Re: [xml-dev] Develop a theory of one's data prior to developing adata m

[Peter Hunsberger <peter.hunsberger@gmail.com>]

I originally thought you were maybe heading down the path of data modelling and theories about domain modelling.  At the more abstract level I think the theories you might be hinting at are structural.  In particular, bits without structure are useless, so we chunk them into bytes, or words, or registers, and assign op codes to them and meaning as pointers and we build up many theories of computing using these things.  Some of them, such as the theories of how Turing machines work being pretty central to any discussion about how algorithms and data work together... I'd also argue that the are many other higher level ways of structuring data that allow us to create very useful theories.  In particular, relational theory and graph theory come to mind.  These two suggest your third point is not true; for example, graph traversal algorithms do not work well over relational data structures.  And then of course there is the more "meta" argument that has been suggested to you on several occasions before: algorithms are data.  There are algorithms that exist only to process other algorithms.  So, I'd say you (still) have it backwards: data is primary, without data algorithms have no use, but data can still be useful after the algorithms are done with it.  For example, this e-mail might (maybe) fall into that category...

On Sun Dec 21 2014 at 7:28:11 AM Costello, Roger L. <costello@mitre.org> wrote:

Rick Jelliffe wrote:

 

Ø  You only have data because you have software.

 

Does that mean that data without software is pointless/meaningless/futile?

 

Does that mean that software should dictate the structure and content of data?

 

As I see it, there are theories about processing data. For example, there are theories about parsing data. But there doesn’t appear to be theories about the data itself. For example, there is no theory about the data that is input into a parser.

 

So this seems to be the situation with regard to theories:

 

 

To take the parsing example, we have this:

 

 

Theory applies to the realm of processing/actions/software. Theory does not apply to the realm of data. Yes?

 

I am led to these conclusions about developing a data model:

 

1. First, identify what processing must be done on the data. For example, “Structure the data (a linear sequence of symbols) into a parse tree.”

 

2. Develop theories and algorithms for that processing. For example, develop theories about grammars and then develop parsing algorithms.

 

3. Model the data so that it is well-suited to consumption by the algorithms. Algorithms dictate the data model, the data model does not dictate the algorithms. Software/processing/algorithms is primary, data is merely stuff to be gobbled up by the software.

 

Do you agree?

 

/Roger