使用Hive中自定义UDAF函数实现统计区域产品用户访问排名

  • loveheping

    2018-01-26 18:42:28
  • 数据分析
  • 原创
UDAF实现方法: 
1，用户的UDAF必须继承了org.apache.hadoop.hive.ql.exec.UDAF；
2，用户的UDAF必须包含至少一个实现了org.apache.hadoop.hive.ql.exec的静态类，诸如实现了 UDAFEvaluator 
3，一个计算函数必须实现的5个方法的具体含义如下：
init()：主要是负责初始化计算函数并且重设其内部状态，一般就是重设其内部字段。一般在静态类中定义一个内部字段来存放最终的结果。
iterate()：每一次对一个新值进行聚集计算时候都会调用该方法，计算函数会根据聚集计算结果更新内部状态。当输 入值合法或者正确计算了，则       就返回true。
terminatePartial()：Hive需要部分聚集结果的时候会调用该方法，必须要返回一个封装了聚集计算当前状态的对象。
merge()：Hive进行合并一个部分聚集和另一个部分聚集的时候会调用该方法。
terminate()：Hive最终聚集结果的时候就会调用该方法。计算函数需要把状态作为一个值返回给用户。

mapreduce阶段调用函数 
MAP 
init() 
iterate() 
terminatePartial()

Combiner 
merge() 
terminatePartial()

REDUCE 
init() 
merge() 
terminate()

一、自定义UDAF函数
点击(此处)折叠或打开
  1. package hive.org.ruozedata;

  3. import java.util.*;

  5. import org.apache.hadoop.hive.ql.exec.UDAF;
  6. import org.apache.hadoop.hive.ql.exec.UDAFEvaluator;
  7. import org.apache.log4j.Logger;

  9.     public class UserClickUDAF extends UDAF {

  11.     // 日志对象初始化
  12.     public static Logger logger = Logger.getLogger(UserClickUDAF.class);

  14.     // 静态类实现UDAFEvaluator
  15.     public static class Evaluator implements UDAFEvaluator {
  16.         // 设置成员变量，存储每个统计范围内的总记录数
  17.         private static Map<String, String> courseScoreMap;

  19.         private static Map<String, String> city_info;
  20.         private static Map<String, String> product_info;
  21.         private static Map<String, String> user_click;

  23.         //初始化函数,map和reduce均会执行该函数,起到初始化所需要的变量的作用
  24.         public Evaluator() {
  25.             init();
  26.         }

  28.         // 初始化函数间传递的中间变量
  29.         public void init() {
  30.             courseScoreMap = new HashMap<String, String>();
  31.             city_info = new HashMap<String, String>();
  32.             product_info = new HashMap<String, String>();
  33.        }

  35.         //map阶段，返回值为boolean类型，当为true则程序继续执行，当为false则程序退出
  36.         public boolean iterate(String pcid, String pcname, String pccount) {
  37.             if (pcid == null || pcname == null || pccount == null) {
  38.                 return true;
  39.             }

  41.             if (pccount.equals("-1")) {
  42.                 // 城市表
  43.                 city_info.put(pcid, pcname);
  44.             }
  45.             else if (pccount.equals("-2")) {
  46.                 // 产品表
  47.                 product_info.put(pcid, pcname);
  48.             }
  49.             else
  50.             {
  51.                 // 处理用户点击关联
  52.                 unionCity_Prod_UserClic1(pcid, pcname, pccount);
  53.            }
  54.             return true;
  55.         }

  57.         // 处理用户点击关联
  58.         private void unionCity_Prod_UserClic1(String pcid, String pcname, String pccount) {
  59.             if (product_info.containsKey(pcid)) {
  60.                 if (city_info.containsKey(pcname)) {
  61.                     String city_name = city_info.get(pcname);
  62.                     String prod_name = product_info.get(pcid);
  63.                     String cp_name = city_name + prod_name;

  65.                     // 如果之前已经Put过Key值为区域信息，则把记录相加处理
  66.                     if (courseScoreMap.containsKey(cp_name)) {
  67.                         int pcrn = 0;
  68.                         String strTemp = courseScoreMap.get(cp_name);
  69.                         String courseScoreMap_pn = strTemp.substring(strTemp.lastIndexOf("\t".toString())).trim();
  70.                         pcrn = Integer.parseInt(pccount) + Integer.parseInt(courseScoreMap_pn);

  72.                         courseScoreMap.put(cp_name, city_name + "\t" + prod_name + "\t" + Integer.toString(pcrn));
  73.                     }
  74.                     else {
  75.                         courseScoreMap.put(cp_name, city_name + "\t" + prod_name + "\t" + pccount);
  76.                     }
  77.                 }
  78.             }
  79.         }

  81.         /**
  82.          * 类似于combiner,在map范围内做部分聚合，将结果传给merge函数中的形参mapOutput
  83.          * 如果需要聚合，则对iterator返回的结果处理，否则直接返回iterator的结果即可
  84.          */
  85.         public Map<String, String> terminatePartial() {
  86.             return courseScoreMap;
  87.         }

  89.         // reduce 阶段，用于逐个迭代处理map当中每个不同key对应的 terminatePartial的结果
  90.         public boolean merge(Map<String, String> mapOutput) {
  91.             this.courseScoreMap.putAll(mapOutput);
  92.             return true;
  93.         }
  94.         // 处理merge计算完成后的结果，即对merge完成后的结果做最后的业务处理
  95.         public String terminate() {
  96.             return courseScoreMap.toString();
  97.         }
  98.     }
  99. }
二、使用添加Jar包与Hive中临时函数
点击(此处)折叠或打开
  1. DROP TEMPORARY FUNCTION user_click;

  3. add jar /data/hive_udf-1.0.jar;

  5. CREATE TEMPORARY FUNCTION user_click AS 'hive.org.ruozedata.UserClickUDAF';
三、调用自定义UDAF函数处理数据

点击(此处)折叠或打开

  1. insert overwrite directory '/works/tmp1' ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
  2. select regexp_replace(substring(rs, instr(rs, '=')+1), '}', '') from (
  3.   select explode(split(user_click(pcid, pcname, type),',')) as rs from (
  4.     select * from (
  5.       select '-2' as type, product_id as pcid, product_name as pcname from product_info
  6.       union all
  7.       select '-1' as type, city_id as pcid,area as pcname from city_info
  8.       union all
  9.       select count(1) as type,
  10.              product_id as pcid,
  11.              city_id as pcname
  12.         from user_click
  13.        where action_time='2016-05-05'
  14.       group by product_id,city_id
  15.     ) a
  16.   order by type) b
  17. ) c
四、创建Hive临时外部表

点击(此处)折叠或打开

  1. create external table tmp1(
  2. city_name string,
  3. product_name string,
  4. rn string
  5. )
  6. ROW FORMAT DELIMITED FIELDS TERMINATED BY '\t'
  7. location '/works/tmp1';
五、-统计最终区域前3产品排名

点击(此处)折叠或打开

  1. select * from (
  2. select city_name,
  3.        product_name,
  4.        floor(sum(rn)) visit_num,
  5.        row_number()over(partition by city_name order by sum(rn) desc) rn,
  6.        '2016-05-05' action_time
  7.   from tmp1
  8.  group by city_name,product_name
  9. ) a where rn <=3
六、统计结果


此方法可能不会很好，但最少可以起到一定的抛砖引玉的功效。希望大家不吝赐教！


若泽大数据交流群：671914634


请使用浏览器的分享功能分享到微信等