阿里天池大数据之移动推荐算法大赛总结及代码全公布

时间：2015-07-10 22:21:32 收藏：1 阅读：21578

移动推荐算法比赛已经结束了一个多星期了，现在写一篇文章来回顾一下自己的参赛历程。

首先，对不了解这个比赛的同学们介绍一下这个比赛（引用自官网）：

赛题简介

2014年是阿里巴巴集团移动电商业务快速发展的一年，例如2014双11大促中移动端成交占比达到42.6%，超过240亿元。相比PC时代，移动端网络的访问是随时随地的，具有更丰富的场景数据，比如用户的位置信息、用户访问的时间规律等。

本次大赛以阿里巴巴移动电商平台的真实用户-商品行为数据为基础，同时提供移动时代特有的位置信息，而参赛队伍则需要通过大数据和算法构面向建移动电子商务的商品推荐模型。希望参赛队伍能够挖掘数据背后丰富的内涵，为移动用户在合适的时间、合适的地点精准推荐合适的内容。

赛制安排

第一赛季，3月20日—4月25日

1. 可下载的少量淘宝数据，并在本地调试算法，提交结果；若队伍一天内多次提交结果，新结果版本将覆盖原版本；

2.4月1日起开放提交结果入口，4月2日出首次排行榜，每天更新排行榜，按照F1分从高到低排序；排行榜将选择选手在本阶段的历史最优成绩进行排名展示；

3.4月20日将进行一次数据切换，排行榜取20日之后成绩排名；

4. 第一赛季截止时，最好成绩排名前500名的队伍进入第二赛季；

第二赛季，4月30日-7月1日

1. 第二赛季分为2个阶段：

1）Part 1，4月30日-6月23日， Part 1截止时，最好成绩排名前200名的队伍进入Part 2；

2）Part 2，6月24日-7月1日， Part 2截止时，最好成绩排名前5名的队伍将受邀参加决赛答辩；

注：

Part 1答案数据包含在观测日期当天有购买行为的用户全集50%的用户购买数据；

Part 2答案数据包含在观测日期当天有购买行为的用户全集100%的用户购买数据，即Part 2阶段的用户量是Part 1阶段的2倍；

2.入围选手需登录天池平台，访问和使用海量淘宝数据，并利用Map&Reduce、SQL及各种平台集成的机器学习算法包调试模型，提交结果；

3.第二赛季提供每天1次的评测机会，提交截止时间为0点，每天更新排行榜，按照F1分从高到低排序；

数据说明

竞赛数据包含两个部分。第一部分是用户在商品全集上的移动端行为数据（D）,表名为tianchi_mobile_recommend_train_user，包含如下字段：

字段	字段说明	提取说明
user_id	用户标识	抽样&字段脱敏
item_id	商品标识	字段脱敏
behavior_type	用户对商品的行为类型	包括浏览、收藏、加购物车、购买，对应取值分别是1、2、3、4。
user_geohash	用户位置的空间标识，可以为空	由经纬度通过保密的算法生成
item_category	商品分类标识	字段脱敏
time	行为时间	精确到小时级别

第二个部分是商品子集（P）,表名为tianchi_mobile_recommend_train_item，包含如下字段：

字段	字段说明	提取说明
item_id	商品标识	抽样&字段脱敏
item_ geohash	商品位置的空间标识，可以为空	由经纬度通过保密的算法生成
item_category	商品分类标识	字段脱敏

训练数据包含了抽样出来的一定量用户在一个月时间（11.18~12.18）之内的移动端行为数据（D），评分数据是这些用户在这个一个月之后的一天（12.19）对商品子集（P）的购买数据。参赛者要使用训练数据建立推荐模型，并输出用户在接下来一天对商品子集购买行为的预测结果。

评分数据格式

具体计算公式如下：参赛者完成用户对商品子集的购买预测之后，需要将结果放入指定格式的数据表（非分区表）中，要求结果表名为：tianchi_mobile_recommendation_predict，包含user_id和item_id两列（均为string类型）,要求去除重复。例如：

技术分享

初赛数据

初赛阶段提供10000用户的完整行为数据以及百万级的商品信息；训练和预测的数据将会在4月20日进行一次切换（即切换为另一批10000用户的数据）。

决赛数据

决赛阶段提供500万用户的完整行为数据以及千万级的商品信息。

评估指标

比赛采用经典的精确度(precision)、召回率(recall)和F1值作为评估指标。具体计算公式如下：

技术分享

其中PredictionSet为算法预测的购买数据集合，ReferenceSet为真实的答案购买数据集合。我们以F1值作为最终的唯一评测标准。

知道了这个比赛的背景及数据之后我们开始介绍第一赛季：

第一赛季用户数为1万，一个月的用户商品交互总共千万级别，商品子集十万级别，我们第一次根据用户的心里行为，觉得前一天的购物车商品很有可能第二天就被购买，所以我们直接提交了12月18号一天的购物车（跟商品子集交），得到的成绩为

技术分享

这个成绩在当时确实不错，特别是好多人还没想到这个侥幸的方法的时候，我们觉得还是挺机智的。这种直接提交购物车的方式不涉及任何算法，我们以后就称为规则了。之后的几天我们开始很兴奋的去寻找各种规则，包括剔除掉在30天里从来不买东西的人

技术分享

包括把前一天加入购物车然后当天购买的剔除掉

技术分享

还有一条云泛天音小哥给我们的指导，就是假设用户在12月18号上午9点加入购物车三件东西，10点买了其中一件，那么剩下两件就可以删掉了，因为如果要买，为何不一起买呢，因为用户做出了选择，F1继续提升

技术分享

这时候F1已经破9了，但是由于我们一直是在一天的数据集上进行操作，所以recall是有瓶颈的。但此时对于我们来说还是想继续找规则，相信那个阶段很多队伍都跟我们一样，原因只有一个，那就是规则提升太快了。

后来实在找不出啥强有力的规则来了,就准备用一下算法,因为LR快速简单,在python里面有包,我们就直接拿来用了,对稀疏问题采用了L1正则化.LR自己写的话也很快，就一个梯度下降定参。同一队的小哥之前用过那个包，我们就直接拿来用了。拆分数据集我们是这样做的

19号：线上测试

18号：线下测试用

17号：线下训练用（用当天购买或没购买的作用对打标签，然后用16号及之前的提特征）

16号之前的：提特征用

其中的具体数据是多少由于时间久远记不太清楚了，但其中最大的问题就是训练的时候正负样本不均衡的问题，我们是在写代码的时候把他当做一个参数设置的，然后选择最优的，第一赛季负样本我们做了1/20的抽样，正负样本比例我记得是在1:5左右。

这里分析一下正负样本比例的问题：

直接量化（1）正：负=1:3

（2）正：负=1:5

（3）正：负=1:7

从（1）到（3）负样本逐渐增多，负样本分的更细，算法预测出的正样本会变少，，准确率会上升，召回率会下降，因为评测指标是两者的均衡F1，所以总能调出一个F1最高的值来。

用LR的时候特征我们主要用了以下这些（直接上代码，都是些简单的特征，包括购物车购买转化率啊，商品热度啊等等）：

selectuser_id,count(behavior_type) from user_item where behavior_type=1 andtime<‘2014-12-18 00:00:00‘ group by user_id into outfile‘D://model//signal_feature//buser_click.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=2 andtime<‘2014-12-18 00:00:00‘ group by user_id into outfile‘D://model//signal_feature//buser_collect.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=3 andtime<‘2014-12-18 00:00:00‘ group by user_id into outfile‘D://model//signal_feature//buser_cart.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=4 andtime<‘2014-12-18 00:00:00‘ group by user_id into outfile‘D://model//signal_feature//buser_buy.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=1 andtime<‘2014-12-18 00:00:00‘ group by item_id into outfile‘D://model//signal_feature//bitem_click.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=2 andtime<‘2014-12-18 00:00:00‘ group by item_id into outfile‘D://model//signal_feature//bitem_collect.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=3 andtime<‘2014-12-18 00:00:00‘ group by item_id into outfile‘D://model//signal_feature//bitem_cart.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=4 andtime<‘2014-12-18 00:00:00‘ group by item_id into outfile‘D://model//signal_feature//bitem_buy.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=1 andtime<‘2014-12-18 00:00:00‘ group by user_id,item_id into outfile‘D://model//signal_feature//bu_it_click.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=2 andtime<‘2014-12-18 00:00:00‘ group by user_id,item_id into outfile‘D://model//signal_feature//bu_it_collect.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=3 andtime<‘2014-12-18 00:00:00‘ group by user_id,item_id into outfile‘D://model//signal_feature//bu_it_cart.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=4 andtime<‘2014-12-18 00:00:00‘ group by user_id,item_id into outfile‘D://model//signal_feature//bu_it_buy.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=1 group byuser_id into outfile ‘D://model//signal_feature//nuser_click.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=2 group byuser_id into outfile ‘D://model//signal_feature//nuser_collect.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=3 group byuser_id into outfile ‘D://model//signal_feature//nuser_cart.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=4 group byuser_id into outfile ‘D://model//signal_feature//nuser_buy.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=1 group byitem_id into outfile ‘D://model//signal_feature//nitem_click.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=2 group byitem_id into outfile ‘D://model//signal_feature//nitem_collect.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=3 group byitem_id into outfile ‘D://model//signal_feature//nitem_cart.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=4 group byitem_id into outfile ‘D://model//signal_feature//nitem_buy.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=1 groupby user_id,item_id into outfile ‘D://model//signal_feature//nu_it_click.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=2 groupby user_id,item_id into outfile ‘D://model//signal_feature//nu_it_collect.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=3 groupby user_id,item_id into outfile ‘D://model//signal_feature//nu_it_cart.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=4 groupby user_id,item_id into outfile ‘D://model//signal_feature//nu_it_buy.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=1 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byuser_id into outfile ‘D://model//signal_feature//buser_recent3day_click.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=2 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byuser_id into outfile ‘D://model//signal_feature//buser_recent3day_collect.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=3 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byuser_id into outfile ‘D://model//signal_feature//buser_recent3day_cart.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=4 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byuser_id into outfile ‘D://model//signal_feature//buser_recent3day_buy.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=1 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byitem_id into outfile ‘D://model//signal_feature//bitem_recent3day_click.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=2 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byitem_id into outfile ‘D://model//signal_feature//bitem_recent3day_collect.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=3 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byitem_id into outfile ‘D://model//signal_feature//bitem_recent3day_cart.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=4 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byitem_id into outfile ‘D://model//signal_feature//bitem_recent3day_buy.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=1 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byuser_id,item_id into outfile‘D://model//signal_feature//bu_it_recent3day_click.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=2 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byuser_id,item_id into outfile‘D://model//signal_feature//bu_it_recent3day_collect.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=3 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byuser_id,item_id into outfile‘D://model//signal_feature//bu_it_recent3day_cart.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=4 andtime<‘2014-12-18 00:00:00‘ and time>‘2014-12-15 00:00:00‘ group byuser_id,item_id into outfile‘D://model//signal_feature//bu_it_recent3day_buy.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=1 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byuser_id into outfile ‘D://model//signal_feature//nuser_recent3day_click.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=2 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byuser_id into outfile ‘D://model//signal_feature//nuser_recent3day_collect.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=3 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byuser_id into outfile ‘D://model//signal_feature//nuser_recent3day_cart.txt‘;

selectuser_id,count(behavior_type) from user_item where behavior_type=4 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byuser_id into outfile ‘D://model//signal_feature//nuser_recent3day_buy.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=1 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byitem_id into outfile ‘D://model//signal_feature//nitem_recent3day_click.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=2 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byitem_id into outfile ‘D://model//signal_feature//nitem_recent3day_collect.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=3 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byitem_id into outfile ‘D://model//signal_feature//nitem_recent3day_cart.txt‘;

selectitem_id,count(behavior_type) from user_item where behavior_type=4 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byitem_id into outfile ‘D://model//signal_feature//nitem_recent3day_buy.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=1 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byuser_id,item_id into outfile‘D://model//signal_feature//nu_it_recent3day_click.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=2 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byuser_id,item_id into outfile‘D://model//signal_feature//nu_it_recent3day_collect.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=3 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byuser_id,item_id into outfile‘D://model//signal_feature//nu_it_recent3day_cart.txt‘;

selectuser_id,item_id,count(behavior_type) from user_item where behavior_type=4 andtime<‘2014-12-19 00:00:00‘ and time>‘2014-12-16 00:00:00‘ group byuser_id,item_id into outfile‘D://model//signal_feature//nu_it_recent3day_buy.txt‘;

这时候我们是在规则的基础上用了两天的数据

技术分享

最后以10.02的分数结束了第一赛季。

第二赛季是相当蛋疼的，刚进入ODPS准备好好探索一番的时候实验室有了一些任务，不得不将大部分注意力转到科研上来，两个队友也有别的事没有再参与，中间甚至有一星期都没看过ODPS，全部做的时间加起来感觉也就10天左右，这个想起来还是非常遗憾的，因为难得的一次机会被浪费了。所以如果想好好做这个比赛的话一定要有充足的时间。因为它确实非常耗时间，只要坚持，就会有好结果的。

好的，进入正题，我来说下第二赛季我是怎么做的：

第二赛季用的是阿里的ODPS平台，提供了SQL和MR来提特征，我主要用的就是SQL。第二赛季第一阶段是500个队，第二阶段是200个队。这一阶段从我用ODPS开始就感到平台很慢，一个select查询语句两个小时都跑不出来，一直在等待资源。不少队伍是资源大户，并且机制是资源是一定的，谁的job占了谁就跑，并不是每一个队伍分固定的资源。想想其实阿里毕竟要先满足自己的业务需求，500个队，一队两台机子就要1000台了。。。。这实在代价很大。

上手ODPS平台可以看看官方文档，SQL语句比较偏向hive，用起来都是些简单的，也不难。

介绍一下第二赛季的数据,第二赛季用户数为500万,商品为千万级别的,30天的交互的条数大约是58亿条,数据量庞大,最后第二赛季第一阶段评测只用结果的一半,第二阶段用全部的结果.区别就是第二阶段precision会翻倍,recall基本不变.

好的，同学们可能对代码比较感兴趣，我就先贴代码(中间写了一些注释)，然后解释：

--------------------------------------------------------------------------------

-- 参数说明 --

--------------------------------------------------------------------------------

--集成的数据处理流程，包含了特征提取，数据融合，加分类标签，归一化和欠采样等

--输入参数：

--label_day：分类标签的日期，如：‘2014-12-18 00‘

--table_label:保存的表名label，例如16

--输出表：

--特征表系列：${table_label}_item_features_1,...,${table_label}_item_features_n,${table_label}_item_features,

-- ${table_label}_ui_features_1,....,${table_label}_ui_features_n,${table_label}_ui_features,

-- ${table_label}_user_features_1,....,${table_label}_user_features_n,${table_label}_user_features,

--数据融合处理：

-- ${table_label}_feature_table,${table_label}_normal,${table_label}_under_sample

--------------------------------------------------------------------------------

-- 特征提取 --

--------------------------------------------------------------------------------

--使用label_day之前的数据生成item特征

--已成功

drop table if exists${table_label}_item_features_1;

create table${table_label}_item_features_1 as

select

item_id,

--1）对不同item点击、收藏、购物车、购买的总计

sum(case when(behavior_type=1 andtime<‘${label_day}‘) then 1 else 0 end) as i1,

sum(case when(behavior_type=2 andtime<‘${label_day}‘) then 1 else 0 end) as i2,

sum(case when(behavior_type=3 andtime<‘${label_day}‘) then 1 else 0 end) as i3,

sum(case when(behavior_type=4 andtime<‘${label_day}‘) then 1 else 0 end) as i4,

--2)对不同item点击、收藏、购物车、购买平均每个user的计数

sum(case when(behavior_type=1 andtime<‘${label_day}‘) then 1 else 0 end)/count(distinct user_id) as i5,

--2015-5-6添加（品牌是否有变热门的征兆）

--最近第1天的行为数与日平均行为数的比值

case when sum(case when behavior_type=1then 1 else 0 end)=0 then 0 else sum(case when behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=1 then 1 else 0 end) end as i6,

case when sum(case when behavior_type=2then 1 else 0 end)=0 then 0 else sum(case when behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=2 then 1 else 0 end) end as i7,

case when sum(case when behavior_type=3then 1 else 0 end)=0 then 0 else sum(case when behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=3 then 1 else 0 end) end as i8,

--最近第2天的行为数与日平均行为数的比值

case when sum(case when behavior_type=1then 1 else 0 end)=0 then 0 else sum(case when behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=2 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=1 then 1 else 0 end) end as i9,

case when sum(case when behavior_type=2then 1 else 0 end)=0 then 0 else sum(case when behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=2 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=2 then 1 else 0 end) end as i10,

case when sum(case when behavior_type=3then 1 else 0 end)=0 then 0 else sum(case when behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=2 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=3 then 1 else 0 end) end as i11,

case when sum(case when behavior_type=4then 1 else 0 end)=0 then 0 else sum(case when behavior_type=4 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=2 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=4 then 1 else 0 end) end as i12,

--最近第3天的行为数与日平均行为数的比值

case when sum(case when behavior_type=1then 1 else 0 end)=0 then 0 else sum(case when behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=3 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=1 then 1 else 0 end) end as i13,

case when sum(case when behavior_type=2then 1 else 0 end)=0 then 0 else sum(case when behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=3 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=2 then 1 else 0 end) end as i14,

case when sum(case when behavior_type=3then 1 else 0 end)=0 then 0 else sum(case when behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=3 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=3 then 1 else 0 end) end as i15,

case when sum(case when behavior_type=4then 1 else 0 end)=0 then 0 else sum(case when behavior_type=4 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=3 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=4 then 1 else 0 end) end as i16,

--最近3天的行为数与日平均行为数的比值

case when sum(case when behavior_type=1then 1 else 0 end)=0 then 0 else sum(case when behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=1 then 1 else 0 end) end as i17,

case when sum(case when behavior_type=2then 1 else 0 end)=0 then 0 else sum(case when behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=2 then 1 else 0 end) end as i18,

case when sum(case when behavior_type=3then 1 else 0 end)=0 then 0 else sum(case when behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=3 then 1 else 0 end) end as i19,

case when sum(case when behavior_type=4then 1 else 0 end)=0 then 0 else sum(case when behavior_type=4 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4 then 1 else 0 end)*datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(‘2014-11-18 00‘,‘yyyy-mm-dd hh‘),‘dd‘)/sum(case whenbehavior_type=4 then 1 else 0 end) end as i20

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

group by item_id;

--已成功

--统计商品在类别中的排序(2015-5-15添加)

drop table if exists${table_label}_item_features_2;

create table${table_label}_item_features_2 as

select item_id,

dense_rank() over(partition byitem_category order by ii1 desc) as i21,

dense_rank() over(partition byitem_category order by ii2 desc) as i22,

dense_rank() over(partition byitem_category order by ii3 desc) as i23,

dense_rank() over(partition byitem_category order by ii4 desc) as i24

from

(

select item_category,item_id,

sum(case when behavior_type=1 then 1 else 0end) as ii1,

sum(case when behavior_type=2 then 1 else 0end) as ii2,

sum(case when behavior_type=3 then 1 else 0end) as ii3,

sum(case when behavior_type=4 then 1 else 0end) as ii4

fromtianchi_lbs.tianchi_mobile_recommend_train_user

where time<‘${label_day}‘

group by item_id,item_category

)t;

#####################05-24

--商品交互的总人数（全部，最近1天，3天）

drop table if exists${table_label}_item_features_3;

create table${table_label}_item_features_3 as

select item_id,

--全部

sum(casewhen behavior_type=1 then 1 else 0 end) as i25,

sum(casewhen behavior_type=2 then 1 else 0 end) as i26,

sum(casewhen behavior_type=3 then 1 else 0 end) as i27,

sum(casewhen behavior_type=4 then 1 else 0 end) as i28,

--最近1天

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)=1 then 1 else 0 end) as i29,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)=1 then 1 else 0 end) as i30,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)=1 then 1 else 0 end) as i31,

sum(casewhen behavior_type=4 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)=1 then 1 else 0 end) as i32,

--最近3天

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)<4 then 1 else 0 end) as i33,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)<4 then 1 else 0 end) as i34,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)<4 then 1 else 0 end) as i35,

sum(casewhen behavior_type=4 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)<4 then 1 else 0 end) as i36

from

(

selectdistinct item_id,user_id,time,behavior_type

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

group by item_id;

--商品行为数（最近1天，3天）

drop table if exists${table_label}_item_features_4;

create table${table_label}_item_features_4 as

select item_id,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)=1 then 1 else 0 end) as i37,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)=1 then 1 else 0 end) as i38,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)=1 then 1 else 0 end) as i39,

sum(casewhen behavior_type=4 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)=1 then 1 else 0 end) as i40,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)<4 then 1 else 0 end) as i41,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)<4 then 1 else 0 end) as i42,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)<4 then 1 else 0 end) as i43,

sum(casewhen behavior_type=4 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)<4 then 1 else 0 end) as i44

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

group by item_id;

--商品的购买转化率及转化率与类别平均转化率的比值

drop table if exists${table_label}_item_features_5;

create table${table_label}_item_features_5 as

select item_id,r1 asi45,r2 as i46,r3 as i47,

casewhen cr1>0 then r1/cr1 else 0 end as i48,

casewhen cr2>0 then r2/cr2 else 0 end as i49,

casewhen cr3>0 then r3/cr3 else 0 end as i50

from

(

selectitem_id,item_category,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=1 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as r1,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=2 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as r2,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=3 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as r3

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

group byitem_id,item_category

) t1

join

(

selectitem_category,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=1 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as cr1,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=1 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as cr2,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=1 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as cr3

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

groupby item_category

) t2

ont1.item_category=t2.item_category;

--商品行为/同类同行为均值（总表，类别行为统计）

drop table if exists${table_label}_item_features_6;

create table${table_label}_item_features_6 as

select item_id,

casewhen t2.click>0 then t1.click/t2.click else 0 end as i51,

casewhen t2.favorite>0 then t1.favorite/t2.favorite else 0 end i52,

casewhen t2.cart>0 then t1.cart/t2.cart else 0 end i53,

casewhen t2.buy>0 then t1.buy/t2.buy else 0 end i54

from

(

--用户的行为数

selectitem_id,item_category,

sum(casewhen behavior_type=1 then 1 else 0 end) as click,

sum(casewhen behavior_type=2 then 1 else 0 end) as favorite,

sum(casewhen behavior_type=3 then 1 else 0 end) as cart,

sum(casewhen behavior_type=4 then 1 else 0 end) as buy

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

groupby item_id,item_category

)t1

join

(

--类别的平均行为数

selectitem_category,

avg(casewhen behavior_type=1 then 1 else 0 end) as click,

avg(casewhen behavior_type=1 then 1 else 0 end) as favorite,

avg(casewhen behavior_type=1 then 1 else 0 end) as cart,

avg(casewhen behavior_type=1 then 1 else 0 end) as buy

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

groupby item_category

)t2

ont1.item_category=t2.item_category;

--合并item特征。此处需要常常做变动，添加新的特征之后都需要做改变

drop table if exists${table_label}_item_features;

create table${table_label}_item_features as

select

${table_label}_item_features_1.item_id,i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15,i16,i17,i18,i19,i20,i21,i22,i23,i24

,i25,i26,i27,i28,i29,i30,i31,i32,i33,i34,i35,i36,i37,i38,i39,i40,i41,i42,i43,i44,i45,i46,i47,i48,i49,i50,i51,i52,i53,i54

from

${table_label}_item_features_1

join

${table_label}_item_features_2

on${table_label}_item_features_1.item_id=${table_label}_item_features_2.item_id

join

${table_label}_item_features_3

on${table_label}_item_features_1.item_id=${table_label}_item_features_3.item_id

join

${table_label}_item_features_4

on${table_label}_item_features_1.item_id=${table_label}_item_features_4.item_id

join

${table_label}_item_features_5

on${table_label}_item_features_1.item_id=${table_label}_item_features_5.item_id

join

${table_label}_item_features_6

on${table_label}_item_features_1.item_id=${table_label}_item_features_6.item_id;

------------------------------------------------------------------------------

使用label_day之前的数据生成用户-商品特征

已成功

drop table if exists${table_label}_ui_features_1;

create table${table_label}_ui_features_1 as

SELECT

user_id,item_id,

--平均每天对商品的行为数

sum(case when (behavior_type=1 andtime<‘${label_day}‘) then 1 else 0 end)/30 as ui1,

sum(case when (behavior_type=2 andtime<‘${label_day}‘) then 1 else 0 end)/30 as ui2,

sum(case when (behavior_type=3 andtime<‘${label_day}‘) then 1 else 0 end)/30 as ui3,

sum(case when (behavior_type=4 andtime<‘${label_day}‘) then 1 else 0 end)/30 as ui4,

--最近第一天的操作

sum(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then 1 else 0 end) as ui5,

sum(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then 1 else 0 end) as ui6,

sum(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then 1 else 0 end) as ui7,

--最近第二天的操作

sum(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=2) then 1 else 0 end) as ui8,

sum(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=2) then 1 else 0 end) as ui9,

sum(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=2) then 1 else 0 end) as ui10,

sum(case when (behavior_type=4 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=2) then 1 else 0 end) as ui11,

--最近第三天的操作

sum(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=3) then 1 else 0 end) as ui12,

sum(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=3) then 1 else 0 end) as ui13,

sum(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=3) then 1 else 0 end) as ui14,

sum(case when (behavior_type=4 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=3) then 1 else 0 end) as ui15,

--最近1周的操作

sum(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<8) then 1 else 0 end) as ui16,

sum(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<8) then 1 else 0 end) as ui17,

sum(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<8) then 1 else 0 end) as ui18,

sum(case when (behavior_type=4 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<8) then 1 else 0 end) as ui19,

--最近一天的最后的操作时间

max(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 0 end) as ui20,

max(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 0 end) as ui21,

max(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 0 end) as ui22,

--最近一天的最早的操作时间

min(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 24 end) as ui23,

min(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 24 end) as ui24,

min(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 24 end) as ui25,

--2015-5-16添加

--用户最近的交互离现在的时间

min(case when behavior_type=1 thendatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘hh‘) else 744 end) as ui26,

min(case when behavior_type=2 thendatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘hh‘) else 744 end) as ui27,

min(case when behavior_type=3 thendatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘hh‘) else 744 end) as ui28,

min(case when behavior_type=4 thendatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘hh‘) else 744 end) as ui29,

--用户商品交互的总天数

count(distinct case when behavior_type=1then substr(time,7,4) end) as ui30,

count(distinct case when behavior_type=4then substr(time,7,4) end) as ui31,

--最近3天的操作

sum(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4) then 1 else 0 end) as ui32,

sum(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4) then 1 else 0 end) as ui33,

sum(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4) then 1 else 0 end) as ui34,

sum(case when (behavior_type=4 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4) then 1 else 0 end) as ui35

FROMtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

GROUP BYuser_id,item_id;

2015-5-21添加

最近24小时用户的各种行为数（4*24）

drop table if exists${table_label}_ui_features_2;

create table${table_label}_ui_features_2 as

selectuser_id,item_id,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=24 then 1 else 0 end) as ui36,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=23 then 1 else 0 end) as ui37,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=22 then 1 else 0 end) as ui38,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=21 then 1 else 0 end) as ui39,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=20 then 1 else 0 end) as ui40,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=19 then 1 else 0 end) as ui41,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=18 then 1 else 0 end) as ui42,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=17 then 1 else 0 end) as ui43,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=16 then 1 else 0 end) as ui44,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=15 then 1 else 0 end) as ui45,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=14 then 1 else 0 end) as ui46,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=13 then 1 else 0 end) as ui47,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=12 then 1 else 0 end) as ui48,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=11 then 1 else 0 end) as ui49,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=10 then 1 else 0 end) as ui50,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=9 then 1 else 0 end) as ui51,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=8 then 1 else 0 end) as ui52,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=7 then 1 else 0 end) as ui53,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=6 then 1 else 0 end) as ui54,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=5 then 1 else 0 end) as ui55,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=4 then 1 else 0 end) as ui56,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=3 then 1 else 0 end) as ui57,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=2 then 1 else 0 end) as ui58,

sum(casewhen behavior_type=1 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=1 then 1 else 0 end) as ui59,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=24 then 1 else 0 end) as ui60,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=23 then 1 else 0 end) as ui61,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=22 then 1 else 0 end) as ui62,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=21 then 1 else 0 end) as ui63,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=20 then 1 else 0 end) as ui64,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=19 then 1 else 0 end) as ui65,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=18 then 1 else 0 end) as ui66,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=17 then 1 else 0 end) as ui67,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=16 then 1 else 0 end) as ui68,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=15 then 1 else 0 end) as ui69,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=14 then 1 else 0 end) as ui70,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=13 then 1 else 0 end) as ui71,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=12 then 1 else 0 end) as ui72,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=11 then 1 else 0 end) as ui73,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=10 then 1 else 0 end) as ui74,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=9 then 1 else 0 end) as ui75,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=8 then 1 else 0 end) as ui76,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=7 then 1 else 0 end) as ui77,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=6 then 1 else 0 end) as ui78,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=5 then 1 else 0 end) as ui79,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=4 then 1 else 0 end) as ui80,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=3 then 1 else 0 end) as ui81,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=2 then 1 else 0 end) as ui82,

sum(casewhen behavior_type=2 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=1 then 1 else 0 end) as ui83,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=24 then 1 else 0 end) as ui84,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=23 then 1 else 0 end) as ui85,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=22 then 1 else 0 end) as ui86,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=21 then 1 else 0 end) as ui87,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=20 then 1 else 0 end) as ui88,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=19 then 1 else 0 end) as ui89,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=18 then 1 else 0 end) as ui90,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=17 then 1 else 0 end) as ui91,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=16 then 1 else 0 end) as ui92,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=15 then 1 else 0 end) as ui93,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=14 then 1 else 0 end) as ui94,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=13 then 1 else 0 end) as ui95,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=12 then 1 else 0 end) as ui96,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=11 then 1 else 0 end) as ui97,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=10 then 1 else 0 end) as ui98,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=9 then 1 else 0 end) as ui99,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=8 then 1 else 0 end) as ui100,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=7 then 1 else 0 end) as ui101,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=6 then 1 else 0 end) as ui102,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=5 then 1 else 0 end) as ui103,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=4 then 1 else 0 end) as ui104,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=3 then 1 else 0 end) as ui105,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=2 then 1 else 0 end) as ui106,

sum(casewhen behavior_type=3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘hh‘)=1 then 1 else 0 end) as ui107

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheredatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1

group byuser_id,item_id;

--用户A对品牌B的总购买数\收藏数\购物车数（总表,测试通过）

--用户A对品牌B的点击数的平方（总表）

--用户A对品牌B的购买数的平方（总表）

--用户A对品牌B的点击购买比（可用总表替代）

drop table if exists${table_label}_ui_features_2;

create table${table_label}_ui_features_2 as

selectuser_id,item_id,

sum(casewhen behavior_type=1 then 1 else 0 end) as ui36,

sum(casewhen behavior_type=2 then 1 else 0 end) as ui37,

sum(casewhen behavior_type=3 then 1 else 0 end) as ui38,

sum(casewhen behavior_type=4 then 1 else 0 end) as ui39,

sum(casewhen behavior_type=1 then 1 else 0 end)*sum(case when behavior_type=1 then 1else 0 end) as ui40,

sum(casewhen behavior_type=2 then 1 else 0 end)*sum(case when behavior_type=2 then 1else 0 end) as ui41,

sum(casewhen behavior_type=3 then 1 else 0 end)*sum(case when behavior_type=2 then 1else 0 end) as ui42,

sum(casewhen behavior_type=4 then 1 else 0 end)*sum(case when behavior_type=4 then 1else 0 end) as ui43,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=1 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as ui44,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=2 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as ui45,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=3 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as ui46,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=4 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as ui47

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

group byuser_id,item_id;

--测试通过

--用户交互本商品前，交互的商品数（4）（时间上为>=）

--用户交互本商品后，交互的商品数（4）（时间上为<=）

drop table if exists${table_label}_ui_features_3;

create table${table_label}_ui_features_3 as

selectt1.user_id,t1.item_id,

sum(casewhen t1.behavior_type=1 and t2.behavior_type=1 and t2.hour<=t1.s then 1 else0 end) as ui48,

sum(casewhen t1.behavior_type=2 and t2.behavior_type=2 and t2.hour<=t1.s then 1 else0 end) as ui49,

sum(casewhen t1.behavior_type=3 and t2.behavior_type=3 and t2.hour<=t1.s then 1 else0 end) as ui50,

sum(casewhen t2.behavior_type=4 and t2.hour<=t1.s then 1 else 0 end) as ui51,

sum(casewhen t1.behavior_type=1 and t2.behavior_type=1 and t2.hour>t1.e then 1 else0 end) as ui52,

sum(casewhen t1.behavior_type=2 and t2.behavior_type=1 and t2.hour>t1.e then 1 else0 end) as ui53,

sum(casewhen t1.behavior_type=3 and t2.behavior_type=1 and t2.hour>t1.e then 1 else0 end) as ui54,

sum(casewhen t2.behavior_type=4 and t2.hour>t1.e then 1 else 0 end) as ui55

from

(--前一天的交互行为最早最晚时间

selectuser_id,item_id,behavior_type,min(cast(substr(time,-2,2) as bigint)) ass,max(cast(substr(time,-2,2) as bigint)) as e

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheredatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1

groupby user_id,item_id,behavior_type

) t1

join

--最近一天用户与商品的交互行为

(

selectdistinct user_id,item_id,behavior_type,cast(substr(time,-2,2) as bigint) ashour

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheredatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1

) t2

ont1.user_id=t2.user_id

group byt1.user_id,t1.item_id;

--合并ui特征，这里需要常常改动

drop table if exists${table_label}_ui_features;

create table${table_label}_ui_features as

SELECT

${table_label}_ui_features_1.user_id,${table_label}_ui_features_1.item_id,

ui1,ui2,ui3,ui4,ui5,ui6,ui7,ui8,ui9,ui10,ui11,ui12,ui13,ui14,ui15,ui16,ui17,ui18,ui19,ui20,ui21,ui22,ui23,ui24,ui25,ui26,ui27,ui28,ui29,ui30,ui31,ui32,ui33,ui34,ui35

--ui_features_2

,case when ui36 is not null then ui36 else0 end as ui36

,casewhen ui37 is not null then ui37 else 0 end as ui37

,casewhen ui38 is not null then ui38 else 0 end as ui38

,casewhen ui39 is not null then ui39 else 0 end as ui39

,casewhen ui40 is not null then ui40 else 0 end as ui40

,casewhen ui41 is not null then ui41 else 0 end as ui41

,casewhen ui42 is not null then ui42 else 0 end as ui42

,casewhen ui43 is not null then ui43 else 0 end as ui43

,casewhen ui44 is not null then ui44 else 0 end as ui44

,casewhen ui45 is not null then ui45 else 0 end as ui45

,casewhen ui46 is not null then ui46 else 0 end as ui46

,casewhen ui47 is not null then ui47 else 0 end as ui47

,casewhen ui48 is not null then ui48 else 0 end as ui48

,casewhen ui49 is not null then ui49 else 0 end as ui49

,casewhen ui50 is not null then ui50 else 0 end as ui50

,casewhen ui51 is not null then ui51 else 0 end as ui51

,casewhen ui52 is not null then ui52 else 0 end as ui52

,casewhen ui53 is not null then ui53 else 0 end as ui53

,casewhen ui54 is not null then ui54 else 0 end as ui54

,casewhen ui55 is not null then ui55 else 0 end as ui55

FROM${table_label}_ui_features_1

left outer join${table_label}_ui_features_2

on${table_label}_ui_features_1.user_id=${table_label}_ui_features_2.user_id and${table_label}_ui_features_1.item_id=${table_label}_ui_features_2.item_id

left outer join${table_label}_ui_features_3

on${table_label}_ui_features_1.user_id=${table_label}_ui_features_3.user_id and${table_label}_ui_features_1.item_id=${table_label}_ui_features_3.item_id;

使用label_day之前的数据生成user特征

已成功

drop table if exists${table_label}_user_features_1;

create table${table_label}_user_features_1 as

select user_id,

--最近1天用户行为统计

sum(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then 1 else 0 end) as u1,

sum(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then 1 else 0 end) as u2,

sum(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then 1 else 0 end) as u3,

sum(case when (behavior_type=4 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then 1 else 0 end) as u4,

--最近3天用户行为统计

sum(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)>0) then 1 else 0 end) as u5,

sum(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)>0) then 1 else 0 end) as u6,

sum(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)>0) then 1 else 0 end) as u7,

sum(case when (behavior_type=4 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<4 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)>0) then 1 else 0 end) as u8,

--过去全部的用户行为统计

sum(case when (behavior_type=1 andtime<‘${label_day}‘) then 1 else 0 end) as u9,

sum(case when (behavior_type=2 andtime<‘${label_day}‘) then 1 else 0 end) as u10,

sum(case when (behavior_type=3 andtime<‘${label_day}‘) then 1 else 0 end) as u11,

sum(case when (behavior_type=4 andtime<‘${label_day}‘) then 1 else 0 end) as u12,

--2015-5-15添加

--用户前一天最早的交互行为时间

min(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 24 end) as u13,

min(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 24 end) as u14,

min(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 24 end) as u15,

--用户前一天最晚的交互行为时间

max(case when (behavior_type=1 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 0 end) as u16,

max(case when (behavior_type=2 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 0 end) as u17,

max(case when (behavior_type=3 anddatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1) then cast(substr(time,-2,2) as bigint) else 0 end) as u18,

--最近一天用户交互了多少种item

count(distinct case when behavior_type=1and datediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1 then item_id end) as u19,

count(distinct case when behavior_type=2and datediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1 then item_id end) as u20,

count(distinct case when behavior_type=3and datediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1 then item_id end) as u21,

count(distinct case when behavior_type=4and datediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1 then item_id end) as u22

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

group by user_id;

--------------------------------------------------------------------------------------------------

--最近1天用户交互了多少种类别(测试通过)

drop table if exists${table_label}_user_features_2;

create table${table_label}_user_features_2 as

selectuser_id,count(item_category) as u23 from

(

selectdistinct user_id,item_category

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheredatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=1

group by user_id;

--用户的购买转化率(测试通过)

drop table if exists${table_label}_user_features_3;

create table${table_label}_user_features_3 as

select user_id,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=1 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as u24,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=2 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as u25,

casewhen sum(case when behavior_type=4 then 1 else 0 end)>0 then sum(case whenbehavior_type=3 then 1 else 0 end)/sum(case when behavior_type=4 then 1 else 0end) else 0 end as u26

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

group by user_id;

--用户对交互过的商品的平均交互数，最大交互数，最小交互数(测试通过)

drop table if exists${table_label}_user_features_4;

create table${table_label}_user_features_4 as

select user_id,

max(cli)as u27,

min(cli)as u28,

avg(cli)as u29,

max(fav)as u30,

min(fav)as u31,

avg(fav)as u32,

max(car)as u33,

min(car)as u34,

avg(car)as u35,

max(buy)as u36,

min(buy)as u37,

avg(buy)as u38

from

(

selectuser_id,item_id,

sum(casewhen behavior_type=1 then 1 else 0 end) as cli,

sum(casewhen behavior_type=2 then 1 else 0 end) as fav,

sum(casewhen behavior_type=3 then 1 else 0 end) as car,

sum(casewhen behavior_type=4 then 1 else 0 end) as buy

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

groupby user_id,item_id

) t

group by user_id;

--用户购买商品的时间（平均，最早，最晚）#生活习惯(测试通过)

drop table if exists${table_label}_user_features_5;

create table${table_label}_user_features_5 as

select user_id,

max(cast(substr(time,-2,2)as bigint)) as u39,

min(cast(substr(time,-2,2)as bigint)) as u40,

avg(cast(substr(time,-2,2)as bigint)) as u41

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

group by user_id;

--用户有交互的天数(测试通过)

drop table if exists${table_label}_user_features_6;

create table${table_label}_user_features_6 as

select user_id,

sum(casewhen behavior_type=1 then 1 else 0 end) as u42,

sum(casewhen behavior_type=2 then 1 else 0 end) as u43,

sum(casewhen behavior_type=3 then 1 else 0 end) as u44,

sum(casewhen behavior_type=4 then 1 else 0 end) as u45

from

(

selectdistinct user_id,behavior_type,substr(time,9,2) as daynum

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheretime<‘${label_day}‘

) t

group by user_id;

--合并user特征

drop table if exists${table_label}_user_features;

create table${table_label}_user_features as

select

${table_label}_user_features_1.user_id,u1,u2,u3,u4,u5,u6,u7,u8,u9,u10,u11,u12,u13,u14,u15,u16,u17,u18,u19,u20,u21,u22

,casewhen u23 is not null then u23 else 0 end as u23

,casewhen u24 is not null then u24 else 0 end as u24

,casewhen u25 is not null then u25 else 0 end as u25

,casewhen u26 is not null then u26 else 0 end as u26

,casewhen u27 is not null then u27 else 0 end as u27

,casewhen u28 is not null then u28 else 0 end as u28

,casewhen u29 is not null then u29 else 0 end as u29

,casewhen u30 is not null then u30 else 0 end as u30

,casewhen u31 is not null then u31 else 0 end as u31

,casewhen u32 is not null then u32 else 0 end as u32

,casewhen u33 is not null then u33 else 0 end as u33

,casewhen u34 is not null then u34 else 0 end as u34

,casewhen u35 is not null then u35 else 0 end as u35

,casewhen u36 is not null then u36 else 0 end as u36

,casewhen u37 is not null then u37 else 0 end as u37

,casewhen u38 is not null then u38 else 0 end as u38

,casewhen u39 is not null then u39 else 0 end as u39

,casewhen u40 is not null then u40 else 0 end as u40

,casewhen u41 is not null then u41 else 0 end as u41

,casewhen u42 is not null then u42 else 0 end as u42

,casewhen u43 is not null then u43 else 0 end as u43

,casewhen u44 is not null then u44 else 0 end as u44

,casewhen u45 is not null then u45 else 0 end as u45

from${table_label}_user_features_1

left outer join${table_label}_user_features_2

on${table_label}_user_features_1.user_id=${table_label}_user_features_2.user_id

left outer join${table_label}_user_features_3

on${table_label}_user_features_1.user_id=${table_label}_user_features_3.user_id

left outer join${table_label}_user_features_4

on${table_label}_user_features_1.user_id=${table_label}_user_features_4.user_id

left outer join${table_label}_user_features_5

on${table_label}_user_features_1.user_id=${table_label}_user_features_5.user_id

left outer join${table_label}_user_features_6

on${table_label}_user_features_1.user_id=${table_label}_user_features_6.user_id;

--------------------------------------------------------------------------------

-- 分类标签 --

--------------------------------------------------------------------------------

--用于预测的作用对（label_day的前两天有交互且无购买）

--这个相当于模型之前的规则过滤，是一个常常需要调整的地方

drop table if existstemp_pairs_table;

create tabletemp_pairs_table as

selectuser_id,item_id

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheredatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)<3 and datediff(to_date(‘${label_day}‘,‘yyyy-mm-ddhh‘),to_date(time,‘yyyy-mm-dd hh‘),‘dd‘)>0

group byuser_id,item_id

having sum(case whenbehavior_type=4 then 1 else 0 end)=0;

--生成分类标签

drop table if existstemp_label_table;

create tabletemp_label_table as

selecta.user_id,a.item_id,case when b.user_id is null and b.item_id is null then 0else 1 end as tag

from

temp_pairs_table a

left outer join

(

--分类标签

select distinct user_id,item_id

fromtianchi_lbs.tianchi_mobile_recommend_train_user

wheredatediff(to_date(‘${label_day}‘,‘yyyy-mm-dd hh‘),to_date(time,‘yyyy-mm-ddhh‘),‘dd‘)=0 and behavior_type=4

) b

ona.user_id=b.user_id and a.item_id=b.item_id;

--------------------------------------------------------------------------------

-- 数据融合 --

--------------------------------------------------------------------------------

--融合特征，生成特征总表（没有归一化处理）

drop table if exists${table_label}_feature_table;

create table${table_label}_feature_table as

selecta.user_id,a.item_id,a.tag,

--item特征

i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15,i16,i17,i18,i19,i20,i21,i22,i23,i24

,i25,i26,i27,i28,i29,i30,i31,i32,i33,i34,i35,i36,i37,i38,i39,i40,i41,i42,i43,i44,i45,i46,i47,i48,i49,i50,i51,i52,i53,i54

--ui特征

,ui1,ui2,ui3,ui4,ui5,ui6,ui7,ui8,ui9,ui10,ui12,ui13,ui14,ui15,ui16,ui17,ui18,ui19,ui20,ui21,ui22,ui23,ui24,ui25,ui26,ui27,ui28,ui29,ui30,ui31,ui32,ui33,ui34,ui35

,ui36,ui37,ui38,ui39,ui40,ui41,ui42,ui43,ui44,ui45,ui46,ui47,ui48,ui49,ui50,ui51,ui52,ui53,ui54,ui55

--user特征

,u1,u2,u3,u4,u5,u6,u7,u8,u9,u10,u11,u12,u13,u14,u15,u16,u17,u18,u19,u20,u21,u22

,u23,u24,u25,u26,u27,u28,u29,u30,u31,u32,u33,u34,u35,u36,u37,u38,u39,u40,u41,u42,u43,u44,u45

fromtemp_label_table a

left outer join${table_label}_item_features

ona.item_id=${table_label}_item_features.item_id

left outer join${table_label}_ui_features

ona.user_id=${table_label}_ui_features.user_id anda.item_id=${table_label}_ui_features.item_id

left outer join${table_label}_user_features

ona.user_id=${table_label}_user_features.user_id;

--删除中间表

drop table if existstemp_label_table;

drop table if existstemp_pairs_table;

--------------------------------------------------------------------------------

-- 归一化 --

--------------------------------------------------------------------------------

--对特征做归一化处理

drop table if exists${table_label}_normal;

PAI -name Normalize-project algo_public -DkeepOriginal="false"-DoutputTableName="${table_label}_normal"-DinputTableName="${table_label}_feature_table"

-DselectedColNames="i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15,i16,i17,i18,i19,i20,i21,i22,i23,i24,i25,i26,i27,i28,i29,i30,i31,i32,i33,i34,i35,i36,i37,i38,i39,i40,i41,i42,i43,i44,i45,i46,i47,i48,i49,i50,i51,i52,i53,i54,ui1,ui2,ui3,ui4,ui5,ui6,ui7,ui8,ui9,ui10,ui12,ui13,ui14,ui15,ui16,ui17,ui18,ui19,ui20,ui21,ui22,ui23,ui24,ui25,ui26,ui27,ui28,ui29,ui30,ui31,ui32,ui33,ui34,ui35,ui36,ui37,ui38,ui39,ui40,ui41,ui42,ui43,ui44,ui45,ui46,ui47,ui48,ui49,ui50,ui51,ui52,ui53,ui54,ui55,u1,u2,u3,u4,u5,u6,u7,u8,u9,u10,u11,u12,u13,u14,u15,u16,u17,u18,u19,u20,u21,u22,u23,u24,u25,u26,u27,u28,u29,u30,u31,u32,u33,u34,u35,u36,u37,u38,u39,u40,u41,u42,u43,u44,u45";

--------------------------------------------------------------------------------

-- 欠采样 --

--------------------------------------------------------------------------------

--只对训练数据集有意义

--拆分数据为正例和反例

drop table if existstemp_pos_table;

create tabletemp_pos_table as

select * from${table_label}_normal where tag=1;

drop table if existstemp_nat_table;

create tabletemp_nat_table as

select * from${table_label}_normal where tag=0;

--对反例数据做欠采样

drop table if existssub_nat_table;

PAI -nameRandomSample -project algo_public -Dreplace="false"-DoutTableName="sub_nat_table" -DsampleSize="2000000"-DinputTableName="temp_nat_table";

--合并采样后的正反例数据，删除零时表

drop table if exists${table_label}_under_sample;

create table${table_label}_under_sample as

select * from

(

select * from sub_nat_table

union all

select * from temp_pos_table

) t;

drop table if existssub_nat_table;

drop table if existstemp_nat_table;

drop table if existstemp_pos_table;

--使用lr-gbdt串联分类器，使用lr过滤掉大部分反例样本，然后再使用二级分类器做分类

--来个约定先

--lr分类器（一级分类器）：lr_model_0622

--gbdt分类器（二级分类器）:gbdt_model_0622

-- --清洁工先上场清扫垃圾

-- --清除上次的分类器

DROP OFFLINEMODEL IFEXISTS lr_model_0626;

drop offlinemodel ifexists gbdt_model_0626;

-- --训练工作开始啦！！！

-- --切分16日数据为两个部分各50%

drop table if existstrain_p1;

drop table if existstrain_p2;

PAI -nameDeclustering -project algo_public -Dfactors="0.5,0.5"-DoutputTableNames="train_p1,train_p2"-DinputTableName="16_under_sample";

--选择训练数据集训练一级分类器（16日train_p1数据）

DROP OFFLINEMODEL IFEXISTS lr_model_0626;

PAI -nameLogisticRegression -project algo_public -DmodelName="lr_model_0626"

-DregularizedLevel="20"-DmaxIter="200" -Depsilon="0.000001"-DlabelColName="tag"

-DfeatureColNames="i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15,i16,i17,i18,i19,i20,i21,i22,i23,i24,i25,i26,i27,i28,i29,i30,i31,i32,i33,i34,i35,i36,i37,i38,i39,i40,i41,i42,i43,i44,i45,i46,i47,i48,i49,i50,i51,i52,i53,i54,ui1,ui2,ui3,ui4,ui5,ui6,ui7,ui8,ui9,ui10,ui12,ui13,ui14,ui15,ui16,ui17,ui18,ui19,ui20,ui21,ui22,ui23,ui24,ui25,ui26,ui27,ui28,ui29,ui30,ui31,ui32,ui33,ui34,ui35,ui36,ui37,ui38,ui39,ui40,ui41,ui42,ui43,ui44,ui45,ui46,ui47,ui48,ui49,ui50,ui51,ui52,ui53,ui54,ui55,u1,u2,u3,u4,u5,u6,u7,u8,u9,u10,u11,u12,u13,u14,u15,u16,u17,u18,u19,u20,u21,u22,u23,u24,u25,u26,u27,u28,u29,u30,u31,u32,u33,u34,u35,u36,u37,u38,u39,u40,u41,u42,u43,u44,u45"

-DgoodValue="1"-DinputTableName="train_p1";

--使用一级分类器预测train_p2

drop table if existsp2_after_lr;

PAI -name Prediction-project algo_public -DdetailColName="prediction_detail1"

-DsplitCharacteristic="1" -DmodelName="lr_model_0626"-DresultColName="prediction_result1"-DoutputTableName="p2_after_lr"-DscoreColName="prediction_score1" -DinputTableName="train_p2"-DlabelValueToPredict="1"

-DappendColNames="user_id,item_id,tag,i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15,i16,i17,i18,i19,i20,i21,i22,i23,i24,i25,i26,i27,i28,i29,i30,i31,i32,i33,i34,i35,i36,i37,i38,i39,i40,i41,i42,i43,i44,i45,i46,i47,i48,i49,i50,i51,i52,i53,i54,ui1,ui2,ui3,ui4,ui5,ui6,ui7,ui8,ui9,ui10,ui12,ui13,ui14,ui15,ui16,ui17,ui18,ui19,ui20,ui21,ui22,ui23,ui24,ui25,ui26,ui27,ui28,ui29,ui30,ui31,ui32,ui33,ui34,ui35,ui36,ui37,ui38,ui39,ui40,ui41,ui42,ui43,ui44,ui45,ui46,ui47,ui48,ui49,ui50,ui51,ui52,ui53,ui54,ui55,u1,u2,u3,u4,u5,u6,u7,u8,u9,u10,u11,u12,u13,u14,u15,u16,u17,u18,u19,u20,u21,u22,u23,u24,u25,u26,u27,u28,u29,u30,u31,u32,u33,u34,u35,u36,u37,u38,u39,u40,u41,u42,u43,u44,u45";

--拆分数据集（正例有2.3，反例有528）

drop table if existstemp_nat;

create tabletemp_nat as

select * fromp2_after_lr where tag=0 and prediction_score1>0.1;

drop table if existstemp_pos;

create tabletemp_pos as

select * fromp2_after_lr where tag=1 and prediction_score1>0.1;

-- 对反例做欠采样160000

drop table if existssub_nat;

PAI -nameRandomSample -project algo_public -DoutTableName="sub_nat"

-DsampleSize="160000"-DinputTableName="temp_nat";

-- 对正例做欠采样80000,60000,40000

drop table if existssub_pos;

PAI -nameRandomSample -project algo_public -DoutTableName="sub_pos"

-DsampleSize="65000"-DinputTableName="temp_pos";

--构建二级训练数据

drop table if existstrain_2;

create table train_2as

select * from

(

select * fromsub_pos

union all

select * fromsub_nat

)t;

--训练二级分类器

drop offlinemodel ifexists gbdt_model_0630;

PAI -name GBDT_LR-project algo_public -DfeatureSplitValueMaxSize="500"-DrandSeed="0"

-Dshrinkage="0.05"-DmaxLeafCount="32" -DlabelColName="tag"-DinputTableName="train_2"

-DminLeafSampleCount="500"-DsampleRatio="0.6" -DmaxDepth="11"-DmodelName="gbdt_model_0630" -DmetricType="2"-DfeatureRatio="0.6" -DtestRatio="0.2" -DtreeCount="500"

-- 评测的分类器性能

--测试数据集使用17日测试数据

drop table if existspre_after_lr;

PAI -name Prediction-project algo_public -DdetailColName="prediction_detail1"

-DsplitCharacteristic="1" -DmodelName="lr_model_0626"-DresultColName="prediction_result1"-DoutputTableName="pre_after_lr"-DscoreColName="prediction_score1" -DinputTableName="17_normal"-DlabelValueToPredict="1"

-- 使用一级分类器过滤数据

drop table if existspre_for_gbdt;

create tablepre_for_gbdt as

selectuser_id,item_id,tag,i1,i2,i3,i4,i5,i6,i7,i8,i9,i10,i11,i12,i13,i14,i15,i16,i17,i18,i19,i20,i21,i22,i23,i24,i25,i26,i27,i28,i29,i30,i31,i32,i33,i34,i35,i36,i37,i38,i39,i40,i41,i42,i43,i44,i45,i46,i47,i48,i49,i50,i51,i52,i53,i54,ui1,ui2,ui3,ui4,ui5,ui6,ui7,ui8,ui9,ui10,ui12,ui13,ui14,ui15,ui16,ui17,ui18,ui19,ui20,ui21,ui22,ui23,ui24,ui25,ui26,ui27,ui28,ui29,ui30,ui31,ui32,ui33,ui34,ui35,ui36,ui37,ui38,ui39,ui40,ui41,ui42,ui43,ui44,ui45,ui46,ui47,ui48,ui49,ui50,ui51,ui52,ui53,ui54,ui55,u1,u2,u3,u4,u5,u6,u7,u8,u9,u10,u11,u12,u13,u14,u15,u16,u17,u18,u19,u20,u21,u22,u23,u24,u25,u26,u27,u28,u29,u30,u31,u32,u33,u34,u35,u36,u37,u38,u39,u40,u41,u42,u43,u44,u45

from pre_after_lrwhere prediction_score1>0.1;

-- 使用二级分类器预测

drop table if existspre_after_gbdt;

PAI -name Prediction-project algo_public -DdetailColName="prediction_detail2"-DsplitCharacteristic="1"-DappendColNames="user_id,item_id,tag"

-DmodelName="gbdt_model_0629"-DresultColName="prediction_result2"-DoutputTableName="pre_after_gbdt"

-DscoreColName="prediction_score2"-DinputTableName="pre_for_gbdt" -DlabelValueToPredict="1";

--计算混淆矩阵

drop table if existsgbdtlr_confusion;

PAI -nameconfusionmatrix -project algo_public-DoutputTableName="gbdtlr_confusion" -DlabelColName="tag"-DpredictionColName="prediction_result2"-DinputTableName="pre_after_gbdt";

--显示混淆矩阵内容

select * from gbdtlr_confusion;

--------------------------------------------------------------------------------

--使用18日测试数据

drop table if existspre_after_lr_181;

PAI -name Prediction-project algo_public -DdetailColName="prediction_detail1"

-DsplitCharacteristic="1" -DmodelName="lr_model_0626"-DresultColName="prediction_result1"-DoutputTableName="pre_after_lr_181"-DscoreColName="prediction_score1" -DinputTableName="18_normal"-DlabelValueToPredict="1"

-- 使用一级分类器过滤数据

drop table if existspre_for_gbdt_181;

create tablepre_for_gbdt_181 as

frompre_after_lr_181 where prediction_score1>0.1;

--使用二级分类器预测

drop table if existspre_after_gbdt_181;

PAI -name Prediction-project algo_public -DdetailColName="prediction_detail2"-DsplitCharacteristic="1"-DappendColNames="user_id,item_id,tag"

-DmodelName="gbdt_model_0630"-DresultColName="prediction_result2"-DoutputTableName="pre_after_gbdt_181"

-DscoreColName="prediction_score2"-DinputTableName="pre_for_gbdt_181"-DlabelValueToPredict="1";

drop table if existstianchi_mobile_recommendation_predict_0630;

create tabletianchi_mobile_recommendation_predict_0630 as

select distinctb.user_id,b.item_id

from

tianchi_lbs.tianchi_mobile_recommend_train_itema

joinpre_after_gbdt_181 b

ona.item_id=b.item_id and b.prediction_result2="1";

1.平台允许有输入参数,点"高级运行即可",这里我设置了两个

--label_day：分类标签的日期，如：‘2014-12-18 00‘ 这是用18号的数据打标签

--table_label:保存的表名label，例如16 这是用16号及之前的数据产生特征

2.我仍然采用的是一天的数据,即18号.由于时间实在太紧,平台速度太慢,直接17号的训练,18号的上线测试,正负样本比例在17号做了选择,负样本16万,正样本选择了65000,效果就是最后的成绩了

技术分享

第二赛季由于投入时间太少,很多想法没实现,比如用两天的数据训练,还有就是试试其他模型,调节参数等等.挺遗憾的.不过参加这个比赛确实学到了很多东西,希望同学们多多参加类似的比赛.我在比赛期间,画了很多图表,整理后会再上传上来,分享给大家!另外我水平有限,有问题及需要交流的地方请及时联系我,我的QQ 1192625541,谢谢!

来自 <http://write.blog.csdn.net/postedit>