- I finished reading the CSP chapter (the PDF file posted on the web
site) over the weekend. I have a comment about one of the sentences
in section 5.2 talking about arc consistency.
On p. 146, it says that "although this [arc consistency] is
substantially more expensive than forward checking, the extra cost is
usually worthwhile." This doesn't seem to agree with the CSP survey
paper by Kumar (1992), which mentions that for loosely constrained
problems like N-queens that arc consistency is too expensive. (This is
also borne out by my own experience writing loosely-constrained CSPs
that have used AC-3.)
In this regard, I think that it would be good for section 5.2 to
actually have a brief mention of tightly constrained problems vs.
loosely constrained problems that would lead into a discussion of why
you would only apply FC vs. MAC/AC-3.
My 2 cents.