Re: Modeling question...

On Oct 23, 3:15 pm, David BL <davi..._at_iinet.net.au> wrote:
> On Oct 23, 9:05 pm, JOG <j..._at_cs.nott.ac.uk> wrote:
>
>
>
> > On Oct 23, 2:01 pm, Roy Hann <specia..._at_processed.almost.meat> wrote:
>
> > > JOG wrote:
> > > > Despite a growing literature, current definitions of "semi-structure"
> > > > are woefully inadequate.
>
> > > A million people can (and evidently will) talk bollocks, but it's still
> > > bollocks.
>
> > > > The standard denotation is of data that "does
> > > > not fit into the relational model".
>
> > > That definition is entirely bogus.  The relational model just applies
> > > set theory to first order predicate logic.  If you have "data" that
> > > doesn't fit into both of these then you better start hiring mystics to
> > > look after it for you.
>
> > Indeed. And yet hundreds of peer-reviewed papers have been published
> > on the topic. I find this incredibly depressing.
>
> Ok, I’ll bite…
>
> No doubt any data can be made to “fit” into the relational model.

Let me state first that I don't believe that the relational model is universally applicable (I'm not sure where you think I have stated that). However, all data can be stated in predicate logic, and all statements of logic can be modelled in the RM. Hence, i consider it absolutely unarguable that there is any data which cannot be structured as a schema of relations. This is my objection to the semistructure literature.

> The more important question is whether it happens /naturally/.
> The relational model works really well when there is a UoD on which many
> propositions can be made without needing to introduce lots of abstract
> identifiers.

The RM handles facts as naturally as stating them in predicate logic. And why would one ever model things other than facts in predicate logic?
I think there is confusion (in general, not simply here!) about what a database is intended to model. It models data as it has been stated in the real world, not the things which that data refers to.

> That’s very common, but it’s not always the case. It
> seems to me the question of whether the RM is generally appropriate
> for heavily nested composite values is unresolved.   Much of the
> world’s data is in this latter form.  Eg abstract syntax trees, rich
> text documents, scene graphs.
>
> If the relational model is universally applicable, why don’t
> programmers enter their programs as relations?   Do you really think
> it’s only because of the tools currently available?

Nope. I think data that requires a high number of predicates compared to the number of statements it models can be cumbersome to manipulate in the RM. Equally I think it misses a trick when it comes to facts that might be represented using logical quantification. However, I do believe this situation can be improved (specifically via greater flexibility in defining predicates and integration of existential quantifiers) and its general declarative principles will be increasingly incorporated into programming languages.

Regards, Jim.

>
> What about automated proof systems?  Is the knowledge base and data
> associated with an ongoing proof best represented using a set of
> relations?  I find that quite unlikely. The RA seems to have more to
> do with set based calculations on known sets of values, rather than
> symbolic manipulation.   Symbolic manipulation involves a lot of
> recursion and the RA on its own is too weak, which suggests it will
> take a backseat role.  Eg to compute the most general unifier of two
> given expressions involves recursion in the nested expressions.
>
> I also find it rather telling that relational queries (ie RA
> expressions) are not themselves represented using relations.  Surely
> if that were useful, many cdt folks would jump at the opportunity to
> further promote the use of relations.
Received on Wed Oct 29 2008 - 01:13:04 CET