, 'w+') as f:
f.write(str(self.best_result_ndcg))
self.logger.info('Max recall rate @ 1: %.4f ndcg @ 1: %.4f' % (self.hr_1, self.ndcg_1))
self.logger.info('Max recall rate @ 5: %.4f ndcg @ 5: %.4f' % (self.hr_5, self.ndcg_5))
self.logger.info('Max recall rate @ 10: %.4f ndcg @ 10: %.4f' % (self.hr_10, self.ndcg_10))
self.logger.info('Max recall rate @ 30: %.4f ndcg @ 30: %.4f' % (self.hr_30, self.ndcg_30))
self.logger.info('Max recall rate @ 50: %.4f ndcg @ 50: %.4f' % (self.hr_50, self.ndcg_50))
self.one_epoch_step = 0
#if self.global_step > 1000:
#lr = lr / 2
#if lr < 0.0005:
#lr = lr * 0.99
#elif self.FLAGS.type == "tmall":
#lr = lr * 0.5
#else:
#lr = lr * 0.98
self.logger.info('Epoch %d DONE\tCost time: %.2f' %
(self.now_epoch, time.time() - start_time))
self.now_epoch = self.now_epoch + 1
self.one_epoch_step = 0
self.model.save(self.sess,self.global_step)
self.logger.info('best test_auc: ' + str(self.best_auc))
self.logger.info('best recall: ' + str(self.best_recall))
self.logger.info('Finished')
#judge to save model
#three result for evaluating model: auc ndcg recall
def save_model(self):
# avg_loss / self.FLAGS.eval_freq, test_auc,test_auc_new))
# result.append((self.model.global_epoch_step.eval(), model.global_step.eval(), avg_loss / FLAGS.eval_freq, _eval(sess, test_set, model), _eval_auc(sess, test_set, model)))avg_loss = 0.0
# only store good model
is_save_model = False
#for bsbe
'''
if self.FLAGS.experiment_type == "bsbe" or self.FLAGS.experiment_type == "bpr":
if (self.test_auc > 0.85 and self.test_auc - self.best_auc > 0.01):
self.best_auc = self.test_auc
is_save_model = True
#recall for istsbp
elif self.FLAGS.experiment_type == "istsbp" or self.FLAGS.experiment_type == "pistrec":
if self.recall_rate > 0.15 and self.recall_rate > self.best_recall:
self.best_recall = self.recall_rate
is_save_model =True
'''
if self.global_step % 50000 == 0:
is_save_model = True
if is_save_model == True:
self.model.save(self.sess, self.global_step)
def save_fig(self):
# save fig
if self.global_step % (self.FLAGS.eval_freq * 1) == 0:
input_dic = self.emb.make_feed_dic(batch_data=self.test_set[:100])
if self.FLAGS.experiment_type == "istsbp" or \
self.FLAGS.experiment_type == "bsbe":
input_dic[self.model.now_bacth_data_size] = len(self.test_set[:100])
behavior = self.sess.run(self.model.user_history_embedding_result_dense, input_dic)
usr_h_fig, short_term_intent_fig, attention_result_fig, item_emb_lookup_table_fig = \
self.sess.run([self.model.user_h, self.model.short_term_intent,
self.model.attention_result,
self.emb.item_emb_lookup_table], input_dic)
# draw picture
generate_pic_class_ins = generate_pic_class(init=False, user_h=usr_h_fig,
short_term_intent=short_term_intent_fig,
attention_result=attention_result_fig,
item_table=item_emb_lookup_table_fig,
item_category_dic=self.item_category_dic)
generate_pic_class_ins.draw_picure(type=self.FLAGS.type,
experiment_type=self.FLAGS.experiment_type,
version=self.FLAGS.version,
global_step=self.global_step)
self.logger.info("save fig finish!!!")
if __name__ == '__main__':
main_process = Train_main_process()
main_process.train()