Re: ISSUE-3

Souri,

I couldn't agree more! :)

As we presented today, Ultrawrap does the Mapper in compile time and Querier
in run time.

We are on the same page

Juan Sequeda, Ph.D Student
Dept. of Computer Sciences
The University of Texas at Austin
www.juansequeda.com
www.semanticwebaustin.org


On Tue, Nov 24, 2009 at 11:55 AM, Souri Das <Souripriya.Das@oracle.com>wrote:

>  Some ramblings from me just to make sure we are on the same page as far as
> understanding of "DDL" mapping goes and also to suggest an alternate term
> for "DDL" for the group's consideration:
>
> Let us assume that there are two roles: 1) the Mapper (specifies the
> mapping from RDB to RDF) and 2) the Querier (designs the SPARQL query
> against the RDF view of RDB data).
>
> To directly work with data in an already-created database, a Mapper has to
> first look at the data dictionary tables to find out the list of definitions
> for types, tables/views (including columns), and constraints. The SQL DDL
> statements (CREATE / ALTER  + TYPE / TABLE / CONSTRAINT) used for creating
> or modifying these items happened over time and are *not *available at the
> time of mapping. So, to determine the RDF (actually RDFS/OWL) classes and
> properties, the Mapper needs to look at the DB-vendor-specific data
> dictionary tables.
>
> Presumably the Mapper knows about the data dictionary tables of the server
> database (i.e., where the data that s/he wants to map to RDF is located) or
> we could have an adapter for each DB-vendor to expose this info in a uniform
> way.
>
> Based upon his/her understanding of the database, s/he specifies a mapping,
> using a RDB2RDF mapping language to be determined (D2RQ / SQL / ... /
> something-very-simple / ...), of the DB objects to RDFS/OWL classes and
> properties. (If there is a simple mapping, one can write a
> DB-vendor-specific program to generate this mapping based upon some
> conventions, e.g., URIfy the table/view/col names.)
>
> Leaving out the DB object references and other such details included in the
> mapping, one can then create an RDFS/OWL ontology (or schema: i.e., classes
> and properties).
>
> The Querier sees the RDFS/OWL ontology/schema and designs his/her SPARQL
> query based upon that. (Translation of the SPARQL query, as I understand, is
> outside the scope of this WG.)
>
> So, instead of calling it a mapping of SQL DDL statements (which the Mapper
> never sees directly), can we consider calling it a mapping of "RDB Data
> Definitions" (DB table/view/"named-"query/column/constraint definitions
> available from the data dictionary).
>
> Thanks,
> - Souri.
>
>
> Juan Sequeda wrote:
>
>
>
> On Mon, Nov 23, 2009 at 10:45 PM, Ezzat, Ahmed <Ahmed.Ezzat@hp.com> wrote:
>
>>
>>
>>
>>
>> Thanks Juan…
>>
>>
>>
>> Tomorrow we are continuing presentations.  It might be a good idea to use
>> email to exchange thoughts till we are done with the presentations next
>> week.
>>
> Sounds good
>
>>  One observation that I have which is kind of border line is the notion
>> of R2RML supporting SQL DDL statements.  I think what we mean is RDFS/OWL
>> equivalence to SQL DDL statements.  In other words, we are not creating
>> SQL table but rather creating a class that maps an existing SQL table to the
>> RDF world, etc..
>>
> Yes!
>
>>  If the above makes sense, one possibility is to have a table with 3
>> columns: SQL column (DDL statement): RDFS column (equivalence): OWL column
>> (equivalence).  In the SQL column you would have create table, then in
>> OWL the equivalent is to create a class, etc…
>>
> That is exactly what I was thinking of. I was just trying to figure out how
> to create a table in a wiki.
>
>>
>>
>   Ahmed
>>
>>
>>
>> *From:* Juan Sequeda [mailto:juanfederico@gmail.com]
>> *Sent:* Monday, November 23, 2009 19:22
>> *To:* Ezzat, Ahmed
>> *Cc:* Marcelo Arenas; Sören Auer; Michael Hausenblas; Harry Halpin;
>> RDB2RDF WG
>> *Subject:* Re: ISSUE-3
>>
>>
>>
>> I have now updated the list and added justification
>>
>>
>>
>> http://www.w3.org/2001/sw/rdb2rdf/wiki/Requirements/DDLCoverage
>>
>>
>>
>> I have based this on our previous work [1] and a technical report of ours
>> which is a survey of different approaches that take DDL and convert it to an
>> ontology [2].
>>
>>
>>
>> Looking forward to a discussion about this tomorrow.
>>
>>
>>
>> [1] http://www.springerlink.com/content/mv58805364k31734/
>>
>> [2] ftp://ftp.cs.utexas.edu/pub/techreports/tr09-04.pdf
>>
>>
>>
>>
>>
>> Juan Sequeda, Ph.D Student
>> Dept. of Computer Sciences
>> The University of Texas at Austin
>> www.juansequeda.com
>> www.semanticwebaustin.org
>>
>>  On Mon, Nov 23, 2009 at 6:09 PM, Juan Sequeda <juanfederico@gmail.com>
>> wrote:
>>
>> I'm sure we will all have different points of views and opinions on this.
>>
>>
>>
>> For example, the way we have done this (and we have shown this in our dexa
>> paper [1]): We have mapped the create table to classes, except if the tables
>> are binary relations, then they would map to object properties. How do we
>> automatically do this? Reading the data dictionary or using describe
>> operation.
>>
>>
>>
>> I apologize for being a bit behind on this but I've been under the
>> weather. I'm in the processing of finishing the list and I will take Ahmed's
>> suggestion of writing justifications. We can then discuss this tomorrow.
>>
>>
>>
>> [1] http://www.springerlink.com/content/mv58805364k31734/
>>
>>
>> Juan Sequeda, Ph.D Student
>> Dept. of Computer Sciences
>> The University of Texas at Austin
>> www.juansequeda.com
>>
>> www.semanticwebaustin.org
>>
>>
>>
>>  On Mon, Nov 23, 2009 at 11:48 AM, Ezzat, Ahmed <Ahmed.Ezzat@hp.com>
>> wrote:
>>
>>
>>
>> Here is my 2 cents:
>>
>> The format of the information Juan is collecting needs to be modified.  I
>> do not know if Juan read his email or not?  The list he is collecting is the
>> list of features that the mapping language needs to support on the SQL side,
>> i.e., ability to find out the PK for a given relation R; this is doable
>> using DESC (describe) operation. I asked Juan to put usage scenarios for
>> each item suggested on his page.  For example, scenario where creating a
>> table is needed?  I did not want to go and edit the format of the page, and
>> instead asked Juan to do that. Then myself and others can edit content.
>>
>> There is another dimension to the mapping issue on the other side, which
>> is not part of what Juan is doing, i.e., RDFS or OWL usage of the
>> information you get from the RDBMS side.
>>
>> Ahmed
>>
>>
>>
>> -----Original Message-----
>> From: Marcelo Arenas [mailto:marcelo.arenas1@gmail.com]
>> Sent: Monday, November 23, 2009 07:56
>> To: Ezzat, Ahmed
>> Cc: Juan Sequeda; Sören Auer; Michael Hausenblas; Harry Halpin; RDB2RDF WG
>> Subject: Re: ISSUE-3
>>
>> Hi,
>>
>> I am a bit confused about what we mean by "supporting a feature" of
>> DDL.  Assume that we are given a relation schema R(A, B) where A is
>> the primary key of R. The list says that primary keys should be
>> supported,  so should attribute A be a primary key of R(A, B) in the
>> RDF representation of this database? The problem with this is that
>> there is no way to enforce a key in RDF (RDFS). Are we just going to
>> describe in RDF that A is a primary key without enforcing it?
>> What about OWL? Are we planning to use owl:FunctionalProperty to
>> indicate that A is a primary key? Thanks!
>>
>> All the best,
>>
>> Marcelo
>>
>> 2009/11/18 Ezzat, Ahmed <Ahmed.Ezzat@hp.com>
>> >
>> >
>> > Hi Juan,
>> >
>> > You can approach this problem from different angles. Request: on the
>> discussion page as an example:
>> >
>> > Tables item:
>> >
>> > Create Table
>> > Must be supported
>> >
>> >
>> > Let me suggest one of two formats: you can list for the Table bullet:
>> create table, delete table, alter table, describe table or just list the
>> ones you want to support?  I see as an example Describe table is the obvious
>> one as a must.
>> >
>> > In either scenario you want to adopt, please have next to any DDL
>> statement you want a justification, i.e., scenario(s) justifying its use.
>> > Thanks,
>> >
>> > Ahmed
>> >
>> > Ahmed K. Ezzat, Ph.D.
>> > HP Fellow, Business Intelligence Software Division
>> > Hewlett-Packard Corporation
>> > 11000 Wolf Road, Bldg 42 Upper, MS 4502, Cupertino, CA 95014-0691
>> > Office:      Email: Ahmed.Ezzat@hp.com Off: 408-447-6380  Fax:
>> 1408796-5427  Cell: 408-504-2603
>> > Personal: Email: AhmedEzzat@aol.com Tel: 408-253-5062  Fax:
>> 408-253-6271
>> > From: Juan Sequeda [mailto:juanfederico@gmail.com]
>> > Sent: Wednesday, November 18, 2009 7:00 AM
>> > To: Sören Auer
>> > Cc: Michael Hausenblas; Harry Halpin; RDB2RDF WG; Ezzat, Ahmed
>> > Subject: Re: ISSUE-3
>> >
>> > On Tue, Nov 17, 2009 at 3:32 PM, Sören Auer <
>> auer@informatik.uni-leipzig.de> wrote:
>> > Michael Hausenblas wrote:
>> > Though I see your point, DDL is the most general form of what we are
>> talking
>> > about, here, covering data model elements
>> >
>> > I actually think DDL is not a very general form, but rather a very
>> specific language for creating and manipulating relational schema objects.
>> >
>> >
>> > IMO, we should stick to the specifics. Hence, using DDL should be
>> appropriate.
>> >
>> >
>> > (for sure DROP, ALTER is not in
>> > scope, but this is a no-brainer, I guess ;)
>> >
>> > Ok, but DDLs consist *only* of such statements, cf. e.g.:
>> >
>> > http://dev.mysql.com/doc/refman/5.1/en/sql-syntax-data-definition.html
>> >
>> >
>> > What is missing from that list, that we should take in account?
>> >
>> > I'm fine with DDL and think we have used it in the discussion throughout
>> as
>> > such ...
>> >
>> > We can use the term DDL if everybody in the group got used to it, but
>> from a conceptual point of view I think this is wrong and in order to avoid
>> confusion with the outside world we should rather talk about /data model
>> elements/ or /schema objects/.
>> >
>> > I think we can combine the two in this list that we are going to make.
>> But we should also be on the same page. I see that you have Foreign Key,
>> Integrity Constraints and Referential Integrity separate. Why? Aren't
>> referential constraints a subset of integrity constraints. And a foreign key
>> is a referential constraints. Those shouldn't be separate, but express them
>> as subclasses.
>> >
>> > Sören
>> >
>> >
>>
>>
>>
>>
>>
>
>
>