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
import tensorflow as tf
tf.logging.set_verbosity(tf.logging.INFO)
import sys
import time
import os
os.environ['TF_FORCE_GPU_ALLOW_GROWTH'] = 'true'
start = time.time()
BASE_DIR = '/data/files/wideAndDeep/'
def input_fn(csv_path, epoch, shuffle, batch_size):
dataset = tf.data.TextLineDataset(csv_path)
def parse_line(line_tensor):
splits = tf.compat.v1.string_split([line_tensor], delimiter='|', skip_empty=False).values
return {
'ITEM_CATEGORY_card_id': splits[0],
'USER_CATEGORY_device_id': splits[2],
'USER_CATEGORY_os': splits[3],
'USER_CATEGORY_user_city_id': splits[4],
'USER_MULTI_CATEGORY_second_solutions': tf.compat.v1.string_split([splits[6]], delimiter=',').values,
'USER_MULTI_CATEGORY_second_demands': tf.compat.v1.string_split([splits[7]], delimiter=',').values,
'USER_MULTI_CATEGORY_second_positions': tf.compat.v1.string_split([splits[8]], delimiter=',').values,
'USER_MULTI_CATEGORY_projects': tf.compat.v1.string_split([splits[9]], delimiter=',').values,
'ITEM_NUMERIC_click_count_sum': tf.compat.v1.string_to_number(splits[10]),
'ITEM_NUMERIC_click_count_avg': tf.compat.v1.string_to_number(splits[11]),
'ITEM_NUMERIC_click_count_stddev': tf.compat.v1.string_to_number(splits[12]),
'ITEM_NUMERIC_exp_count_sum': tf.compat.v1.string_to_number(splits[13]),
'ITEM_NUMERIC_exp_count_avg': tf.compat.v1.string_to_number(splits[14]),
'ITEM_NUMERIC_exp_count_stddev': tf.compat.v1.string_to_number(splits[15]),
'ITEM_NUMERIC_discount': tf.compat.v1.string_to_number(splits[16]),
'ITEM_NUMERIC_case_count': tf.compat.v1.string_to_number(splits[17]),
'ITEM_NUMERIC_sales_count': tf.compat.v1.string_to_number(splits[18]),
'ITEM_CATEGORY_service_type': splits[19],
'ITEM_CATEGORY_merchant_id': splits[20],
'ITEM_CATEGORY_doctor_type': splits[21],
'ITEM_CATEGORY_doctor_id': splits[22],
'ITEM_CATEGORY_doctor_famous': splits[23],
'ITEM_CATEGORY_hospital_id': splits[24],
'ITEM_CATEGORY_hospital_city_tag_id': splits[25],
'ITEM_CATEGORY_hospital_type': splits[26],
'ITEM_CATEGORY_hospital_is_high_quality': splits[27],
'ITEM_MULTI_CATEGORY_second_demands': tf.compat.v1.string_split([splits[28]], delimiter=',').values,
'ITEM_MULTI_CATEGORY_second_solutions': tf.compat.v1.string_split([splits[29]],
delimiter=',').values,
'ITEM_MULTI_CATEGORY_second_positions': tf.compat.v1.string_split([splits[30]],
delimiter=',').values,
'ITEM_MULTI_CATEGORY_projects': tf.compat.v1.string_split([splits[31]], delimiter=',').values,
'ITEM_NUMERIC_sku_price': tf.compat.v1.string_to_number(splits[32]),
# 'label': tf.compat.v1.string_to_number(splits[5])
}, tf.compat.v1.string_to_number(splits[5])
padded_shapes = ({'ITEM_CATEGORY_card_id': (), 'USER_CATEGORY_device_id': (), 'USER_CATEGORY_os': (),
'USER_CATEGORY_user_city_id': (), 'USER_MULTI_CATEGORY_second_solutions': [-1],
'USER_MULTI_CATEGORY_second_demands': [-1], 'USER_MULTI_CATEGORY_second_positions': [-1],
'USER_MULTI_CATEGORY_projects': [-1], 'ITEM_NUMERIC_click_count_sum': (),
'ITEM_NUMERIC_click_count_avg': (), 'ITEM_NUMERIC_click_count_stddev': (),
'ITEM_NUMERIC_exp_count_sum': (), 'ITEM_NUMERIC_exp_count_avg': (),
'ITEM_NUMERIC_exp_count_stddev': (), 'ITEM_NUMERIC_discount': (), 'ITEM_NUMERIC_case_count': (),
'ITEM_NUMERIC_sales_count': (), 'ITEM_CATEGORY_service_type': (), 'ITEM_CATEGORY_merchant_id': (),
'ITEM_CATEGORY_doctor_type': (), 'ITEM_CATEGORY_doctor_id': (), 'ITEM_CATEGORY_doctor_famous': (),
'ITEM_CATEGORY_hospital_id': (), 'ITEM_CATEGORY_hospital_city_tag_id': (),
'ITEM_CATEGORY_hospital_type': (), 'ITEM_CATEGORY_hospital_is_high_quality': (),
'ITEM_MULTI_CATEGORY_second_demands': [-1], 'ITEM_MULTI_CATEGORY_second_solutions': [-1],
'ITEM_MULTI_CATEGORY_second_positions': [-1], 'ITEM_MULTI_CATEGORY_projects': [-1],
'ITEM_NUMERIC_sku_price': ()}, ())
padding_values = ({'ITEM_CATEGORY_card_id': '-1', 'USER_CATEGORY_device_id': '-1', 'USER_CATEGORY_os': '-1',
'USER_CATEGORY_user_city_id': '-1', 'USER_MULTI_CATEGORY_second_solutions': '-1',
'USER_MULTI_CATEGORY_second_demands': '-1', 'USER_MULTI_CATEGORY_second_positions': '-1',
'USER_MULTI_CATEGORY_projects': '-1', 'ITEM_NUMERIC_click_count_sum': 0.0,
'ITEM_NUMERIC_click_count_avg': 0.0, 'ITEM_NUMERIC_click_count_stddev': 0.0,
'ITEM_NUMERIC_exp_count_sum': 0.0, 'ITEM_NUMERIC_exp_count_avg': 0.0,
'ITEM_NUMERIC_exp_count_stddev': 0.0, 'ITEM_NUMERIC_discount': 0.0,
'ITEM_NUMERIC_case_count': 0.0, 'ITEM_NUMERIC_sales_count': 0.0,
'ITEM_CATEGORY_service_type': '-1', 'ITEM_CATEGORY_merchant_id': '-1',
'ITEM_CATEGORY_doctor_type': '-1', 'ITEM_CATEGORY_doctor_id': '-1',
'ITEM_CATEGORY_doctor_famous': '-1', 'ITEM_CATEGORY_hospital_id': '-1',
'ITEM_CATEGORY_hospital_city_tag_id': '-1', 'ITEM_CATEGORY_hospital_type': '-1',
'ITEM_CATEGORY_hospital_is_high_quality': '-1', 'ITEM_MULTI_CATEGORY_second_demands': '-1',
'ITEM_MULTI_CATEGORY_second_solutions': '-1', 'ITEM_MULTI_CATEGORY_second_positions': '-1',
'ITEM_MULTI_CATEGORY_projects': '-1', 'ITEM_NUMERIC_sku_price': 0.0}, 0.0)
dataset = dataset.map(parse_line, num_parallel_calls = 8).cache()
dataset = dataset.padded_batch(batch_size, padded_shapes, padding_values=padding_values)
if shuffle:
dataset = dataset.shuffle(2048).prefetch(512 * 100).repeat(epoch)
else:
dataset = dataset.prefetch(512 * 100).repeat(epoch)
return dataset
boundaries = [0, 10, 100]
ITEM_NUMERIC_click_count_sum_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_click_count_sum'), boundaries)
ITEM_NUMERIC_exp_count_sum_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_exp_count_sum'), boundaries)
ITEM_NUMERIC_click_count_avg_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_click_count_avg'), boundaries)
ITEM_NUMERIC_exp_count_avg_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_exp_count_avg'), boundaries)
boundaries = [0, 0.01, 0.1]
ITEM_NUMERIC_click_count_stddev_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_click_count_stddev'), boundaries)
ITEM_NUMERIC_exp_count_stddev_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_exp_count_stddev'), boundaries)
boundaries = [0, 0.01, 0.1, 1]
ITEM_NUMERIC_discount_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_discount'), boundaries)
boundaries = [0, 10, 100]
ITEM_NUMERIC_case_count_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_case_count'), boundaries)
ITEM_NUMERIC_sales_count_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_sales_count'), boundaries)
ITEM_NUMERIC_sku_price_fc = tf.feature_column.bucketized_column(tf.feature_column.numeric_column('ITEM_NUMERIC_sku_price'), boundaries)
USER_CATEGORY_device_id_fc = tf.feature_column.categorical_column_with_vocabulary_file('USER_CATEGORY_device_id', BASE_DIR + 'USER_CATEGORY_device_id_vocab.csv')
USER_CATEGORY_os_fc = tf.feature_column.categorical_column_with_vocabulary_file('USER_CATEGORY_os', BASE_DIR + 'USER_CATEGORY_os_vocab.csv')
USER_CATEGORY_user_city_id_fc = tf.feature_column.categorical_column_with_vocabulary_file('USER_CATEGORY_user_city_id', BASE_DIR + 'USER_CATEGORY_user_city_id_vocab.csv')
USER_MULTI_CATEGORY__second_solutions_fc = tf.feature_column.categorical_column_with_vocabulary_file('USER_MULTI_CATEGORY_second_solutions', BASE_DIR + 'USER_MULTI_CATEGORY_second_solutions_vocab.csv')
USER_MULTI_CATEGORY__second_positions_fc = tf.feature_column.categorical_column_with_vocabulary_file('USER_MULTI_CATEGORY_second_positions', BASE_DIR + 'USER_MULTI_CATEGORY_second_positions_vocab.csv')
USER_MULTI_CATEGORY__second_demands_fc = tf.feature_column.categorical_column_with_vocabulary_file('USER_MULTI_CATEGORY_second_demands', BASE_DIR + 'USER_MULTI_CATEGORY_second_demands_vocab.csv')
USER_MULTI_CATEGORY__projects_fc = tf.feature_column.categorical_column_with_vocabulary_file('USER_MULTI_CATEGORY_projects', BASE_DIR + 'USER_MULTI_CATEGORY_projects_vocab.csv')
ITEM_CATEGORY_card_id_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_card_id', BASE_DIR + 'ITEM_CATEGORY_card_id_vocab.csv')
ITEM_CATEGORY_service_type_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_service_type', BASE_DIR + 'ITEM_CATEGORY_service_type_vocab.csv')
ITEM_CATEGORY_merchant_id_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_merchant_id', BASE_DIR + 'ITEM_CATEGORY_merchant_id_vocab.csv')
ITEM_CATEGORY_doctor_type_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_doctor_type', BASE_DIR + 'ITEM_CATEGORY_doctor_type_vocab.csv')
ITEM_CATEGORY_doctor_id_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_doctor_id', BASE_DIR + 'ITEM_CATEGORY_doctor_id_vocab.csv')
ITEM_CATEGORY_doctor_famous_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_doctor_famous', BASE_DIR + 'ITEM_CATEGORY_doctor_famous_vocab.csv')
ITEM_CATEGORY_hospital_id_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_hospital_id', BASE_DIR + 'ITEM_CATEGORY_hospital_id_vocab.csv')
ITEM_CATEGORY_hospital_city_tag_id_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_hospital_city_tag_id', BASE_DIR + 'ITEM_CATEGORY_hospital_city_tag_id_vocab.csv')
ITEM_CATEGORY_hospital_type_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_hospital_type', BASE_DIR + 'ITEM_CATEGORY_hospital_type_vocab.csv')
ITEM_CATEGORY_hospital_is_high_quality_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_CATEGORY_hospital_is_high_quality', BASE_DIR + 'ITEM_CATEGORY_hospital_is_high_quality_vocab.csv')
ITEM_MULTI_CATEGORY__second_solutions_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_MULTI_CATEGORY_second_solutions', BASE_DIR + 'ITEM_MULTI_CATEGORY_second_solutions_vocab.csv')
ITEM_MULTI_CATEGORY__second_positions_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_MULTI_CATEGORY_second_positions', BASE_DIR + 'ITEM_MULTI_CATEGORY_second_positions_vocab.csv')
ITEM_MULTI_CATEGORY__second_demands_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_MULTI_CATEGORY_second_demands', BASE_DIR + 'ITEM_MULTI_CATEGORY_second_demands_vocab.csv')
ITEM_MULTI_CATEGORY__projects_fc = tf.feature_column.categorical_column_with_vocabulary_file('ITEM_MULTI_CATEGORY_projects', BASE_DIR + 'ITEM_MULTI_CATEGORY_projects_vocab.csv')
def embedding_fc(categorical_column, dim):
return tf.feature_column.embedding_column(categorical_column, dim)
linear_feature_columns = [
ITEM_NUMERIC_click_count_sum_fc,
ITEM_NUMERIC_exp_count_sum_fc,
ITEM_NUMERIC_click_count_avg_fc,
ITEM_NUMERIC_exp_count_avg_fc,
ITEM_NUMERIC_click_count_stddev_fc,
ITEM_NUMERIC_exp_count_stddev_fc,
ITEM_NUMERIC_discount_fc,
ITEM_NUMERIC_case_count_fc,
ITEM_NUMERIC_sales_count_fc,
ITEM_NUMERIC_sku_price_fc,
embedding_fc(ITEM_CATEGORY_card_id_fc, 1),
embedding_fc(ITEM_CATEGORY_service_type_fc, 1),
embedding_fc(ITEM_CATEGORY_merchant_id_fc, 1),
embedding_fc(ITEM_CATEGORY_doctor_type_fc, 1),
embedding_fc(ITEM_CATEGORY_doctor_id_fc, 1),
embedding_fc(ITEM_CATEGORY_doctor_famous_fc, 1),
embedding_fc(ITEM_CATEGORY_hospital_id_fc, 1),
embedding_fc(ITEM_CATEGORY_hospital_city_tag_id_fc, 1),
embedding_fc(ITEM_CATEGORY_hospital_type_fc, 1),
embedding_fc(ITEM_CATEGORY_hospital_is_high_quality_fc, 1),
embedding_fc(ITEM_MULTI_CATEGORY__projects_fc, 1),
embedding_fc(ITEM_MULTI_CATEGORY__second_demands_fc, 1),
embedding_fc(ITEM_MULTI_CATEGORY__second_positions_fc, 1),
embedding_fc(ITEM_MULTI_CATEGORY__second_solutions_fc, 1),
]
dnn_feature_columns = [
embedding_fc(USER_CATEGORY_device_id_fc, 8),
embedding_fc(USER_CATEGORY_os_fc, 8),
embedding_fc(USER_CATEGORY_user_city_id_fc, 8),
embedding_fc(USER_MULTI_CATEGORY__second_solutions_fc, 8),
embedding_fc(USER_MULTI_CATEGORY__second_positions_fc, 8),
embedding_fc(USER_MULTI_CATEGORY__second_demands_fc, 8),
embedding_fc(USER_MULTI_CATEGORY__projects_fc, 8),
embedding_fc(ITEM_NUMERIC_click_count_sum_fc, 8),
embedding_fc(ITEM_NUMERIC_exp_count_sum_fc, 8),
embedding_fc(ITEM_NUMERIC_click_count_avg_fc, 8),
embedding_fc(ITEM_NUMERIC_exp_count_avg_fc, 8),
embedding_fc(ITEM_NUMERIC_click_count_stddev_fc, 8),
embedding_fc(ITEM_NUMERIC_exp_count_stddev_fc, 8),
embedding_fc(ITEM_NUMERIC_discount_fc, 8),
embedding_fc(ITEM_NUMERIC_case_count_fc, 8),
embedding_fc(ITEM_NUMERIC_sales_count_fc, 8),
embedding_fc(ITEM_NUMERIC_sku_price_fc, 8),
embedding_fc(ITEM_CATEGORY_card_id_fc, 8),
embedding_fc(ITEM_CATEGORY_service_type_fc, 8),
embedding_fc(ITEM_CATEGORY_merchant_id_fc, 8),
embedding_fc(ITEM_CATEGORY_doctor_type_fc, 8),
embedding_fc(ITEM_CATEGORY_doctor_id_fc, 8),
embedding_fc(ITEM_CATEGORY_doctor_famous_fc, 8),
embedding_fc(ITEM_CATEGORY_hospital_id_fc, 8),
embedding_fc(ITEM_CATEGORY_hospital_city_tag_id_fc, 8),
embedding_fc(ITEM_CATEGORY_hospital_type_fc, 8),
embedding_fc(ITEM_CATEGORY_hospital_is_high_quality_fc, 8),
embedding_fc(ITEM_MULTI_CATEGORY__projects_fc, 8),
embedding_fc(ITEM_MULTI_CATEGORY__second_demands_fc, 8),
embedding_fc(ITEM_MULTI_CATEGORY__second_positions_fc, 8),
embedding_fc(ITEM_MULTI_CATEGORY__second_solutions_fc, 8),
]
import os
os.environ["CUDA_VISIBLE_DEVICES"]="0, 1, 2"
from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())
distribution = tf.distribute.MirroredStrategy()
# session_config = tf.compat.v1.ConfigProto(log_device_placement = True, allow_soft_placement = True)
session_config = tf.compat.v1.ConfigProto(allow_soft_placement = True)
session_config.gpu_options.allow_growth = True
# config = tf.estimator.RunConfig(save_checkpoints_steps = 10000, train_distribute = distribution, eval_distribute = distribution)
config = tf.estimator.RunConfig(save_checkpoints_steps = 3000, session_config = session_config)
wideAndDeepModel = tf.estimator.DNNLinearCombinedClassifier(model_dir = BASE_DIR + 'model_csv',
linear_feature_columns = linear_feature_columns,
dnn_feature_columns = dnn_feature_columns,
dnn_hidden_units = [128, 32],
dnn_dropout = 0.5,
config = config)
# early_stopping = tf.contrib.estimator.stop_if_no_decrease_hook(wideAndDeepModel, eval_dir = wideAndDeepModel.eval_dir(), metric_name='auc', max_steps_without_decrease=1000, min_steps = 100)
# early_stopping = tf.contrib.estimator.stop_if_no_increase_hook(wideAndDeepModel, metric_name = 'auc', max_steps_without_increase = 1000, min_steps = 1000)
hooks = []
train_spec = tf.estimator.TrainSpec(input_fn = lambda: input_fn(BASE_DIR + 'train_samples.csv', 20, True, 1024), hooks = hooks)
serving_feature_spec = tf.feature_column.make_parse_example_spec(
linear_feature_columns + dnn_feature_columns)
serving_input_receiver_fn = (
tf.estimator.export.build_parsing_serving_input_receiver_fn(
serving_feature_spec))
exporter = tf.estimator.BestExporter(
name = "best_exporter",
compare_fn = lambda best_eval_result, current_eval_result: current_eval_result['auc'] > best_eval_result['auc'],
serving_input_receiver_fn = serving_input_receiver_fn,
exports_to_keep = 3)
eval_spec = tf.estimator.EvalSpec(input_fn = lambda: input_fn(BASE_DIR + 'eval_samples.csv', 1, False, 2 ** 15), steps = None, throttle_secs = 120, exporters = exporter)
# def my_auc(labels, predictions):
# return {'auc_pr_careful_interpolation': tf.metrics.auc(labels, predictions['logistic'], curve='ROC',
# summation_method='careful_interpolation')}
# wideAndDeepModel = tf.contrib.estimator.add_metrics(wideAndDeepModel, my_auc)
tf.estimator.train_and_evaluate(wideAndDeepModel, train_spec, eval_spec)
wideAndDeepModel.evaluate(lambda: input_fn(BASE_DIR + 'eval_samples.csv', 1, False, 2 ** 15))
print("训练耗时: {}s".format(time.time() - start))