阿里天池大数据竞赛（一）用ODPS提取特征

//2015年4月30日

提取特征之所以难，是因为我们写出的sql语句往往只能提取一个特征。

而在ODPS上却不一样，一条语句可以提取很多特征。

如提取最近1、2、3、4天四种行为的数量

重点是decode函数，具体用法详见ODPS-sql文档

<span style="white-space:pre">		</span>int days[]={0,-1,-2,-3};
		String sql="create table if not exists  user_set1 as select user_id";
		featurenum=0;
		for(int i=1;i<5;i++){
			for(int j:days){
				String date2=processdate(date, j);
			sql+=",\t sum(decode(behavior_type="+i+" and mydate=\""+date2+"\",true,1,0)) as feature"+featurenum++ +"\t";
			}
			sql+=",\t sum(decode(behavior_type="+i+" and mydate<=\""+date+"\" and mydate>\""+processdate(date, -4)+"\",true,1,0)) as feature"+featurenum++ +"\t";
			sql+=",\t sum(decode(behavior_type="+i+" and mydate<=\""+date+"\" and mydate>\""+processdate(date, -7)+"\",true,1,0)) as feature"+featurenum++ +"\t";
			sql+=",\t sum(decode(behavior_type="+i+" and mydate<=\""+date+"\" and mydate>\""+processdate(date, -14)+"\",true,1,0)) as feature"+featurenum++ +"\t";
			sql+=",\t sum(decode(behavior_type="+i+" and mydate<=\""+date+"\",true,1,0)) as feature"+featurenum++ +"\t";
		
		}
		sql+="from "+table+" where mydate<= \""+date+"\" group by user_id;";