foreach 内嵌的使用

foreach内部处理数据流的每条记录，进行关系操作，最后用generate返回数据给外部。但注意关系操作符不能作用于表达式，要将表达式提取成关系。

foreach内部只支持distinct, filter, limit, order；最后必须是generate；

foreach内部处理数据每次处理一条。对group后的关系（流对象），foreach内嵌每次传入一个group子集。

所以foreach内嵌之前都是先group关系，

也就是说，group后的数据，一次传进来一个group子集，可以分别按每个group进行统计等操作。（证明在最后。）

如统计每个exchange对应的symbol数：

--distinct_symbols.pig

daily   = load 'NYSE_daily' as (exchange, symbol); -- not interested in otherfields

grpd    = group daily by exchange;

uniqcnt = foreach grpd {

          sym      = daily.symbol;

          uniq_sym = distinct sym;

          generate group, COUNT(uniq_sym);

};

Foreach分别提取grpd的每个group，表达式变关系，被关系操作符处理。

 

 

使用forecah内嵌，去重的两个例子

1，分别统计两列的唯一数：group内提取两个子集，分别去重，统计

--double_distinct.pig

divs = load 'NYSE_dividends' as(exchange:chararray, symbol:chararray);

grpd = group divs all;

uniq = foreach grpd {

           exchanges      =divs.exchange;--this 'divs' is the $1 colum

           uniq_exchanges = distinct exchanges;

           symbols        = divs.symbol;

           uniq_symbols   = distinct symbols;

           generate COUNT(uniq_exchanges), COUNT(uniq_symbols);

};

 

另一种用limit返回1一条：日志数据通过hour分组，统计cookie:

(1)   首先对（hour，cookie）按照cookie进行去重；

(2)   按hour分组；

(3)   hour分组进行局部count

(4)   合并各个count，得到最终结果

下面是具体代码示例：

                   showlogs= LOAD '$inpath' USING PigStorage(',');

necessaryfields= FOREACH showlogs GENERATE (chararray)$1 AS fhour,$13 AS fallyesid, $19 ASfip, $20 AS fuseragent , CONCAT((chararray)$19,(chararray)$20) AS fipuseragent;

--1.生成hour+cookie : hourcookie

hourcookie= FOREACH necessaryfields GENERATE $0 AS hour, $1 AS cookie;

 

--2.对hourcookie按照cookie进行去重: dist_hour_cookie

ta= GROUP hourcookie BY cookie;

dist_hour_cookie= FOREACH ta {

       temp = limit hourcookie 1;

       GENERATE FLATTEN(temp);

};

 

--3.对dist_hour_cookie按照hour进行分组:dist_hour_cookie_grp

dist_hour_cookie_grp= GROUP dist_hour_cookie BY $0;

 

--4.对dist_hour_cookie_grp进行count:dist_hour_cookie_grp_count

dist_hour_cookie_grp_count= FOREACH dist_hour_cookie_grp {

       GENERATE group AS hour, COUNT(dist_hour_cookie) AS cookie_count;

};

 

--5.对dist_hour_cookie_grp_count进行sum

sumtemp= GROUP  dist_hour_cookie_grp_count all;

total_cookie= FOREACH sumtemp GENERATE SUM(dist_hour_cookie_grp_count.cookie_count);

dump total_cookie;

 

 

针对foreach内嵌，每次处理一条数据。我做了一个实例测试，证明关系group后，foreach每次处理一个group子集。

数据按guid分组，每个记录都是一个独立的group。
去重统计个数，
在foreach最后输出guid和group，做对比，看到重复数据：
my_data = foreach origin_cleaned_data generate wizad_ad_id,guid,os_version,log_type;
test_data = limit my_data 100;
g_log = group test_data by guid;
uniq = foreach g_log{
        guid = test_data.guid;
        os_v = test_data.os_version;
        uniq_guid = distinct guid;
        generate group,COUNT(uniq_guid),COUNT(os_v),guid;
        };
dump uniq;
describe uniq;


结果
g_log结构: {group: chararray,test_data: {wizad_ad_id: chararray,guid: chararray,os_version: chararray,log_type: chararray}}
输出结果：
(351794060670802,1,1,{(351794060670802)})
(352246063893286,1,1,{(352246063893286)})
(352274018390729,1,1,{(352274018390729)})
(352315053649659,1,1,{(352315053649659)})
......
(354710052256050,1,1,{(354710052256050)})
(355065053261297,1,1,{(355065053261297)})
(861202021584958,1,1,{(861202021584958)})
(861276027634215,1,1,{(861276027634215)})
(861288000290493,1,1,{(861288000290493)})
(861372020081247,1,3,{(861372020081247),(861372020081247),(861372020081247)})
(862011024062881,1,1,{(862011024062881)})
(862040020713619,1,1,{(862040020713619)})
(862055100027987,1,1,{(862055100027987)})
(862106010206458,1,1,{(862106010206458)})
(862191016593489,1,1,{(862191016593489)})
(862283020830914,1,1,{(862283020830914)})
(862324016545965,1,2,{(862324016545965),(862324016545965)})
(862565010397387,1,1,{(862565010397387)})
(862620028211136,1,1,{(862620028211136)})
(862663027090333,1,1,{(862663027090333)})
(862703020059792,1,1,{(862703020059792)})
(862910026533684,1,1,{(862910026533684)})
(862966020482112,1,1,{(862966020482112)})
(863077025294442,1,1,{(863077025294442)})
(863139026459463,1,1,{(863139026459463)})
......
uniq: {group: chararray,long,long,guid: {guid: chararray}}


从中看出：
按guid分组所以前后，group和guid基本一致，但因为guid有重复，所以多了几条；
可如果使用uniq_guid，即generate group,COUNT(uniq_guid),COUNT(os_v),uniq_guid;那么前后将完全一致。


对比，在generate中去掉group和guid，即generate group,COUNT(uniq_guid),COUNT(os_v),guid;
输出结果:
(1,1)
(1,1)
(1,1)
(1,1)
(1,1)
.......


如果按log_type分组，2为click，1为show日志，
只limit 100的话，只有1的日志，
g_log = group test_data by log_type;
uniq = foreach g_log{
        guid = test_data.guid;
        os_v = test_data.os_version; 
        uniq_guid = distinct guid;
        generate COUNT(uniq_guid),COUNT(os_v),guid;
        };

结构为
uniq: {long,long,guid: {guid: chararray}}
(97,100,{(351794060670802),(352246063893286),(352274018390729),(352315053649659),(352316055186377),(352903052496931),(352956061574924),(353721059706006),(354710052256050),(355065053261297),(355310044316523),(355431800349753),(355594050655870),(355868054005229),(356405053894342),(356521051392830),(356524057953100),(356845052409198),(356988053301574),(357070009612856),(357116040052477),(357747051110463),(357784057979674),(358197058147473),(358373048304967),(358585052003587),(358968041526969),(359092058109426),(359357054529855),(359786051477264),(359899046896710),(860173013415348),(860570023370282),(860602020630317),(860813022906346),(860892020832126),(861022007037726),(861060010353755),(861118000103844),(861133029749824),(861202021584958),(861276027634215),(861288000290493),(861372020081247),(861372020081247),(861372020081247),(862011024062881),(862040020713619),(862055100027987),(862106010206458),(862191016593489),(862283020830914),(862324016545965),(862324016545965),(862565010397387),(862620028211136),(862663027090333),(862703020059792),(862910026533684),(862966020482112),(863077025294442),(863139026459463),(863150020224084),(863151028063706),(863177029375831),(863235016133314),(863343023501554),(863427022570130),(863431020260215),(863735010063193),(863777020966957),(863827011094952),(864260024902038),(864264021430787),(864264021548489),(864299028966482),(864301029532170),(864375022648902),(864500025549803),(864505000231848),(864573012942174),(864789028603903),(864958020327031),(864989010679594),(865030012667169),(865316020164285),(865369022512415),(865369029331074),(865407016261415),(866805010343957),(867064010125951),(867064016986521),(867163010209227),(867264010644205),(867739012450792),(868201005627091),(868629010372156),(868880017937570),(869226012126760),(869642009320895)})


 在generate COUNT(uniq_guid),COUNT(os_v),guid;中去掉guid的结果为：
(97,100)
uniq: {long,long}



另外，用sample可以按日志数据的整体分布取样，这样可以看到log_type的所有取值类型数据，

但是每次取样不一样，所以多次执行的统计结果不一样
my_data = foreach origin_cleaned_data generate wizad_ad_id,guid,os_version,log_type;
test_data = sample my_data 0.01;
g_log = group test_data by log_type;
uniq = foreach g_log{
        guid = test_data.guid;
        os_v = test_data.os_version;
        uniq_guid = distinct guid;
        generate group, COUNT(uniq_guid), COUNT(os_v);
        };



g_log结构为: {group: chararray,test_data: {wizad_ad_id: chararray,guid: chararray,os_version: chararray,log_type: chararray}}


生成的数据结果为
(1,13455,14246)
(2,74,74)
uniq: {group: chararray,long,long}


语句 generate中去掉group：generate COUNT(uniq_guid), COUNT(os_v)结果为
(13407,14178)
(73,73)
uniq: {long,long}