[pet] PET: managing resource limits

[Stephan Oepen]

thanks for the quick reply, yi!

> i find the reasoning behind this to be appealing. one potential criticism is that we do NOT have any proper ranking mechanism in the (default, exhaustive packed) forest creation phase. that didn't matter much since we assumed it to be exhaustive.  but when the timeout occurs in (6), there is no guarantee on the quality of the analysis in the partial forest.  selective unpacking will still get the best unpacked reading out of the forest though. it would be great if we prioritize the agenda in forest creation according to, say a PCFG (something similar to what Bart did in his thesis).  anyway, this is probably digressing ...

hmm, yes, in the sense that the question of how to best structure phase (6)
internally can be separated from the strategy for resource allotment.  but i do
see your point.  on the other hand, the current agenda strategy for creation
of the parse forest was optimized to reduce the number of (costly) retroactive
packings.  so different use cases might call for different configurations here.

> as to the fixed allocation of time between (6) and (7), it is a practical question of, i) whether the same fraction works for different grammars, ii) whether the same fraction works for different length of timeout (say strict 5s timeout in real-time parsing scenario)?

ah, sorry, i had maybe oversimplified for exposition: i was assuming the
sub-allotment ratio would have to be configurable, e.g. minimally through a
command line option or setting that specified a percentage of resources; in
principle, one could imagine different percentages for different types of limits
(10% of the time but only 5% of passive edges); not sure how complex we
need to make things in the beginning, though?

>> in terms of next steps, maybe we could try a top-down development model, for
>> once :-).  i.e. write up what we would like to have (somewhere on the wiki), and
>> then see whether we can adapt the code (on top of the cleaned-up NG branch,
>> i would suggest).
>>
> wow, that is unconventional. but i'm in :-)

:-), oe

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+++ Universitetet i Oslo (IFI); Boks 1080 Blindern; 0316 Oslo; (+47) 2284 0125
+++    --- oe at ifi.uio.no; stephan at oepen.net; http://www.emmtee.net/oe/ ---
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++