You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@pig.apache.org by Tobias Schulze-Heinrichs <t....@xplosion.de> on 2013/12/10 18:00:35 UTC

Re: Count empty relation after filtering

What might be a proper workaround: 


id, valid
(1, false)
(2, false)

records = LOAD 'test.csv' USING PigStorage(',') AS (id:long,valid:boolean);
total = GROUP records ALL;

test = FILTER records BY valid == true;
test_count = FOREACH (GROUP test ALL) GENERATE COUNT(test) as count;
test_count_normalized = FOREACH total GENERATE ( (test_count.count is null) ? 0 : test_count.count) ) as count;
The new total relation is just something which I abuse to iterate and apply my 'is null' check.

At least for my problem - which is pretty much the same as yours it works quite fine ;)

Cheers Tobi

> Great analysis. Couldn't agree more.


2013/5/29 mehmet <me...@yahoo.com>

> I tried your code on 0.10 and it gives the same result. I can logically
> explain why it gives you this result, although I am not convinced that
> would be the desired outcome.
>
> If you think of it as there is a synthetic implicit key 'all' in each
> tuple, and you are grouping over that, you can see why there is no output:
> no tuples, nothing to group over (no reducer sees the key 'all', because
> it doesn't exist). Although, I would contend that when there are no
> tuples, it is might be ideal to output (all,{}) as the output of the group
> all.
>
>
> ________________________________
>  From: Marco Brinkmann <ma...@cope.io>
> To: user@pig.apache.org
> Sent: Wednesday, May 29, 2013 8:43 AM
> Subject: Re: Count empty relation after filtering
>
>
> I tried to explain why in my basic understanding an operation in a foreach
> (count, count_star or anything else) will not leed to any success. And I
> still appreciate any hints or tricks to achieve the above.
>
>
> 2013/5/29 Shahab Yunus <sh...@gmail.com>
>
> > So basically this means that we were trying to look at this from RDBMS'
> SQL
> > perspective where 'SELECT COUNT(*) FROM TABLE' returns 0 even if there is
> > nothing in the result set and that is why we ignored the possibility that
> > FOREACH might not being executed at all (which could be by design)?
> >
> > -Shahab
> >
> >
> > On Wed, May 29, 2013 at 10:13 AM, Marco Brinkmann
> > <ma...@cope.io>wrote:
> >
> > > Thanks, but this does not change anything. My personal guess (and I
> only
> > > work for a few days with pig) is that FOREACH will never be executed,
> > > because the relation 'test' is empty.
> > >
> > >
> > > 2013/5/29 Shahab Yunus <sh...@gmail.com>
> > >
> > > > Try COUNT_STAR.
> > > >
> > > > -Shahab
> > > >
> > > >
> > > > On Wed, May 29, 2013 at 9:55 AM, Marco Brinkmann <
> > > marco.brinkmann@cope.io
> > > > >wrote:
> > > >
> > > > > Hi everybody,
> > > > >
> > > > > I have a rather simple question and scenario, but still I could not
> > > find
> > > > an
> > > > > answer in the documention or in other resource:
> > > > >
> > > > > id, valid
> > > > > (1, false)
> > > > > (2, false)
> > > > >
> > > > > records = LOAD 'test.csv' USING PigStorage(',') AS (id:long,
> > > > > valid:boolean);
> > > > >
> > > > > test = FILTER records BY valid == true;
> > > > > test_count = FOREACH (GROUP test ALL) GENERATE COUNT(test);
> > > > >
> > > > > DUMP test_count;
> > > > >
> > > > >
> > > > > I would expect that 'valid_count' nows contains '0'. But the dump
> is
> > > > > completely empty (with 'valid == false' I get '(2)' as expected).
I
> > use
> > > > pig
> > > > > 0.11.1.
> > > > >
> > > > > Could someone point me in the right direction?
> > > > >
> > > > > Cheers, Marco
> > > > >
> > > >
> > >
> >
>