Fwd: ACTION item(s): definition(s) and an algorithm for postprocessing minimization

Sigh. Got the reply header wrong.

Cheers,
Bijan.

Begin forwarded message:

> From: Bijan Parsia <bparsia@cs.man.ac.uk>
> Date: August 25, 2006 12:37:18 PM BDT
> To: Pat Hayes <phayes@ihmc.us>
> Subject: Re: ACTION item(s): definition(s) and an algorithm for  
> postprocessing minimization
>
> On Aug 24, 2006, at 5:52 PM, Pat Hayes wrote:
>
>>> On Aug 21, 2006, at 10:53 AM, Seaborne, Andy wrote:
>>>
>>>> Bijan Parsia wrote:
> [snip]
>>>>> So, first qualification: These algorithms are only minimizing  
>>>>> with respect to BNodes. You have to plug in your own account of  
>>>>> literals.  I use "row" for "answer" avoid confusion:
>>>>> DEFINITION 1: Answer graph template
>>>>> 	Let A be an answer set and Avar be the set of column headings  
>>>>> of A.
>>>>> 	The answer graph template of A is the set of triple patterns,  
>>>>> such  that:
>>>>> 		{tp | _:row ('http://var.org/#" ++ var) var. & var \in Avar}
>>>>                                                         ^^^
>>>> value?
>>>
>>> I just needed a random URI prefix. Substitute any you like.
>>
>> We did once consider a proposal to have a special 'anonymous'  
>> prefix, basically an IRI space solely for skolem names, as part of  
>> the spec.
>
> This isn't the same. The prefix is used to turn a query variable  
> name into a property. There are no *values* in the picture yet. By  
> "random" I just mean I needed *any* one in order to form the  
> template and then the analogous graph. If you skolemed the bnodes  
> in this way, the resulting graph would *not* be lean (I believe).
>
>> That idea kind of died, I think (?) because it was felt to be  
>> artificial and not in the IRI spirit of rock-solid eternal  
>> universal identifiers that retain their meaning throughout the  
>> known universe,
>
> Ah, it died of foolish prejudice.
>
>> but it might be worth reconsidering it. There are emails about  
>> this in the archive somewhere but I confess I can't actually find  
>> them now.
>>
>>>> (Just to show I'm reading this all!)
>>>
>>> Heh.
>>>
>>> Oh, if we had (do we have?) CONSTRUCT DISTINCT, I would  
>>> personally expect a lean graph as output.
>>
>> I agree, if we have this, but I hope we don't :-)
>
> Well, I was trying to pump the intuition that DISTINCT was  
> connected to leanness of output, not of input. I see I've succeeded.
>
> Cheers,
> Bijan.

Received on Friday, 25 August 2006 16:57:54 UTC