Re: Relation or attribute and why

x wrote:
> "dawn" <dawnwolthuis_at_gmail.com> wrote in message
> news:1147990242.573989.100810_at_g10g2000cwb.googlegroups.com...
> > x wrote:
> > > "dawn" <dawnwolthuis_at_gmail.com> wrote in message
> > > news:1147826073.046909.174590_at_u72g2000cwu.googlegroups.com...
> > > > Let's say that we have a noun present in a conceptual data model, but
> > > > not identified as a strong entitiy (recognizing not all conceptual
> > > > modeling techniques use such distinctions). What are the conditions
> > > > under which this noun will/should translate into an attribute in a
> > > > logical data model? What are the conditions under which this noun
> > > > will/should translate into a relation in the logical data model?
> > >
> > > It will always translate to an attribute or a set of attributes.
>
> > Hey, thanks x ;-) but are you sure? It might be dropped entirely when
> > going from the CDM to the LDM, although less likely if the LDM includes
> > derived sets and attributes too.
>
> I just told you my opinion. I'm not Codd, you know...
>
> > We could take the noun "name" from our CDM and model it as a derived
> > attribute or user-defined function, perhaps associating it with a
> > derived set so we have something like
>
> > select function('fullName') as name from Person;
> > perhaps placing this in a create view statement so we can then do
> > select name from PersonView;
> >
> > If we have a noun A that has-a B in our CDM, then we often model A as a
> > relation and if B doesn't have-a C, then we might model it as an
> > attribute in set A. In the case of "A:name has-a B:firstName" we often
> > model "name" as a derived attribute, perhaps leaving it out of the LDM
> > altogether so that each application can paste together the parts to
> > form a name.
>
> Sorry, but it's hard for me to follow you above.

In a conceptual data model we might have three nouns: name, firstName, and lastName. name has-a firstName and name has-a lastName. So, we dump "name" from the LDM (or include it as derived if we are modeling derived attributes too). What is the "rule of thumb" that causes us to not make name either a relation or an attribute in our LDM, but to ditch it?

> > Is there a guideline for this that we could state in terms of
> > functional dependencies or otherwise that would lead us to a noun such
> > as "name" becoming a derived attribute rather than a set?
>
> Have you studied abstract algebra ?

I am hard pressed to see how algebra is related to eliminating "name" when going from the CDM to the LDM, I'm all ears if you have an algebraic response [but yes once upon a time, roughly a quarter of a century ago]

> > I'm trying to gain clarity in how one goes from a conceptual data model
> > to a logical data model (when employing the RDM) and why.
>
> When employing RDM one must use relations for all data.
> I encountered several opinions regarding SQL-DBMSs.
> 1) one should be able to work with relations as described by the RDM at the
> level of an SQL-DBMS
> 2) the schema in an SQL-DBMS is just an implementation and one should work
> with RDM in a layer above the DBMS
> 3) SQL is a language that must enable one to use a DBMS regardless of the
> data model of the DBMS
> What a mess.

The prevailing opinion on cdt seems to be that SQL is a useful, yet incomplete and/or inaccurate implementation of the RM.

> >What is the
> > technique (is it a normalization technique, is it based on FDs?) that
> > might eliminate "name" in going from the conceptual to the logical data
> > model?
>
> I don't understand your example (I don't have the patience to decode it
> right now), but I'll say that if you need "name" why do you want to remove
> it ?

Because it is composed to two attributes that are "easier" to implement on their own without dragging the immediate "relation" to which they belong into the picture. Instead of Person(personId, nameId, ...) plus Name(nameId, firstName, lastName) it is more often modeled as Person(personId, firstName, lastName...)

Why? What rule or practice is being employed when we drop a noun like "name" when going from a CDM to an LDM? Thanks. --dawn

> > What rules or rules of thumb might make either A or B a
> > relation or an attribute? For example, Marshall mentioned the
> > technique once known as 1NF in the case where B is multivalued. In
> > that case, B becomes a set instead of an attribute. Are there other
> > instances when B which has no C would become a set instead of an
> > attribute? What rule are we employing when we drop B from the LDM or
> > make it derived (A:date has-a B:month would be an example)?
>
> What operators would you use if the relations are not in 1NF ?
>
> > Has-a is just one type of relationship in a conceptual data model, but
> > maybe it is worth zeroing in on. Thanks. --dawn
>
> Has-a is a relation like any other.
Received on Tue May 23 2006 - 01:30:54 CEST