Re: filling in missing dates in a time series

Bob Badour wrote:

> paul c wrote:
> 

> 
> 
> Paul,
> 
> I suspect you misunderstand the restrictions Codd proposed. 

Could well be, that's why I'm here (and there). He didn't write a lot about the restrictions, much of what I know is hear-say. After many years, I'm still trying to understand a talk he gave about Christie Brinkley's telephone number!

> He did not 
> restrict a dbms from having a virtual relation with every discrete time 
> within some time representation or a virtual relation with all 4 billion 
> or so values representable by a 32-bit integer etc.

I know the quote mentioned infinity, but I try to stay away from the term, preferring words like 'lengthy'. I would like to see a real product that supported such virtual relations as I find definitions easier to understand than procedural code - I certainly wasn't advocating such code, just saying that I think it is current orthodoxy.

> 
> He only restricted infinite sets. Due to the nature of infinite sets, 
> one cannot represent all of the members of the set in a finite space. 
> Essentially, once the number reaches a certain size, the number of bits 
> required to represent it becomes too big for whatever store you have.
> 
> First order logic allows one to describe infinite sets. For instance, 
> the set of integers is infinite. The set of integers not equal to 1 is 
> likewise infinite.
> 
> To avoid those things, Codd restricted against infinite sets and limited 
> negation to difference.
> 
> Thus you cannot specify the infinite set of integers or the infinite set 
> of integers not equal to 1.
> 
> You can specify the finite set of integers representable in 32-bits or 
> 64-bits or even 128-bits, 

> and you can specify the finite set of such > integers that are not equal to 1.

That's one of the things I want in a product. Admittedly my orientation is towards what a machine can do without disrespecting theory. To change the example a bit, I've personally never felt that it was the machine's responsibility to catch situations like overflow or division by zero, feeling that this is where humans have their natural place. I've come to realize that this is controversial because I've often been compared to a 'theoretical philistine' for saying it.

> 
> Of course, even though the model allows it, one will likely have to wait 
> too long for any query that evaluates every 128-bit integer not equal to 1.

Yes, have been doing some timing tests lately and wish I had been born thirty years later. Saw the interview with Eckart on computerworld today and he quoted a speed for some arithmetic operation of .00002 seconds on the 1946 Eniac, which in rough terms was about the speed of the S360/30, twenty years after the Eniac. Now, on this little Pentium Mobile I have here, which the hardware people consider medium speed on the desktop I am getting results that are roughly hundreds of times faster, sometimes thousands. Doesn't take very long now to something a billion times - I have to run my tests for many many seconds because there aren't enough timer pops to get accurate measurements on the speed of this thing. Plus, I could fit many of the multi-user apps I used to work on in memory. I believe this undercuts the traditional starting point for much of the concurrency theory of the last thirty years, not to mention many of the physical comprises that today surround Codd's basic relational operators.

I guess my point in this reply is that the practical limits have changed dramatically since Codd's first papers - the practical ceilings are much higher now.

pc Received on Wed Feb 15 2006 - 21:17:57 CET