Pig 0.11.1 - Count groups in a time range

Question

I have a dataset, A, that has timestamp, visitor, URL:

(2012-07-21T14:00:00.000Z, joe, hxxp:///www.aaa.com) 
(2012-07-21T14:01:00.000Z, mary, hxxp://www.bbb.com) 
(2012-07-21T14:02:00.000Z, joe, hxxp:///www.aaa.com) 

I want to measure number of visits per user per URL in a time window of say, 10 minutes, but as a rolling window that increments by the minute. Output would be:

(2012-07-21T14:00 to 2012-07-21T14:10, joe, hxxp://www.aaa.com, 2)
(2012-07-21T14:01 to 2012-07-21T14:11, joe, hxxp://www.aaa.com, 1)

To make the arithmetic easy, I change the timestamp to minute of the day, as:

(840, joe, hxxp://www.aaa.com) /* 840 = 14:00 hrs x 60 + 00 mins) */

To iterate over 'A' by a moving time window, I create a dataset B of minutes in the day:

(0)
(1)
(2)
.
.
.
.
(1440)

Ideally, I want to do something like:

A = load 'dataset1' AS (ts, visitor, uri)
B = load 'dataset2' as (minute)

foreach B {
C = filter A by ts > minute AND ts < minute + 10;
D = GROUP C BY (visitor, uri);
foreach D GENERATE group, count(C) as mycnt;
}

DUMP B;

I know "GROUP" isn't allowed inside a "FOREACH" loop but is there a workaround to achieve the same result?

Thanks!

Answer 1

Maybe you can do something like this?

NOTE: This is dependent on the minutes you create for the logs being integers. If they are not then you can round to the nearest minute.

myudf.py

#!/usr/bin/python

@outputSchema('expanded: {(num:int)}')
def expand(start, end):
        return [ (x) for x in range(start, end) ]

myscript.pig

register 'myudf.py' using jython as myudf ;

-- A1 is the minutes. Schema:
-- A1: {minute: int}
-- A2 is the logs. Schema:
-- A2: {minute: int,name: chararray}
-- These schemas should change to fit your needs.

B = FOREACH A1 GENERATE minute, 
                        FLATTEN(myudf.expand(minute, minute+10)) AS matchto ;
-- B is in the form:
-- 1 1
-- 1 2
-- ....
-- 2 2
-- 2 3
-- ....
-- 100 100
-- 100 101
-- etc.

-- Now we join on the minute in the second column of B with the 
-- minute in the log, then it is just grouping by the minute in
-- the first column and name and counting
C = JOIN B BY matchto, A2 BY minute ;
D = FOREACH (GROUP C BY (B::minute, name)) 
            GENERATE FLATTEN(group), COUNT(C) as count ;

I'm a little worried about speed for larger logs, but it should work. Let me know if you need me to explain anything.