Haha, when I say naive I don't mean bad... plenty of my scripts use that
approach, and often it's unavoidable, so it's good to understand.

as far as the naming issue, when you flatten it is usually a good idea to
give the resultant columns a name. so your example would become:


A = load 'one.txt' as (a:int, b:int);
B = load 'two.txt' as (a:int, b:int);
A_1 = foreach A generate flatten(TOTUPLE(a,b)) as (a,b);
B_1 = foreach B generate flatten(TOTUPLE(a,b)) as (x,y);
C = join A_1 by a full, B_1 by x;
describe C

That will get rid of the org.apache.pig.builtin.totuple_b etc. But let's
say that you still want them to have the same name, you can do that:


A = load 'one.txt' as (a:int, b:int);
B = load 'two.txt' as (a:int, b:int);
A_1 = foreach A generate flatten(TOTUPLE(a,b)) as (a,b);
B_1 = foreach B generate flatten(TOTUPLE(a,b)) as (a,b);
C = join A_1 by a full, B_1 by a;
describe C

And in the join result, you can disambiguate A_1::a and B_1::a, and so on.

2012/4/19 Rajgopal Vaithiyanathan <[email protected]>

> I knew it would sound naive :P I didn't even know a schema parser exists.!
>
> `it can only return a tuple, which you then flatten into columns.`
>
>
> Isn't this bad..?   For example see this, ( for simplicity i'm using
> TOTUPLE instead of my UDF.. )
>
> A = load 'one.txt' as (a:int, b:int);
> B = load 'two.txt' as (a:int, b:int);
> A_1 = foreach A generate flatten(TOTUPLE(a,b));
> B_1 = foreach B generate flatten(TOTUPLE(a,b));
> C = join A_1 by a full, B_1 by a;
> describe C
>
> The schema description is like this.
>
> C: {A_1::org.apache.pig.builtin.totuple_b_18::a:
> int,A_1::org.apache.pig.builtin.totuple_b_18::b:
> int,B_1::org.apache.pig.builtin.totuple_b_19::a:
> int,B_1::org.apache.pig.builtin.totuple_b_19::b: int}
>
> and totuple_b_** in the description obviously changes every time  i
> describe because it is based on a counter....
> Now how do i disambiguate between The A_1's a,b and B_1's a,b ?
>
>
> On Thu, Apr 19, 2012 at 12:07 PM, Jonathan Coveney <[email protected]
> >wrote:
>
> > Dmitriy's suggestion is spot on, but just to be pedantic, you'd do:
> >
> > public Schema outputSchema(Schema input) {
> >   List<FieldSchema> list = new ArrayList<FieldSchema>();
> >   list.add(new FieldSchema("one", DataType.CHARARRAY));
> >   list.add(new FieldSchema("two", DataType.CHARARRAY))
> >
> >    return new Schema(new Schema.FieldSchema("t", new Schema(list),
> > DataType.TUPLE));
> > }
> >
> > That said, in your question you asked: "how can you get it without the
> > parenthesis." Short answer is that you can't. A UDF can't return multiple
> > columns -- it can only return a tuple, which you then flatten into
> columns.
> >
> > 2012/4/18 Dmitriy Ryaboy <[email protected]>
> >
> > > It's messy. Easier to use the schema parser:
> > >
> > >
> >
> org.apache.pig.impl.util.Utils.getSchemaFromString("t:tuple(len:int,word:chararray)");
> > >
> > > Even easier to use the @OutputSchema annotation (coming in 0.11 I
> > believe)
> > >
> > > -D
> > >
> > >
> > > On Wed, Apr 18, 2012 at 7:02 PM, Rajgopal Vaithiyanathan
> > > <[email protected]> wrote:
> > > > Hey all,
> > > >
> > > > Sorry if i  sound naive, but how should one implement outputSchema of
> >  an
> > > > eval Func that returns tuple.
> > > > The way i do it is ,
> > > >
> > > > public Schema outputSchema(Schema input) {
> > > >    List<FieldSchema> list = new ArrayList<FieldSchema>();
> > > >    list.add(new FieldSchema("one", DataType.CHARARRAY));
> > > >    list.add(new FieldSchema("two", DataType.CHARARRAY))
> > > >
> > > >    return new Schema(list);
> > > > }
> > > >
> > > > but in the front end, If i use
> > > >  B = foreach A generate flatten(FUNC());
> > > >  describe B
> > > > I get the schema like this:
> > > >    { ( one:chararray, two:chararray ) }
> > > > Now i use a flatten on this like :
> > > >    B = foreach A generate flatten(FUNC());
> > > >  and i get { null::one : chararray, null::two : chararray }
> > > >
> > > > The question is,
> > > > How should i implement the outputSchema so that i get the schema
> like {
> > > one
> > > > : chararray, two : chararray }  // NOTE: without the parenthesis
> > >
> >
>
>
> Raj :)
>

Reply via email to