虽然你可以在纯Pig身上做这件事，但用自定义项做这件事应该效率更高。大致如下：

@outputschema('wordcounts: {T:(word:chararray, count:int)}')
def generate_wordcount(BAG):
    d = {}
    for word in BAG:
        if word in d:
            d[word] += 1
        else:
            d[word] = 1
    return d.items()

您可以这样使用这个自定义项：

REGISTER 'myudfs.py' USING jython AS myudfs ;

-- A: (id, words: {T:(word:chararray)})

B = FOREACH A GENERATE id, FLATTEN(myudfs.generate_wordcount(words)) ;

试试这个：

$ cat input 
foobar  foo
foobar  foo
foobar  foobar
foobar  bar
foo bar
foo bar

--preparing
inputs = LOAD 'input' AS (first: chararray, second: chararray);
grouped = GROUP inputs BY first;
formatted = FOREACH grouped GENERATE group, inputs.second AS second;
--what you need
flattened = FOREACH formatted GENERATE group, FLATTEN(second);
result = FOREACH (GROUP flattened BY (group, second)) GENERATE FLATTEN(group), COUNT(flattened);
DUMP result;

输出：

(foo,bar,2)
(foobar,bar,1)
(foobar,foo,2)
(foobar,foobar,1)