Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uunet!mcsun!ukc!edcastle!aiai!!jkk From: jkk@.uucp (John Kingston) Newsgroups: comp.ai Subject: Re: NEXPERT: Its Use In Larger System Development Keywords: Nexpert frames commercial objects Message-ID: <3770@skye.ed.ac.uk> Date: 16 Nov 90 09:41:02 GMT References: <1990Nov14.192306.27466@PacBell.COM> <133558.21166@timbuk.cray.com> <1990Nov15.212044.19864@portia.Stanford.EDU> Sender: news@aiai.ed.ac.uk Reply-To: jkk@aiai.uucp (John Kingston) Organization: AI Applications Institute, University of Edinburgh Lines: 212 >In article <133558.21166@timbuk.cray.com> sga@redwood19.cray.com (Steve Anderson) writes: > > I disagree with Mr. >Kingston's comment that Nexpert is a midrange tool, and also with most of the Cons listed for Nexpert from the previous discussion. > In article <1990Nov15.212044.19864@portia.Stanford.EDU> johner@portia.Stanford.EDU (John Lynch) writes: >I also have some thoughts on Nexpert I'd like to pass on but my mail is >bouncing as well. could you please post the good email path to this group? Since this discussion is getting lively, I am posting all the comments on Nexpert which I saved from the CLIPS vs Nexpert discussion. Hope this is useful. John K. COMMENTS ON NEXPERT: We have both Clips and Nexpert. As far as documention goes, if you want to understand how the inference engine works - CLIPS wins out hands down. Nexpert is a good rule-based system, but I strongly believe their object-oriented features are WAY oversold. I have been a KEE user for several years and it is what I would consider to be a true object-oriented tool. My experience with Nexpert is that you get very tired of the spreadsheet like rule and object editors, once you learn their commands. BTW, the Lisp-like syntax of Nexpert comes from its origin. The tool was originally written in Lisp, and then recoded in C. =============== regards 'mark ============================================= Mark Ahlenius voice:(708)-632-5346 email: uunet!motcid!ahleniusm Motorola Inc. fax: (708)-632-2413 Arlington, Hts. IL, USA 60004 I've been using Nexpert for several years and have a number of objections. The documentation is opaque or missing on a number of areas that are essential to building a large real-world application. In particular, the internal mechanics of the inference engine are not described at all. Hence it is impossible to determine how mechanisms such as normal rule-firing and if-change actions interact. The user interface is often cumbersome to use. We have built much of our knowledge base in a normal text editor (using Nexpert's lispy internal syntax) rather than use the rule editor. There are also some strikingly bad design decisions, such as organizing the knowledge base according to the name of the hypothesis rather than the name of the rule. Debugging can be painful. If the inference engine enters a loop it will often not respond to user commands, forcing an interrupt and restart. Many of the error messages are nearly useless. If a loop is detected during backward chaining, the user gets a message to that effect - but no indication as to what rule caused the message! Overall, Nexpert strikes me as having been designed by the marketing department and never tested on real problems. (I know, it is the best selling tool. What can I say? I'm merely reporting my experiences.) Don't be too impressed by the spiffy features such as the graphic display of the inference tree. That's the kind of thing that looks great during a demo and is rarely used during development. I do not recommend Nexpert. Our own group is looking for a different tool for our future projects. -- Scott Turner >From: srt@aerospace.aero.org (Scott "TCB" Turner) This posting reflects my opinions, and not those of The Aerospace Corporation. >From: srt@aerospace.aero.org (Scott "TCB" Turner) As long as I'm bitching about Nexpert... I'm in the process of debugging a problem with two interacting knowledge bases, so I load them up, open up the "Current Rule" and "Current Hypotheses" windows, and run the inference engine (called "Knowcess" for some unknown marketing reason) so I can watch what's happening. Result: absolutely nothing in the "Current Rule" and "Current Hypotheses" windows. Makes them kind of useless. Decided to check the documentation to see if I'm doing anything wrong. That turns out to be difficult to do; the documentation doesn't have an index. And from examining the table of contents of each chapter (which are numbered independently!) I can't find any mention of these windows. This is the kind of thing I run into every time I use Nexpert. Not recommended. -- Scott Turner >From: vcr@ROUGE.EDRC.CMU.EDU (V C Ramesh) nexpert-users@cam.edrc.cmu.edu is the mailing list set up for nexpert users for correspondence regarding nexpert object. Those who would like to be added to this list, can send mail to: nexpert-users-request@cam.edrc.cmu.edu Thanks, Ramesh vcr@cs.cmu.edu >From: vcr@ROUGE.EDRC.CMU.EDU (V C Ramesh) To clarify my earlier post, nexpert-users@cam.edrc.cmu.edu is a distribution list. Mail sent to this address gets automatically forwarded to all the users in the list (which is a mail alias). This mailing list is intended to serve as a bboard- like forum, for communicating problems (and soliciting solutions), initiating discussions, and offering suggestions concerning usage of Nexpert Object expert system shell. Anything pertaining to Nexpert Object should be fair game. Again, users who would like to be added on to the list, can send mail to: nexpert-users-request@cam.edrc.cmu.edu Any questions, comments or suggestions concerning this list may be sent to this (nexpert-users-request) address (or to the list, nexpert-users@cam.edrc.cmu.edu) as well. Ramesh vcr@cs.cmu.edu From: barshay@Stars.Reston.Unisys.COM ( Unisys) Hi, Thanks to everyone who posted and mailed responses to my CLIPS v. NEXPERT question! ... I am curious about the SQL interface that I heard NEXPERT is incorporating in their latest release Thanks again. Jane Summary: CLIPS Pros ------ C Source available makes it very flexible Easy to learn - you can "Get going" right away Easy to use for embedded applications (with good documention) Easy to call C routines from within CLIPS Uses Rete algorithm Cons ------ Not object-oriented (changing in 5.0) One user found it slow and a memory hog on the Sun 3/50 (this was with using only 10 rules) Lack of backward chaining ---------------------------------------------------------------------------- NEXPERT Pros ------ Easy to embed. Agenda Monitor to observe inference engine Graphical Interface Cons _____ This was a big complaint - Documentation hard to understand (suggest taking formal training) One user found forward chaining "tortuously" slow" with medium to large rule base (doesn't use Rete algorithm) Object-Oriented features overrated because does not have message sending capabilities Rule editor is "cumbersome" Debugging painful John Kingston, AI Applications Institute, University of Edinburgh, 80 South Bridge, Edinburgh EH1 1HN, Scotland E-mail jkk@uk.ac.ed.aiai, phone 031-225 4464 ext. 229 FAX: 031 226 2730 Arpanet: J.Kingston%uk.ac.ed@nfsnet-relay.ac.uk TELEX: 727442 UNIVED G