[developers] [pet] unpacking

[Sergio Roa]

Hi Yi,

On Thursday 12 July 2007, Yi Zhang wrote:
> Which  version of  the ERG  are you  using? And  by ``don't  use the
> -packing option'', do you mean setting -packing to 7? or 0? or leave
> it as default?  Do you sill have -nsolutions=10 when you ``don't use
> the -packing''. If  -nsolutions is set to be  non-zero, the -packing
> is defaulted to be 15.

I mean leaving it as default, so, the options now are:

  cheap -limit=100000 -tok=fsr -nsolutions=10

I also got confused, because  in the options help documentation stands
that  "packing=15" is  the default,  but observing  this  behaviour it
doesn't seem to be the case.

Thanks,

Sergio.

> I just experimented with the ERG nov-06. It seems that without packing, the
> parser hits the memory limit very soon and aborts. Similar situation is
> observed for packing with exhaustive unpacking (-packing=7 and/or
> -nsolutions=0). With selective unpacking (-packing=15 -nsolutions=10), the
> parser seems to be struggling to unpack within the memory limit. However,
> due to the huge search space, it seems to take forever.
> I'm really not sure this is actually a bug in the selective unpacking code.
> But in case the above configuration is different from what you have meant,
> do correct me :-)
>
> To PET developers, maybe we need some extra mechanism to control the
> behavior of the parser, say a time-out option?
>
> Best,
> Yi
>
> On 7/11/07, Sergio Roa <sergior at coli.uni-sb.de> wrote:
> > Hello,
> >
> > I am  using the  svn repository version  of cheap, with  the following
> > options:
> >
> > cheap -limit=100000 -tok=fsr -packing=15 -nsolutions=10
> >
> > When I try to parse the ambiguous sentence:
> >
> > "It 's hard  to explain to a 17-year-old why someone  they like had to
> > go," says Mrs. Ward.
> >
> > the cheap  process takes  many hours in  the process of  unpacking.
> >
> > I wonder why this happens. When  I don't use the -packing option I get
> > the results  in short  time.
> >
> > Thanks,
> >
> > Sergio.



-- 
Sergio Roa Ovalle
http://www.informatik.uni-freiburg.de/~roa