Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Plain View
Pig, mail # user - Problems with union, projection producing unexpected results
Jonathan Coveney
2011-02-17, 15:41
James Kebinger
2011-02-17, 17:32
James Kebinger
2011-02-17, 19:01
Copy link to this message
-
Problems with union, projection producing unexpected results
James Kebinger 2011-02-16, 23:57
Hello all, I've been scratching my head over a problem with a pig script I'm
having, and hoping another set of eyeballs will help. I'm using pig 0.8, in
local mode

Here's my simplified use case:

I have a log file with events on pages, and the id of the event can be a
users login or a users numeric id:

2010-05-14,james
2010-05-15,123
2010-05-15,23
2010-05-15,456
2010-05-15,notjames

So i want to join a set of users on either the login or user id.

Here's my users:

123,james,11
234,notjames,11
456,someoneelse,11
So I thought I would be clever and load the user list, union it with itself
to generate a relation where each user is represented twice, once by login,
once by id:

logins = FOREACH users GENERATE LOWER(login) as matching_id, user_id as
actual_user_id;
user_ids = FOREACH users GENERATE user_id as matching_id, user_id as
actual_user_id;
user_id_or_login_lookup = UNION logins, user_ids;

user_id_or_login_lookup: {matching_id: chararray,actual_user_id: chararray}
(123,123)
(234,234)
(456,456)
(james,123)
(notjames,234)
(someoneelse,456)

Then join on that, by the first column, and project that away, leaving just
the event info and the numeric id.

views_with_id = JOIN profile_views by viewed_user_id,
user_id_or_login_lookup by matching_id;

That is not working however. My joined relation looks like this (which is
what I expect)

views_with_id: {profile_views::date:
chararray,profile_views::viewed_user_id:
chararray,user_id_or_login_lookup::matching_id:
chararray,user_id_or_login_lookup::actual_user_id: chararray}

(2010-05-15,123,123,123)
(2010-05-15,456,456,456)
(2010-05-14,james,james,123)
(2010-05-15,notjames,notjames,234)

But when I project as follows: views_with_id_projected = FOREACH
views_with_id GENERATE date, viewed_user_id,
user_id_or_login_lookup::actual_user_id;

The result is not what I expect

(2010-05-15,123,123)
(2010-05-15,456,456)
(2010-05-14,james,james)
(2010-05-15,notjames,notjames)

To be clear, I expect

(2010-05-15,123,123)
(2010-05-15,456,456)
(2010-05-14,james,123)
(2010-05-15,notjames,456)
Can anyone give me a push in the right direction?

Thanks, James

Here's my full pig script:

users = LOAD 'patients-test.txt' USING PigStorage(',') AS
(user_id:chararray, login:chararray, disease_id: chararray);
profile_views = LOAD 'patient-views-test.txt' USING PigStorage(',') AS(date:
chararray, viewed_user_id:chararray);

dump users;
dump profile_views;

-- build a relation so that users are present to join by login or user_id
logins = FOREACH users GENERATE LOWER(login) as matching_id, user_id as
actual_user_id;
user_ids = FOREACH users GENERATE user_id as matching_id, user_id as
actual_user_id;
user_id_or_login_lookup = UNION logins, user_ids;

dump user_id_or_login_lookup;
describe user_id_or_login_lookup;

views_with_id = JOIN profile_views by viewed_user_id,
user_id_or_login_lookup by matching_id;

describe views_with_id;
--STORE views_with_id into 'ep-views.txt';

dump views_with_id;

views_with_id_projected = FOREACH views_with_id GENERATE date,
viewed_user_id, user_id_or_login_lookup::actual_user_id;
dump views_with_id_projected;
+
Jonathan Coveney 2011-02-17, 15:41
+
James Kebinger 2011-02-17, 17:32
+
James Kebinger 2011-02-17, 19:01