Re: PRD (Prod Rule Dialect) breakout scribe notes from F2F8 6 Nov 07

Paul Vincent wrote:

> PRD BREAKOUT F2F8 Boston - 6Nov07
> 
> =================================
> 
>  
> 
> CSMA: Do we want PRD as simple as possible but missing major features 
> like negation?
> 
>  
> 
> Gary: Most rules' NOT are simple terms and constants rather than frames.
> 
>  
> 
> CSMA: Another type of common negation is NOT(pattern).
> 
>  
> 
> CSMA: Should design for negation.
> 
>  
> 
> Gary: No model theory for PR.

Indeed!


> CSMA: Model theory for the operational semantics may be possible.


Model Theory and Operational Semantics are orthogonal concepts. Model Theory
fits unchanging truth. In this way, stateful computation such as that performed
by PR systems is at odds with Model Theory. This is what Gary meant - I think
(Gary?) Formal operational semantics, such as based on Gordon Plotkin's SOS
semantics (http://citeseer.ist.psu.edu/plotkin81structural.html), doea not
care so much for models, but for effective computation with changing effects
- such as performed by PR systems. In that framework, formality is just as
rigorous as any Model Theory's (i.e., theorems may be proved about the systems
and their computations, abstract interpretation may be devised, typing and
module system may be rogorously conceived, etc.). I believe that this is
the best way to render PR computation. For an example of such a formal
operational semantics, see a presentation I did this past summer for IFIP
group on Rewriting (http://rewriting.loria.fr/IFIP-WG1.6/. Slides of my
talk: http://koala.ilog.fr/wiki/bin/view/Main/HassanAitKaci#17.)



> Adrian: Select language based on conditions in BLD.
> 
>  
> 
> CSMA: PRD needs to re-use as much of BLD as possible.
> 
>  
> 
> Gary: PRD should reference BLD doc. Rather than repeat it. 1st draft can 
> concentrate on actions rather than extending conditions.
> 
>  
> 
> Gary: EG PRD Core with BLD Condition Language + Actions.
> 
>  
> 
> CSMA: Concern: better not to repeat text, but people won't read 2 docs 
> instead of 1. Could do a doc refactoring exercise so BLD and PRD can 
> share elements.
> 
>  
> 
> Gary: BLD doc is not easy to understand. Its a theoretician point of 
> view rather than end-user or implementor. So comments applies to BLD too.
> 
>  
> 
> Bob: Model-theory semantics will always be a challenge to read.
> 
>  
> 
> CSMA: ACTION: to discuss BLD restructure to allow easier re-use / 
> sharing by PRD etc with RIF WG
> 
>  
> 
>  
> 
>  
> 
>  
> 
> CSMA: Human readable syntax / presentation syntax: not restructured in 
> draft as this is not useful
> 
>  
> 
> Gary: Presentation syntax = very useful. EG test cases as examples for 
> implementors.
> 
>  
> 
> Bob: Fundamental is the XML syntax. Reading XML docs is not best for 
> examples. Can manage with informal presentation language with no formal 
> / explained syntax,
> 
> provided we include the XML doc
> 
>  
> 
> Gary: But we already have the presentation syntax in BLD so this is 
> done. No motivation to drop it.
> 
>  
> 
> CSMA: Do not like a Working Draft defined by reference.
> 
>  
> 
> Gary: Do not like not repeat-by-copy vs repeat-by-reference.
> 
>  
> 
>  
> 
>  
> 
>  
> 
> Gary: Can we repeat semantics in operational terms, or do a hybrid 
> approach (model-theory conditions, operational rules and actions)?
> 
>  
> 
> CSMA: Need operational semantics for conditions as conditions are 
> executed per the instantiated rules.
> 
> Instantiating rules is an operational semantics in terms of patter matching.
> 
>  
> 
> CSMA: Should limit amount of work moving conditions to PRD. But work is 
> already done (conditions in operational semantics).
> 
>  
> 
> Gary: Issue of differences in behavior: JESS re-fires a rule if a fact 
> in condition changes
> 
>  
> 
> Bob: In FIC the change must be in the state of the condition, not the 
> state of the fact / data.
> 
>  
> 
> CSMA: If JESS does not conform, then the doc needs to change.
> 
>  
> 
> CSMA: How should PRD handle such "tricks".
> 
>  
> 
> Bob: Real business rule systems rarely have ambiguity or chaining.
> 
>  
> 
> Gary: Still find need for conflict resolution.
> 
>  
> 
> CSMA: Problems like configuration still require solutions like inference 
> and chaining.
> 
>  
> 
> CSMA: In ILOG the resolution strategy is overridable to some extent. So 
> if the interpretation of RIF strategy is an engine command this makes 
> RIF default easier.
> 
>  
> 
> Gary: JESS / CLIPS: multiple rulesets / stack. Only rulesets in top of 
> stack can fire. Actions can load the stack. Within a ruleset can have 
> priorities.
> 
> Can select first or last recent rule in same priority.
> 
>  
> 
> CSMA: ACTION: All vendors to provide their rule selection strategy and 
> options, in a table set up by Christian on the Wiki.
> 
> 
> Paul Vincent
> 
> TIBCO | ETG/Business Rules


I believe that although we should proceed formally, what we formalize
should stay as simple and intuitive as possible and stay close to the
functioning of the majority of PR systems.

My 2 cents,

-hak
-- 
Hassan Aït-Kaci  *  ILOG, Inc. - Product Division R&D
http://koala.ilog.fr/wiki/bin/view/Main/HassanAitKaci

Received on Wednesday, 7 November 2007 06:17:46 UTC