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
# -*- coding: utf-8 -*-
import pymysql
from pyspark.conf import SparkConf
import pytispark.pytispark as pti
from pyspark.sql import SparkSession
import datetime
import pandas as pd
import time
def app_list_func(x,l):
b = str(x).split(",")
e = []
for i in b:
if i in l.keys():
e.append(l[i])
else:
e.append(0)
return e
def multi_hot(df,column,n):
a = time.time()
v = list(set(df.select(column).rdd.map(lambda x: x[0]).collect()))
b = time.time()
print(column)
print("cost time 分钟")
print((b-a)/60)
app_list_value = [str(i).split(",") for i in v]
app_list_unique = []
for i in app_list_value:
app_list_unique.extend(i)
app_list_unique = list(set(app_list_unique))
number = len(app_list_unique)
app_list_map = dict(zip(app_list_unique, list(range(n, number + n))))
return number,app_list_map
def get_map():
db = pymysql.connect(host='172.16.40.158', port=4000, user='root', passwd='3SYz54LS9#^9sBvC', db='jerry_test')
sql = "select max(stat_date) from esmm_train_data"
validate_date = con_sql(db, sql)[0].values.tolist()[0]
def feature_engineer():
db = pymysql.connect(host='172.16.40.158', port=4000, user='root', passwd='3SYz54LS9#^9sBvC', db='jerry_test')
sql = "select max(stat_date) from esmm_train_data"
validate_date = con_sql(db, sql)[0].values.tolist()[0]
print("validate_date:" + validate_date)
temp = datetime.datetime.strptime(validate_date, "%Y-%m-%d")
start = (temp - datetime.timedelta(days=300)).strftime("%Y-%m-%d")
print(start)
sql = "select e.y,e.z,e.stat_date,e.ucity_id,feat.level2_ids,e.ccity_name,u.device_type,u.manufacturer," \
"u.channel,c.top,cut.time,dl.app_list,feat.level3_ids,doctor.hospital_id," \
"wiki.tag as tag1,question.tag as tag2,search.tag as tag3,budan.tag as tag4," \
"ot.tag as tag5,sixin.tag as tag6,cart.tag as tag7," \
"k.treatment_method,k.price_min,k.price_max,k.treatment_time,k.maintain_time,k.recover_time " \
"from jerry_test.esmm_train_data e left join jerry_test.user_feature u on e.device_id = u.device_id " \
"left join jerry_test.cid_type_top c on e.device_id = c.device_id " \
"left join jerry_test.cid_time_cut cut on e.cid_id = cut.cid " \
"left join jerry_test.device_app_list dl on e.device_id = dl.device_id " \
"left join jerry_test.diary_feat feat on e.cid_id = feat.diary_id " \
"left join jerry_test.train_Knowledge_network_data k on feat.level2 = k.level2_id " \
"left join jerry_test.wiki_tag wiki on e.device_id = wiki.device_id " \
"left join jerry_test.question_tag question on e.device_id = question.device_id " \
"left join jerry_test.search_tag search on e.device_id = search.device_id " \
"left join jerry_test.budan_tag budan on e.device_id = budan.device_id " \
"left join jerry_test.order_tag ot on e.device_id = ot.device_id " \
"left join jerry_test.sixin_tag sixin on e.device_id = sixin.device_id " \
"left join jerry_test.cart_tag cart on e.device_id = cart.device_id " \
"left join eagle.src_zhengxing_api_service service on e.diary_service_id = service.id " \
"left join eagle.src_zhengxing_api_doctor doctor on service.doctor_id = doctor.id " \
"where e.stat_date >= '{}'".format(start)
df = spark.sql(sql)
df = df.drop_duplicates(["ucity_id", "level2_ids", "ccity_name", "device_type", "manufacturer",
"channel", "top", "time", "stat_date", "app_list", "hospital_id", "level3_ids",
"tag1","tag2","tag3","tag4","tag5","tag6","tag7"])
features = ["ucity_id", "ccity_name", "device_type", "manufacturer",
"channel", "top", "time", "stat_date", "hospital_id",
"treatment_method", "price_min", "price_max", "treatment_time", "maintain_time", "recover_time"]
df = df.na.fill(dict(zip(features, features)))
apps_number, app_list_map = multi_hot(df, "app_list", 1)
level2_number, leve2_map = multi_hot(df, "level2_ids", 1 + apps_number)
level3_number, leve3_map = multi_hot(df, "level3_ids", 1 + apps_number + level2_number)
unique_values = []
for i in features:
a = time.time()
unique_values.extend(list(set(df.select(i).rdd.map(lambda x: x[0]).collect())))
b = time.time()
print(i)
print((b-a)/60)
temp = list(range(2 + apps_number + level2_number + level3_number,
2 + apps_number + level2_number + level3_number + len(unique_values)))
value_map = dict(zip(unique_values, temp))
c = time.time()
rdd = df.select("stat_date","y", "z","app_list","level2_ids","level3_ids",
"tag1","tag2","tag3","tag4","tag5","tag6","tag7",
"ucity_id", "ccity_name","device_type", "manufacturer", "channel", "top", "time",
"hospital_id","treatment_method", "price_min", "price_max", "treatment_time",
"maintain_time","recover_time").rdd.repartition(200).map(lambda x: (x[0],float(x[1]),float(x[2]),app_list_func(x[3], app_list_map), app_list_func(x[4], leve2_map),
app_list_func(x[5], leve3_map), app_list_func(x[6], leve2_map),app_list_func(x[7], leve2_map),
app_list_func(x[8], leve2_map), app_list_func(x[9], leve2_map),app_list_func(x[10], leve2_map),
app_list_func(x[11], leve2_map),app_list_func(x[12], leve2_map),
[value_map[x[0]], value_map[x[13]],value_map[x[14]], value_map[x[15]], value_map[x[16]],
value_map[x[17]],value_map[x[18]], value_map[x[19]], value_map[x[20]],value_map[x[21]],
value_map[x[22]], value_map[x[23]], value_map[x[24]],value_map[x[25]],value_map[x[26]]]))
d = time.time()
print("rdd")
print((d-c)/60)
rdd.persist()
# TODO 上线后把下面train fliter 删除，因为最近一天的数据也要作为训练集
# train = rdd.filter(lambda x: x[0] != validate_date) \
# .map(lambda x: (float(x[1]),float(x[2]),app_list_func(x[3], app_list_map), app_list_func(x[4], leve2_map),
# app_list_func(x[5], leve3_map), app_list_func(x[6], leve2_map),app_list_func(x[7], leve2_map),
# app_list_func(x[8], leve2_map), app_list_func(x[9], leve2_map),app_list_func(x[10], leve2_map),
# app_list_func(x[11], leve2_map),app_list_func(x[12], leve2_map),
# [value_map[x[0]], value_map[x[13]],value_map[x[14]], value_map[x[15]], value_map[x[16]],
# value_map[x[17]],value_map[x[18]], value_map[x[19]], value_map[x[20]],value_map[x[21]],
# value_map[x[22]], value_map[x[23]], value_map[x[24]],value_map[x[25]],value_map[x[26]]]))
train = rdd.filter(lambda x: x[0] != validate_date).map(lambda x:(x[1],x[2],x[3],x[4],x[5],x[6],x[7],x[8],x[9],
x[10],x[11],x[12],x[13]))
f = time.time()
spark.createDataFrame(train).toDF("y","z","app_list","level2_list","level3_list",
"tag1_list","tag2_list","tag3_list","tag4_list",
"tag5_list","tag6_list","tag7_list","ids") \
.write.format("tfrecords").save(path=path + "tr/", mode="overwrite")
h = time.time()
print("train tfrecord done")
print((h-f)/60)
test = rdd.filter(lambda x: x[0] == validate_date).map(lambda x:(x[1],x[2],x[3],x[4],x[5],x[6],x[7],x[8],x[9],
x[10],x[11],x[12],x[13]))
spark.createDataFrame(test).toDF("y", "z", "app_list", "level2_list", "level3_list",
"tag1_list", "tag2_list", "tag3_list", "tag4_list",
"tag5_list", "tag6_list", "tag7_list", "ids") \
.write.format("tfrecords").save(path=path+"va/", mode="overwrite")
print("va tfrecord done")
rdd.unpersist()
return validate_date,value_map,app_list_map,leve2_map,leve3_map
def get_predict(date,value_map,app_list_map,leve2_map,leve3_map):
sql = "select e.y,e.z,e.label,e.ucity_id,feat.level2_ids,e.ccity_name," \
"u.device_type,u.manufacturer,u.channel,c.top,e.device_id,e.cid_id,cut.time," \
"dl.app_list,e.hospital_id,feat.level3_ids," \
"wiki.tag as tag1,question.tag as tag2,search.tag as tag3,budan.tag as tag4," \
"ot.tag as tag5,sixin.tag as tag6,cart.tag as tag7," \
"k.treatment_method,k.price_min,k.price_max,k.treatment_time,k.maintain_time,k.recover_time " \
"from jerry_test.esmm_train_data_dur e " \
"left join jerry_test.user_feature u on e.device_id = u.device_id " \
"left join jerry_test.cid_type_top c on e.device_id = c.device_id " \
"left join jerry_test.cid_time_cut cut on e.cid_id = cut.cid " \
"left join jerry_test.device_app_list dl on e.device_id = dl.device_id " \
"left join jerry_test.diary_feat feat on e.cid_id = feat.diary_id " \
"left join jerry_test.wiki_tag wiki on e.device_id = wiki.device_id " \
"left join jerry_test.question_tag question on e.device_id = question.device_id " \
"left join jerry_test.search_tag search on e.device_id = search.device_id " \
"left join jerry_test.budan_tag budan on e.device_id = budan.device_id " \
"left join jerry_test.order_tag ot on e.device_id = ot.device_id " \
"left join jerry_test.sixin_tag sixin on e.device_id = sixin.device_id " \
"left join jerry_test.cart_tag cart on e.device_id = cart.device_id " \
"left join jerry_test.train_Knowledge_network_data k on feat.level2 = k.level2_id"
features = ["ucity_id", "ccity_name", "device_type", "manufacturer",
"channel", "top", "time", "hospital_id",
"treatment_method", "price_min", "price_max", "treatment_time", "maintain_time", "recover_time"]
df = spark.sql(sql)
df = df.na.fill(dict(zip(features, features)))
c = time.time()
rdd = df.select("label", "y", "z","ucity_id","device_id","cid_id","app_list", "level2_ids", "level3_ids",
"tag1", "tag2", "tag3", "tag4", "tag5", "tag6", "tag7",
"ucity_id", "ccity_name", "device_type", "manufacturer", "channel", "top", "time",
"hospital_id", "treatment_method", "price_min", "price_max", "treatment_time",
"maintain_time", "recover_time") \
.rdd.repartition(200).map(lambda x: (x[0],float(x[1]),float(x[2]),x[3],x[4],x[5],
app_list_func(x[6], app_list_map),app_list_func(x[7], leve2_map),
app_list_func(x[8], leve3_map), app_list_func(x[9], leve2_map),
app_list_func(x[10], leve2_map),app_list_func(x[11], leve2_map),
app_list_func(x[12], leve2_map), app_list_func(x[13], leve2_map),
app_list_func(x[14], leve2_map), app_list_func(x[15], leve2_map),
[value_map.get(date, 299999),value_map.get(x[16], 299998),
value_map.get(x[17], 299997),value_map.get(x[18], 299996),
value_map.get(x[19], 299995), value_map.get(x[20], 299994),
value_map.get(x[21], 299993), value_map.get(x[22], 299992),
value_map.get(x[23], 299991), value_map.get(x[24], 299990),
value_map.get(x[25], 299989), value_map.get(x[26], 299988),
value_map.get(x[27], 299987), value_map.get(x[28], 299986),
value_map.get(x[29], 299985)
]))
rdd.persist()
d = time.time()
print("rdd")
print((d-c)/60)
native_pre = spark.createDataFrame(rdd.filter(lambda x:x[0] == 0).map(lambda x:(x[3],x[4],x[5])))\
.toDF("city","uid","cid_id")
print("native csv")
native_pre.toPandas().to_csv(local_path+"native.csv", header=True)
# TODO 写成csv文件改成下面这样
# native_pre.coalesce(1).write.format('com.databricks.spark.csv').save(path+"native/",header = 'true')
# 预测的tfrecord必须写成一个文件，这样可以摆保证顺序
f = time.time()
spark.createDataFrame(rdd.filter(lambda x: x[0] == 0)
.map(lambda x: (x[1],x[2],x[6],x[7],x[8],x[9],x[10],x[11],x[12],x[13],x[14],x[15],x[16]))) \
.toDF("y","z","app_list", "level2_list", "level3_list","tag1_list", "tag2_list", "tag3_list", "tag4_list",
"tag5_list", "tag6_list", "tag7_list", "ids").coalesce(1).write.format("tfrecords") \
.save(path=path+"native/", mode="overwrite")
print("native tfrecord done")
h = time.time()
print((h-f)/60)
native_pre = spark.createDataFrame(rdd.filter(lambda x: x[0] == 1).map(lambda x: (x[3], x[4], x[5]))) \
.toDF("city", "uid", "cid_id")
print("nearby csv")
native_pre.toPandas().to_csv(local_path + "nearby.csv", header=True)
# TODO 写成csv文件改成下面这样
# nearby_pre.coalesce(1).write.format('com.databricks.spark.csv').save(path+"nearby/",header = 'true')
spark.createDataFrame(rdd.filter(lambda x: x[0] == 1)
.map(
lambda x: (x[1], x[2], x[6], x[7], x[8], x[9], x[10], x[11], x[12], x[13], x[14], x[15], x[16]))) \
.toDF("y", "z", "app_list", "level2_list", "level3_list", "tag1_list", "tag2_list", "tag3_list", "tag4_list",
"tag5_list", "tag6_list", "tag7_list", "ids").coalesce(1).write.format("tfrecords") \
.save(path=path + "nearby/", mode="overwrite")
print("nearby tfrecord done")
rdd.unpersist()
def con_sql(db,sql):
cursor = db.cursor()
cursor.execute(sql)
result = cursor.fetchall()
df = pd.DataFrame(list(result))
db.close()
return df
if __name__ == '__main__':
sparkConf = SparkConf().set("spark.hive.mapred.supports.subdirectories", "true") \
.set("spark.hadoop.mapreduce.input.fileinputformat.input.dir.recursive", "true") \
.set("spark.tispark.plan.allow_index_double_read", "false") \
.set("spark.tispark.plan.allow_index_read", "true") \
.set("spark.sql.extensions", "org.apache.spark.sql.TiExtensions") \
.set("spark.tispark.pd.addresses", "172.16.40.158:2379").set("spark.io.compression.codec", "lzf")\
.set("spark.driver.maxResultSize", "8g").set("spark.sql.avro.compression.codec","snappy")
spark = SparkSession.builder.config(conf=sparkConf).enableHiveSupport().getOrCreate()
ti = pti.TiContext(spark)
ti.tidbMapDatabase("jerry_test")
ti.tidbMapDatabase("eagle")
spark.sparkContext.setLogLevel("WARN")
path = "hdfs:///strategy/esmm/"
local_path = "/home/gmuser/esmm/"
validate_date, value_map, app_list_map, leve2_map, leve3_map = feature_engineer()
get_predict(validate_date, value_map, app_list_map, leve2_map, leve3_map)