[developers] Chart pruning: update

[Dan Flickinger]

Hi Stephan -

While experimenting with the added chart-pruning facility in the main cheap branch, I remembered that I had never sent you the cpu definition I'm using with pruning for the ERG.  Since I discovered by trial and error that it's important to include the "-cm" option (for reasons I do not fully understand), I thought it worthwhile to send you this definition, even so belatedly, which works pretty well for me using the current head revision of the LOGON universe and the trunk version of the grammar in $LOGONROOT/lingo/terg:

     (make-cpu 
      :host (short-site-name)
      :spawn cheap
      :options (list "-n" "-tsdb" "-yy" "-packing"
                     "-cm" "-default-les=all"
                     "-memlimit=1024" "-timeout=60"
		     "-cp=p400"
                     (registry:lookup :terg "~a~a~a" :ln :rt :cp))
      :preprocessor '("lkb::repp-for-pet"
                      :repp :tokenizer :calls (:xml :ascii :quotes))
      :tagger (list :tnt tnt at wsj :n 2) :reader "tsdb::yy-read-input"
      :class :terg-cp :grammar erg :name "pet"
      :task '(:parse)  :flags '(:generics t)
      :wait wait :quantum quantum)

Here the interesting options are "-n" to invoke the chart-pruning facility in cheap, and "-cp=p400" to restrict the pruning to passive edges with a chart cell limit of 400.  I would be glad to see this definition added to $LOGONROOT/dot.tsdbrc and I'd be glad to see what other grammar developers are using.

Best,

  Dan

----- Original Message -----
From: "Stephan Oepen" <oe at ifi.uio.no>
To: "Bart Cramer" <bart.cramer at gmail.com>
Cc: developers at delph-in.net, "Montserrat Marimon" <montserrat.marimon at ub.edu>, "Antonio Branco" <Antonio.Branco at di.fc.ul.pt>, "johan benum" <johan.benum at gmail.com>
Sent: Wednesday, July 14, 2010 11:07:05 AM
Subject: Re: [developers] Chart pruning: update

hei,

the ‘ncheap’ binary (invoked as ‘cheap -n’) in the LOGON tree  
is now updated; if the maintainers of grammars in the LOGON tree email  
me their preferred cpu definition for chart pruning, i can include  
these in the global ‘dot.tsdbrc’.

cheers, oe


On 13. juli 2010, at 14.03, Bart Cramer <bart.cramer at gmail.com> wrote:

> Dear devs,
>
> a little update on chart pruning, as since today there is a small  
> change in the way chart pruning works. The ncheap binary will be  
> updated by Stephan in due time as well, for inclusion in the Paris  
> snapshot of the LOGON tree.
>
> There are now four ways of calling the chart pruning mechanism:
> -cp=a1000 (all tasks, cell size = 1000)
> -cp=s1000 (successful tasks, cell size = 1000)
> -cp=p1000 (passive tasks, cell size = 1000)
> -cp=1000 (defaults to passive)
> The -local-cap and -count-tasks are now deprecated, so you will need  
> to update your cpu profiles in dot.tsdbrc. You always have to  
> specify the limit; no default is specified for that aspect.
>
> Furthermore, I am very much interested in the results you have  
> reached so far porting the technique to your own grammars. Help can  
> be provided whenever needed!
>
> Best,
>
> Bart.