Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!swrinde!elroy.jpl.nasa.gov!lll-winken.llnl.gov!ames!kronos.arc.nasa.gov!polya.arc.nasa.gov!minton
From: minton@polya.arc.nasa.gov (Steve Minton)
Subject: Re: Problem: Constraint Based Scheduling
Message-ID: <1995May12.180043.16208@ptolemy-ethernet.arc.nasa.gov>
Sender: usenet@ptolemy-ethernet.arc.nasa.gov (usenet@ptolemy.arc.nasa.gov)
Nntp-Posting-Host: polya.arc.nasa.gov
Organization: NASA/ARC Computational Sciences Division
References:  <smithd.800249825@borealis>
Date: Fri, 12 May 1995 18:00:43 GMT
Lines: 36

Donald Smith writes:
> I am working on my MSc thesis where I am examining some heuristics
>for Constraint based intelligent scheduling.  My implementation
>language is CLP(FD) and my chosen scheduling domain is the classic 
>"job shop" model.
>
> I'm having some problems trying to scientifically "interpret" my 
>results.
> [rest of post omitted]

It seems to me that you are asking exactly the right questions. Unfortunately,
I don't think there are any cut-and-dry answers. These are exactly the
questions that people who do empirical experiments in AI must ask themselves.
As you pointed out, scheduling is NP-hard, so (assuming P /= NP), the problem
is, in some sense, "hopeless".  But, in practice, there are many interesting
reseach avenues. Some tactics that people have taken include:
   -- identifying provably tractable subclasses of problems
   -- Finding a heuristic which works well in practice (better than other
      methods on some problems) and trying to identify what sort of 
      problem distributions it works well on. (This is a laudable goal, but
      its often hard to provide a precise statement of the result.)
   -- applying machine learning (or other AI techniques) to "discover" 
      interesting domain-specific heuristics. 
   -- creating a problem solver  that has a variety of heuristics and having it 
      automatically select a good heuristic for a given problem.

This is not a complete list (OK, I mentioned the things I'm personally most 
interested in.) I'd recommend that you keep on thinking about 
wht sort of contribution you'd like to make.  

Once it comes down to doing experiments, there are a variety of statistical
methods  hat you can use. But first, it's useful to think about WHAT
you want to show...

- Steve Minton
  minton@Ptolemy.arc.nasa.gov
