So, in your opinion...
I was asked to review a SQL feature under consideration. It looks pretty neat - pattern matching over partitioned, ordered rows. Answer questions like "find me the stocks that are exhibiting a 'W' pattern" (trying to catch them on the up-swing). Or maybe "find me things that started out expensive, but have (rapidly) over time become very cheap". Or network usage / audit trail analysis (looking for specific patterns of behavior).
So, I got the paper (you can get it here) and read it. I gave my feedback, but I'd like to hear yours - specifically:
- Do you think you'd find a SQL feature like this useful, truly useful.
- Do you think the suggested implementation is "complete" enough.
- Anything else you'd like to say
I can assure you the people working on this will be reading your feedback - definitely.
A word of warning though - the paper is a bit "dense", it might well take more than a single reading to 'get it'. A working knowledge of the existing set of analytic functions and their syntax would be really useful. If I didn't know analytics as well as I do - I'm not sure how far I'd make it through the paper and still be able to understand it.I can say - I've seen the SQL needed to recognize a "W" using analytics... It is not "pretty", this proposed language feature would make it pretty easy - once you 'got it'.