像这样的办法也许行得通。但是，我从来没有必要在pigMap中查找变量键，而另一个问题没有答案，因此您需要进行一些尝试和错误操作以使其正常工作：

--Load your dynamic percentages as a map
A = LOAD 'percentages' AS (categ_name:chararray, perc:float);
PERCENTAGES = FOREACH A GENERATE TOMAP(categ_name, perc);

PRODUCT_DATA = LOAD ...;
GRP_PROD_DATA = GROUP PRODUCT_DATA BY categ_name;

--Count the elements per group; needed to calculate pecentages
C = FOREACH GRP_PROD_DATA generate FLATTEN(group) AS categ_name, COUNT(*) as count;
c_MAP = FOREACH C GENERATE TOMAP(categ_name, count);

TOP_PROD_LIST = FOREACH GRP_PROD_DATA {
    SORTED_TOP_PROD = ORDER PRODUCT_DATA BY impression_cnt DESC;
    SAMPLED_DATA = LIMIT SORTED_TOP_PROD (C_MAP#group * PERCENTAGES#group);
    GENERATE flatten(SAMPLED_DATA);
}

你也可以试着用Pig的 TOP 函数而不是 ORDER + LIMIT .

我想我用一种稍微不同的方法解决了这个问题。我不确定它有多优化，也许有更好的方法来组织/优化脚本。基本上，如果我把每个类别的产品名称 ASC 当 RANK <= SAMPLE LIMIT 然后我可以模拟动态采样。这个 SAMPLE LIMIT 只不过是 COUNT 每个类别的标题数量* PERCENT WEIGHT 按类别定义。至 RANK 元组，我利用linkedin的datafu开源jar提供 ENUMERATE 自定义项。
再说一次，如果有人对改进/更好地组织代码有什么建议，我洗耳恭听：）谢谢你的输入cabad，它真的很有帮助！
脚本：

REGISTER '/tmp/udf/datafu-1.0.0.jar';
define Enumerate datafu.pig.bags.Enumerate('1');
set default_parallel 10;

LKP_DATA = LOAD '/tmp/lkp.dat' USING PigStorage('|') AS (categ_name:chararray, perc:float);
PRODUCT_DATA = LOAD '/tmp/meta.dat' USING PigStorage('|') AS (categ_name:chararray, product_titl:chararray, impression_cnt:long, click_through_cnt:long);

GRP_PROD_DATA = GROUP PRODUCT_DATA BY categ_name;

CATEG_COUNT = FOREACH GRP_PROD_DATA generate FLATTEN(group) AS categ_name, COUNT(PRODUCT_DATA) as cnt;

CATEG_JOINED = JOIN CATEG_COUNT BY categ_name, LKP_DATA BY categ_name USING 'replicated';

CATEG_PERCENT = FOREACH CATEG_JOINED GENERATE CATEG_COUNT::categ_name AS categ_name, CATEG_COUNT::cnt AS record_cnt, LKP_DATA::perc AS  percentage;

PRCNT_PROD_DATA = JOIN PRODUCT_DATA BY categ_name, CATEG_PERCENT BY categ_name;

PRCNT_PROD_DATA = FOREACH PRCNT_PROD_DATA GENERATE PRODUCT_DATA::categ_name AS categ_name, PRODUCT_DATA::product_titl AS product_titl, PRODUCT_DATA::impression_cnt AS impression_cnt, PRODUCT_DATA::click_through_cnt AS click_through_cnt,  CATEG_PERCENT::record_cnt*CATEG_PERCENT::percentage AS sample_size;

GRP_PRCNT_PROD_DATA = GROUP PRCNT_PROD_DATA BY categ_name;

ORDRD_PROD_LIST = FOREACH GRP_PRCNT_PROD_DATA {
                             SORTED_TOP_PROD = ORDER PRCNT_PROD_DATA BY impression_cnt DESC;
                             GENERATE flatten(SORTED_TOP_PROD);
                          }

GRP_PROD_LIST = GROUP ORDRD_PROD_LIST BY categ_name;

GRP_PRCNT_PROD_DATA = FOREACH GRP_PROD_LIST GENERATE flatten(Enumerate(ORDRD_PROD_LIST)) AS (categ_name, product_titl, impression_cnt, click_through_cnt,  sample_size, rnk);

SAMPLED_DATA = FILTER GRP_PRCNT_PROD_DATA BY rnk <= sample_size;

SAMPLED_DATA = FOREACH SAMPLED_DATA GENERATE categ_name, product_titl, impression_cnt, click_through_cnt, rnk;

DUMP SAMPLED_DATA;