hive中简单的表格数据清洗操作（实例+图解+代码 一看就懂 一做就废）

创建项目

在清洗之前先介绍环境：hive 、beeline、Zeppelin 同时所需的文件在下面的链接中，我祝你心情好😘：
链接：https://pan.baidu.com/s/123qr-BuTa7nxLxd1LXvH8A 提取码：cz12

1、第一步导入json

按照下图操作：


如图在Zeppelin的note中已经添加了项目json

2、上传csv表格至hdfs

先将文件上传到linux中然后下指令上传至hdfs：

3、创建hdfs文件夹

# hdfs dfs -mkdir -p /tmp/shopping/data/customer
# hdfs dfs -mkdir -p /tmp/shopping/data/transaction
# hdfs dfs -mkdir -p /tmp/shopping/data/store
hdfs dfs -mkdir -p /tmp/shopping/data/review

一句一句执行。

4、上传表格

hdfs dfs -put /opt/shop/customer_details.csv /tmp/shopping/data/customer
hdfs dfs -put /opt/shop/transaction_details.csv /tmp/shopping/data/transaction
hdfs dfs -put /opt/shop/store_details.csv /tmp/shopping/data/store
hdfs dfs -put /opt/shop/store_review.csv /tmp/shopping/data/review

可一次性执行

5、清理库

%hive
drop database if exists shopping cascade
%hive
create database shopping

6、修改interpreter

此为Zeppelin的一个缺点 在创建库之后需要修改url 否则创建的所有表都会在default的库中。

在url地址后添加创建的库shopping，之后才会真正的进入shopping库中创建表格。

7、创建原始数据表并且上传csv文件

建表：

%hive
create external table if not exists ext_customer_details(
customer_id string,
first_name string,
last_name string,
email string,
gender string,
address string,
country string,
language string,
job string,
credit_type string,
credit_no string
)
row format delimited fields terminated by ','
location '/tmp/shopping/data/customer'

上传：

%hive 
load data inpath '/tmp/shopping/data/customer/customer_details.csv' overwrite into table ext_customer_details

简单查询

将另外的三个表并且分别上传数据

%hive
create external table if not exists ext_transaction_details(
transaction_id string,
customer_id string,
store_id string,
price string,
product string,
`date` string,
time string
)
row format delimited fields terminated by ','
stored as textfile tblproperties("skip,header.line.count"="1")

%hive
create external table if not exists ext_store_details(
store_id string,
store_name string,
employee_number int
)
row format delimited fields terminated by ','
stored as textfile tblproperties("skip.header.line.count"="1")

%hive
create external table if not exists ext_store_review(
 transaction_id string,
 store_id string,
 review_score string
)
row format delimited fields terminated by ','
stored as textfile tblproperties("skip,header.line.count"="1")

上传：

%hive
-- load data inpath '/tmp/shopping/data/transaction/transaction_details.csv' overwrite into table ext_transaction_details
-- load data inpath '/tmp/shopping/data/store/store_details.csv' overwrite into table ext_store_details
load data inpath '/tmp/shopping/data/review/store_review.csv' overwrite into table ext_store_review

分别查询表是否有数据

%hive
select * from ext_transaction_details limit 20
--select * from ext_transaction_details limit 20
--select * from ext_store_details limit 20
--select * from ext_store_review limit 20

数据问题分析.

在开始之前先简单分析数据的问题，

• 解决以下有问题的数据

1. 对transaction_details中的重复数据生成新ID
2. 过滤掉store_review中没有评分的数据
3. 可以把清洗好的数据放到另一个表或者用View表示
4. 找出PII (personal information identification) 或PCI (personal confidential information) 数据进行加密或hash
5. 重新组织transaction数据按照日期YYYY-MM做分区

表格处理

对于transaction_details表格需要进行如下处理：

1. 对transaction_details中的重复数据生成新ID
2. 过滤掉store_review中没有评分的数据
3. 通过Zepplin实现交易数据按月分区
   首先先创建一个新表存储清洗后的数据：

%hive
create table if not exists transaction_details(
  transaction_id string,
  customer_id string,
  store_id string,
  price string,
  product string,
  `date` string,
  time string
)
partitioned by (year int,month int) 
row format delimited fields terminated by ','
stored as textfile

问题1：8001-8100数据的有重复

解决方案： 先用窗口函数排序 然后将id+（排序号-1）*10000
代码实现：

with
t1 as (select *,row_number() over(partition by transaction_id) as sq from ext_transaction_details where transaction_id>8000)
select (sq-1)*10000+transaction_id as transaction_id,
customer_id,
store_id,
price,
product,
date,
time,
year(date) as year,
month(date) as month
from t1

但是在实现过程中发现数据的分割出现了问题 如图

解决方案：见我的博客csv处理。
重新建表清洗之后：

问题2：过滤掉store_review中没有评分的数据

解决方案：清理无用数据并把数据重新导入至新表
代码实现：
首先创建新表以接收数据清洗之后的数据：

建表
create table store_review

%hive
create table if not exists store_review(
 transaction_id string,
 store_id string,
 review_score string
)
row format delimited fields terminated by ','
stored as textfile

清理无用数据：
思路：

select * from ext_store_review s where  
exists (select * from ext_transaction_details t where cast(s.transaction_id as int)=cast(t.transaction_id as int) and cast(s.store_id as int)=cast(t.store_id as int)) and s.review_score !=""

如图就将所需的有评论并且id对应双表都有的数据筛选出来的

在写hql语句时候，并不止一个写法，但是第一个写法是优化的：

select * from (select * from ext_store_review s where exists(select * from ext_transaction_details t where s.transaction_id=t.transaction_id )) as a where a.review_score >0

接下来将数据导入：

insert overwrite table store_review select * from ext_store_review s where  
exists 
(select * from ext_transaction_details t where cast(s.transaction_id as int)=cast(t.transaction_id as int) 
and 
cast(s.store_id as int)=cast(t.store_id as int)) and s.review_score !=""

问题3：credit_no的加密

解决方案：md5单向加密，当然有需求的话可以采用双向的base64式双向加密解密法.
代码实现

%hive
create table if not exists customer_details as
select customer_id,first_name,last_name,email,gender,address,country,job,credit_type,md5(credit_no) credit_no from ext_customer_details

问题4：transaction数据按照日期YYYY-MM做分区

解决方案 开启分区 并且按照年月分区
代码实现

set hive.exec.dynamic.partition=true   //开启动态分区

insert overwrite table transaction_details partition(year,month)
select (sq-1)*10000+transaction_id as transaction_id,
customer_id,
store_id,
price,
product,
date,
time,
year(date) as year,
month(date) as month from (select *,row_number() over(partition by transaction_id) as sq from ext_transaction_details ) t1