Hello

I'm trying to find a SUM of a range of fields, and am having difficulty.

I have the following data structure (from the movielens public dataset)
where there's a "fixed" field of "Name" and there's a denormalized "genres" list
(for example, the first column is "action", second is "comedy", etc.

Name|Genres
Toy Story|0|0|0|1|1|1|0|0|0|0|0|0|0|0|0|0|0|0|0
GoldenEye|0|1|1|0|0|0|0|0|0|0|0|0|0|0|0|0|1|0|0

This seems like an ideal use for the project-range feature of Pig,
where it would be trivial to find movies that belonged to two or more genres.

I'm trying to use this code:
movies = load 'movies' USING PigStorage('|');
movie_and_genres = FOREACH movies GENERATE $0, TOBAG($2 ..) AS genres;
DUMP movie_and_genres;
This works, and gives me:
(Toy Story,{(0),(1),(1),(1),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0)})
(GoldenEye,{(1),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(1),(0),(0)})

However, if I try to run a SUM on the genres bag, I receive the error:

"Could not infer the matching function for org.apache.pig.builtin.SUM
as multiple or none of them fit. Please use an explicit cast."

I've tried to flatten and cast the genres bag like this:

movies = load 'movies' USING PigStorage('|');
movie_and_genres = FOREACH movies GENERATE $0, FLATTEN(TOBAG($2 ..));
movie_and_int_genres = FOREACH movie_and_genres GENERATE $0, (int) $1;

However, then I receive the error:
Cannot cast bytearray to int

Any ideas what to try next?  Or, would I be better off trying to use a STREAM
or custom loader to do something like this?

Thanks,
--Nate

Try:

movie_and_genres = FOREACH movies GENERATE $0,
(b:bag{t:tuple(i:int)})TOBAG($2 ..) AS genres:b:bag{t:tuple(i:int)};
foo = foreach movies_and_genres generate SUM(genres) as genre_total;
Russell Jurney http://datasyndrome.com

On Mar 10, 2013, at 7:45 AM, Nathan Neff <[EMAIL PROTECTED]> wrote:

> movie_and_genres = FOREACH movies GENERATE $0, TOBAG($2 ..) AS genres;

On Sun, Mar 10, 2013 at 11:15 PM, Russell Jurney
<[EMAIL PROTECTED]> wrote:
> Try:
>
> movie_and_genres = FOREACH movies GENERATE $0,
> (b:bag{t:tuple(i:int)})TOBAG($2 ..) AS genres:b:bag{t:tuple(i:int)};
> foo = foreach movies_and_genres generate SUM(genres) as genre_total;

Hi Russell

Thanks for your help, but I couldn't get this variation to work -- I get

Syntax error, unexpected symbol at or near 'b'

I'm using Pig 0.10

>
>
> Russell Jurney http://datasyndrome.com
>
> On Mar 10, 2013, at 7:45 AM, Nathan Neff <[EMAIL PROTECTED]> wrote:
>
>> movie_and_genres = FOREACH movies GENERATE $0, TOBAG($2 ..) AS genres;

It seems like I'm getting closer:

With this data:

Toy Story|0
GoldenEye|0|1|0|1

And this script:

movies = load 'movies' USING PigStorage('|');
movie_and_genres = FOREACH movies GENERATE $0, TOTUPLE($1 ..);
DUMP movie_and_genres;
describe movie_and_genres;

I get this output:

(Toy Story,(0))
(GoldenEye,(0,1,0,1))
movie_and_genres: {bytearray,()}

Russell's code works with a little modification. (The cast to int doesn't
work though.)

movie_and_genres = FOREACH movies GENERATE $0 as movie_name,
(bag{tuple()})TOBAG($2 ..) AS genres: bag{genre_bit: tuple()};
foo = foreach movies_and_genres generate movie_name, (int)SUM(genres) as
genre_total;
Having said that, it appears from your problem description that there're a
fixed number of genres and every movie record would contain either a 0 or 1
corresponding to that genre. Ergo, every record has the same number of
columns. (Is that right? I see your second example doesn't follow this
though.) Then you could specify the detailed schema in your load statement
and simplify matters.

Secondly, it appears that order matters in your genre bitmap (you say the
first column corresponds to action movies and so on). Bags are unordered,
so it makes sense to make a tuple out of your genre bitmap first because
the TOBAG operation will throw away all column order information.
You need to FLATTEN your tuple before TOBAG-ging and SUM-ming it though.

HTH,
Abhinav
On 19 March 2013 07:20, Nathan Neff <[EMAIL PROTECTED]> wrote:

> It seems like I'm getting closer:
>
> With this data:
>
> Toy Story|0
> GoldenEye|0|1|0|1
>
> And this script:
>
> movies = load 'movies' USING PigStorage('|');
> movie_and_genres = FOREACH movies GENERATE $0, TOTUPLE($1 ..);
> DUMP movie_and_genres;
> describe movie_and_genres;
>
> I get this output:
>
> (Toy Story,(0))
> (GoldenEye,(0,1,0,1))
> movie_and_genres: {bytearray,()}
>

It works!!!!

This confirms that Pig is better than Java MapReduce :-)

Thanks everyone for their help.

Input:
Toy Story|0|0|0|0|1|1|0|0|0
GoldenEye|0|1|1|0|0|0|0|0|0|0|0|0|0|0|0|0|1|0|0
SomeNewMovie|0|0|0|0|1|1|0|0|0|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1|1

Script:
movies = load 'movies' USING PigStorage('|');
movie_and_genres = FOREACH movies GENERATE $0 as movie_name,
        (bag{tuple()})TOBAG($1 ..) AS genres: bag{genre_bit: tuple()};
movies_sum_genres = foreach movie_and_genres generate movie_name,
(int)SUM(genres) as genre_total;
DUMP movies_sum_genres;

Output:
(Toy Story,2)
(GoldenEye,3)
(SomeNewMovie,41)

On Tue, Mar 19, 2013 at 12:43 PM, Abhinav Neelam
<[EMAIL PROTECTED]> wrote:
> Russell's code works with a little modification. (The cast to int doesn't
> work though.)
>
> movie_and_genres = FOREACH movies GENERATE $0 as movie_name,
> (bag{tuple()})TOBAG($2 ..) AS genres: bag{genre_bit: tuple()};
> foo = foreach movies_and_genres generate movie_name, (int)SUM(genres) as
> genre_total;
>
>
> Having said that, it appears from your problem description that there're a
> fixed number of genres and every movie record would contain either a 0 or 1
> corresponding to that genre. Ergo, every record has the same number of
> columns. (Is that right? I see your second example doesn't follow this
> though.) Then you could specify the detailed schema in your load statement
> and simplify matters.

The second example is the main reason for using the range, in that new genres
could be added arbitrarily at the end of each record.  Thus movie#1 could have
been added when there were 10 known genres in the format, but
movie#1000 has 11 known genres
in the record format.

> Secondly, it appears that order matters in your genre bitmap (you say the
> first column corresponds to action movies and so on).

that's correct -- for all movies, the first zero is whether or not it belongs
to an 'action' genre.

>Bags are unordered,
> so it makes sense to make a tuple out of your genre bitmap first because
> the TOBAG operation will throw away all column order information.
> You need to FLATTEN your tuple before TOBAG-ging and SUM-ming it though.

Hmm in my use-case (find the movies that belong to 2 or more genres)
this wouldn't matter, but
that's a very interesting (and tricky) point to note.  Thank you very much!

> HTH,
> Abhinav
>
>
>
>
> On 19 March 2013 07:20, Nathan Neff <[EMAIL PROTECTED]> wrote:
>
>> It seems like I'm getting closer:
>>
>> With this data:
>>
>> Toy Story|0
>> GoldenEye|0|1|0|1
>>
>> And this script:
>>
>> movies = load 'movies' USING PigStorage('|');
>> movie_and_genres = FOREACH movies GENERATE $0, TOTUPLE($1 ..);
>> DUMP movie_and_genres;
>> describe movie_and_genres;
>>
>> I get this output:
>>
>> (Toy Story,(0))
>> (GoldenEye,(0,1,0,1))
>> movie_and_genres: {bytearray,()}
>>