Skip to content

F
ffm-baseline
Commit b4d5aaaf authored by 张彦钊's avatar 张彦钊
Browse files
Options

change test file

parent 1963df08
Show whitespace changes
Inline Side-by-side
Showing with 33 additions and 119 deletions
tensnsorflow/multi.py
View file @ b4d5aaaf
......@@ -6,7 +6,6 @@ import pytispark.pytispark as pti
from pyspark.sql import SparkSession
import datetime
import pandas as pd
import numpy as np
def app_list_func(x,l):
......@@ -42,147 +41,76 @@ def feature_engineer():
start = (temp - datetime.timedelta(days=2)).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," \
"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 " \
sql = "select e.y,e.z,e.stat_date,e.ucity_id,feat.level2_ids " \
"from jerry_test.esmm_train_data e " \
"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 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"])
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"]
features = ["ucity_id","stat_date"]
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)
apps_number, app_list_map = multi_hot(df,"level2_ids",1)
unique_values = []
for i in features:
unique_values.extend(df.select(i).distinct().rdd.map(lambda x: x[0]).collect())
temp = list(range(2 + apps_number + level2_number + level3_number,
2 + apps_number + level2_number + level3_number + len(unique_values)))
temp = list(range(2 + apps_number,
2 + apps_number + len(unique_values)))
value_map = dict(zip(unique_values, temp))
rdd = df.select("app_list","level2_ids","level3_ids","stat_date","ucity_id", "ccity_name",
"device_type", "manufacturer", "channel", "top", "time", "hospital_id",
"treatment_method", "price_min","price_max", "treatment_time","maintain_time",
"recover_time","y","z").rdd
rdd = df.select("level2_ids","stat_date","ucity_id","y","z").rdd
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[3]!= validate_date)\
.map(lambda x: (app_list_func(x[0], app_list_map), app_list_func(x[1], leve2_map),
app_list_func(x[2], leve3_map),value_map[x[3]],value_map[x[4]],
value_map[x[5]],value_map[x[6]],value_map[x[7]],value_map[x[8]],
value_map[x[9]],value_map[x[10]],value_map[x[11]],value_map[x[12]],
value_map[x[13]],value_map[x[14]],value_map[x[15]],value_map[x[16]],
value_map[x[17]], x[18],x[19]))
test = rdd.filter(lambda x: x[3] == validate_date)\
.map(lambda x: (app_list_func(x[0], app_list_map), app_list_func(x[1], leve2_map),
app_list_func(x[2], leve3_map), value_map[x[3]], value_map[x[4]],
value_map[x[5]], value_map[x[6]], value_map[x[7]], value_map[x[8]],
value_map[x[9]], value_map[x[10]], value_map[x[11]], value_map[x[12]],
value_map[x[13]], value_map[x[14]], value_map[x[15]], value_map[x[16]],
value_map[x[17]], x[18], x[19]))
spark.createDataFrame(test).toDF("app_list","level2_ids","level3_ids","stat_date","ucity_id", "ccity_name", "device_type", "manufacturer",
"channel", "top", "time", "hospital_id","treatment_method", "price_min",
"price_max", "treatment_time","maintain_time", "recover_time","y","z")\
.repartition(1).write.format("tfrecords").option("recordType", "SequenceExample").save(path=path+"va/", mode="overwrite")
train = rdd.filter(lambda x: x[1]!= validate_date)\
.map(lambda x: (app_list_func(x[0], app_list_map),[value_map[x[2]],value_map[x[1]]], x[3],x[4]))
test = rdd.filter(lambda x: x[1]== validate_date)\
.map(lambda x: (app_list_func(x[0], app_list_map),[value_map[x[2]],value_map[x[1]]], x[3],x[4]))
spark.createDataFrame(test).toDF("level2_ids","ids","y","z")\
.repartition(1).write.format("tfrecords").save(path=path+"va/", mode="overwrite")
print("va write done")
spark.createDataFrame(train).toDF("app_list","level2_ids","level3_ids","stat_date","ucity_id", "ccity_name", "device_type", "manufacturer",
"channel", "top", "time", "hospital_id","treatment_method", "price_min",
"price_max", "treatment_time","maintain_time", "recover_time","y","z") \
.repartition(1).write.format("tfrecords").option("recordType", "SequenceExample").save(path=path+"tr/", mode="overwrite")
spark.createDataFrame(train).toDF("level2_ids","ids","y","z") \
.repartition(1).write.format("tfrecords").save(path=path+"tr/", mode="overwrite")
print("done")
rdd.unpersist()
return validate_date,value_map,app_list_map,leve2_map,leve3_map
return validate_date,value_map,app_list_map
def get_predict(date,value_map,app_list_map,level2_map,level3_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," \
"k.treatment_method,k.price_min,k.price_max,k.treatment_time,k.maintain_time,k.recover_time " \
"from esmm_pre_data e left join user_feature u on e.device_id = u.device_id " \
"left join cid_type_top c on e.device_id = c.device_id " \
"left join cid_time_cut cut on e.cid_id = cut.cid " \
"left join device_app_list dl on e.device_id = dl.device_id " \
"left join diary_feat feat on e.cid_id = feat.diary_id " \
"left join train_Knowledge_network_data k on feat.level2 = k.level2_id"
def get_predict(date,value_map,app_list_map):
sql = "select e.y,e.z,e.label,e.ucity_id,feat.level2_ids from esmm_pre_data e " \
"left join diary_feat feat on e.cid_id = feat.diary_id"
features = ["app_list", "level2_ids", "level3_ids","ucity_id", "ccity_name", "device_type", "manufacturer",
"channel", "top", "time", "hospital_id",
"treatment_method", "price_min", "price_max", "treatment_time", "maintain_time", "recover_time"]
features = ["ucity_id"]
df = spark.sql(sql)
df = df.na.fill(dict(zip(features, features)))
df = df.drop_duplicates(["ucity_id", "level2_ids", "ccity_name", "device_type", "manufacturer",
"device_id","cid_id","label",
"channel", "top", "time", "app_list", "hospital_id", "level3_ids"])
rdd = df.select("app_list", "level2_ids", "level3_ids","ucity_id","device_id","cid_id","label", "y", "z",
"ccity_name", "device_type","manufacturer", "channel", "time", "hospital_id",
"treatment_method", "price_min", "price_max", "treatment_time", "maintain_time",
"recover_time","top") \
.rdd.map(lambda x: (app_list_func(x[0], app_list_map), app_list_func(x[1], level2_map),
app_list_func(x[2], level3_map), x[3],x[4],x[5],x[6],x[7],x[8],
value_map.get(x[3], 300000),value_map.get(x[9], 299999),
value_map.get(x[10], 299998), value_map.get(x[11], 299997),
value_map.get(x[12], 299996), value_map.get(x[13], 299995),
value_map.get(x[14], 299994),value_map.get(x[15], 299993),
value_map.get(x[16], 299992),value_map.get(x[17], 299991),
value_map.get(x[18], 299990),value_map.get(x[19], 299989),
value_map.get(x[20], 299988),value_map.get(x[21], 299987),
value_map[date]))
rdd = df.select("level2_ids","ucity_id","device_id","cid_id","label", "y", "z") \
.rdd.map(lambda x: (app_list_func(x[0], app_list_map),x[1],x[2],x[3],x[4],x[5],x[6],
[value_map.get(x[1], 300000),value_map.get(date, 299999)]))
rdd.persist()
native_pre = spark.createDataFrame(rdd.filter(lambda x:x[6] == 0).map(lambda x:(x[3],x[4],x[5])))\
native_pre = spark.createDataFrame(rdd.filter(lambda x:x[4] == 0).map(lambda x:(x[1],x[2],x[3])))\
.toDF("city","uid","cid_id")
print("native")
native_pre.toPandas().to_csv(local_path+"native.csv", header=True)
spark.createDataFrame(rdd.filter(lambda x: x[6] == 0)
.map(lambda x: (x[0], x[1], x[2],x[7],x[8],x[9],x[10],x[11],x[12],
x[13],x[14],x[15],
x[16],x[17],x[18],x[19],x[20],x[21],x[22],x[23]))) \
.toDF("app_list", "level2_ids", "level3_ids","y","z","ucity_id",
"ccity_name", "device_type","manufacturer", "channel", "time", "hospital_id",
"treatment_method", "price_min", "price_max", "treatment_time", "maintain_time",
"recover_time", "top","stat_date").repartition(1).write.format("tfrecords").option("recordType", "SequenceExample") \
spark.createDataFrame(rdd.filter(lambda x: x[4] == 0).map(lambda x: (x[0],x[5],x[6],x[7]))) \
.toDF("level2_ids","y","z","ids").repartition(1).write.format("tfrecords") \
.save(path=path+"native/", mode="overwrite")
nearby_pre = spark.createDataFrame(rdd.filter(lambda x: x[6] == 1).map(lambda x: (x[3], x[4], x[5]))) \
nearby_pre = spark.createDataFrame(rdd.filter(lambda x: x[4] == 1).map(lambda x:(x[1],x[2],x[3]))) \
.toDF("city", "uid", "cid_id")
print("nearby")
nearby_pre.toPandas().to_csv(local_path+"nearby.csv", header=True)
spark.createDataFrame(rdd.filter(lambda x: x[6] == 1)
.map(lambda x: (x[0], x[1], x[2], x[7], x[8], x[9], x[10], x[11], x[12],
x[13], x[14], x[15],
x[16], x[17], x[18], x[19], x[20], x[21], x[22], x[23]))) \
.toDF("app_list", "level2_ids", "level3_ids","y","z", "ucity_id",
"ccity_name", "device_type", "manufacturer", "channel", "time", "hospital_id",
"treatment_method", "price_min", "price_max", "treatment_time", "maintain_time",
"recover_time","top","stat_date").repartition(1).write.format("tfrecords").option("recordType", "SequenceExample") \
spark.createDataFrame(rdd.filter(lambda x: x[4] == 1).map(lambda x: (x[0], x[5], x[6], x[7]))) \
.toDF("level2_ids","y","z","ids").repartition(1).write.format("tfrecords") \
.save(path=path+"nearby/", mode="overwrite")
rdd.unpersist()
......@@ -197,20 +125,6 @@ def con_sql(db,sql):
return df
def test():
sql = "select stat_date,cid_id,y,ccity_name from esmm_train_data limit 60"
rdd = spark.sql(sql).select("stat_date","cid_id","y","ccity_name").rdd.map(lambda x:(x[0],x[1],x[2],x[3]))
df = spark.createDataFrame(rdd)
df.show(6)
# spark.sql("use online")
# spark.sql("ADD JAR /srv/apps/brickhouse-0.7.1-SNAPSHOT.jar")
# spark.sql("ADD JAR /srv/apps/hive-udf-1.0-SNAPSHOT.jar")
# spark.sql("CREATE TEMPORARY FUNCTION json_map AS 'brickhouse.udf.json.JsonMapUDF'")
# spark.sql("CREATE TEMPORARY FUNCTION is_json AS 'com.gmei.hive.common.udf.UDFJsonFormatCheck'")
#
# spark.sql("select cl_type from online.tl_hdfs_maidian_view where partition_date = '20190312' limit 6").show()
if __name__ == '__main__':
sparkConf = SparkConf().set("spark.hive.mapred.supports.subdirectories", "true") \
.set("spark.hadoop.mapreduce.input.fileinputformat.input.dir.recursive", "true") \
......@@ -228,8 +142,8 @@ if __name__ == '__main__':
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)
validate_date, value_map, app_list_map = feature_engineer()
get_predict(validate_date, value_map, app_list_map)
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment