Forums before death by AOL, social media and spammers... "We can't have nice things"
|    comp.ai    |    Awaiting the gospel from Sarah Connor    |    1,954 messages    |
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
|    Message 1,199 of 1,954    |
|    Ted Dunning to Jeff Barnett    |
|    Re: RETE rules design......    |
|    04 Oct 06 08:40:51    |
      From: ted.dunning@gmail.com              Jeff Barnett wrote:        ...       > If I understand your message correctly, it seems that you believe that       > RETE has something to do with the problem-solving that is apparent to       > the developer of the application system. It doesn't.              The original poster was indeed a bit confused on this point.              But their real question was to do with advice to a rule builder.              My advice is "don't".              Seriously, the single most important criterion that determines whether       you can effectively maintain your system is how large the largest set       of interacting rules is. If it gets above 20-30 rules (other than       simple fact rules), most people will have serious problems maintaining       the system. In some special cases, even 10 rules can be hard to       maintain.              If you are able to simplify the graph of rule interaction, you may be       able to increase this number significantly. The simplest ways to       simplify the rule set are stratification in which each class of rules       only refers to "lower" classes and separation in which multiple clumps       of rules operate relatively independently.              One of the major problems with rule systems is that there are few tools       that perform the analogical information hiding functions that       object-orientation provides for imperative programming. Without a       strong structuring paradigm like this and with the much more subtle       interactions that you get between rules, you can put yourself into a       world of hurt very, very quickly.              My own experience has to do with supporting large machine translation       systems written in Prolog and various business logic components for       fraud detection systems. In general, I would equate the complexity and       maintenance difficulty of a thousand line Prolog program with a 100,000       line moderately structured program in an imperative language (and this       is not a good comparison).              [ comp.ai is moderated ... your article may take a while to appear. ]              --- SoupGate-Win32 v1.05        * Origin: you cannot sedate... all the things you hate (1:229/2)    |
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
(c) 1994, bbs@darkrealms.ca