pig中查询top k，返回每个hour和ad_network_id下最大两个记录（SUBSTRING，order，COUNT_STAR，limit）

pig里面是有TOP函数，不知道为什么用不了。有时间要去看看pig源码了。

SET job.name 'top_k';

SET job.priority HIGH;





--REGISTER piggybank.jar;

REGISTER wizad-etl-udf-0.1.jar;



--DEFINE SequenceFileLoader org.apache.pig.piggybank.storage.SequenceFileLoader();

DEFINE SequenceFileLoader com.vpon.wizad.etl.pig.SequenceFileCSVLoader();



--%default cleanedLog /user/wizad/data/wizad/cleaned/2014-07-30/*/part*

%default cleanedLog /user/wizad/data/wizad/cleaned/$date/*/part*



%default output_path /user/wizad/tmp/hour_count



origin_cleaned_data = LOAD '$cleanedLog' USING SequenceFileLoader 

AS (ad_network_id:chararray,

    wizad_ad_id:chararray,

    guid:chararray,

    id:chararray,

    create_time:chararray,

    action_time:chararray,

    log_type:chararray, 

    ad_id:chararray,

    positioning_method:chararray,

    location_accuracy:chararray,

    lat:chararray, 

    lon:chararray,

    cell_id:chararray,

    lac:chararray,

    mcc:chararray,

    mnc:chararray,

    ip:chararray,

    connection_type:chararray,

    imei:chararray,

    android_id:chararray,

    android_advertising_id:chararray,

    udid:chararray,

    openudid:chararray,

    idfa:chararray,

    mac_address:chararray,

    uid:chararray,

    density:chararray,

    screen_height:chararray,

    screen_width:chararray,

    user_agent:chararray,

    app_id:chararray,

    app_category_id:chararray,

    device_model_id:chararray,

    carrier_id:chararray,

    os_id:chararray,

    device_type:chararray,

    os_version:chararray,

    country_region_id:chararray,

    province_region_id:chararray,

    city_region_id:chararray,

    ip_lat:chararray,

    ip_lon:chararray,

    quadkey:chararray);





show_log= FILTER origin_cleaned_data by log_type=='1';



--extract column for analyzing，提取子字段做为新属性

original_hour = FOREACH show_log GENERATE ad_network_id,wizad_ad_id,guid,app_category_id,log_type,SUBSTRING(create_time,11,13) AS hour;  --(wizad_ad_id,guid,log_type,hour)



hour_group = GROUP original_hour BY (hour,app_category_id);--按属性分类，



hour_count = foreach hour_group{

                                --guid_data = $1.guid;

                                --uniq_guid = distinct guid_data;--去重处理。查唯一个数。

                                ad_network_ids = original_hour.ad_network_id;

uniq_ad_network_ids = distinct ad_network_ids;

--统计每个包下的个数，将后面uniq_ad_network_ids分成单个记录。比如，uniq_ad_network_ids原值{3,5}，现变成两条记录，分为（xx，3）（xx，5）两条记录

generate flatten(group), COUNT_STAR($1) AS pv, flatten(uniq_ad_network_ids);

}



describe hour_count;

--查看结构为：hour_count: {group::hour: chararray,group::app_category_id: chararray,pv: long,uniq_ad_network_ids::ad_network_id: chararray}





group_hour_count = group hour_count by (hour,ad_network_id);

top_2_data = foreach group_hour_count {

                        --top_dataset = TOP(2,hour_count.pv, hour_count);--top函数 不能用。有谁用过告诉一声，我就不用看源码拉，哈哈

--hour_data = hour_count;

--top k实现方式，order排序，limit返回前k个。

order_hour_count = order hour_count by pv DESC;

top2_hour_count = limit order_hour_count  2;

--generate group, top2_hour_count.pv, top2_hour_count.app_category_id;-- 注意，后面是两个bag，分开的。

                        generate flatten(top2_hour_count );

}