Re: [Gnu-arch-users] Re: -L option for tag?

[Andrew Suffield]

On Fri, Sep 19, 2003 at 05:56:00PM -0700, Tom Lord wrote:
> 
> 
>     > From: Robert Collins <address@hidden>
> 
>     > As we get a good selection of low level merge tools, I think we will be
>     > able to start describing algorithms for choosing amongst those tools at
>     > a given merge point, and from there start to build higher layer 'magic
>     > tools'.
> 
> Good luck.
> 
> I don't see how you can do it w/out AI but, in trying, perhaps you'll
> come up with some (additional) interesting approximations.

"AI" is a misleading label.

"It is impossible to create artificial intelligence, due to the lack
of natural intelligences for use as a model". That's more serious than
it sounds - we've no idea what "intelligence" is, so we don't know how
to set about creating it. Or even how to recognise that we have done
so.

There are two basic approaches I can think of, offhand. The first is
heuristic - a large group of unrelated rules which each check "easily
recognisable" features, and a training database of scenerios and the
decisions made by a human. From these, there are a set of methods for
generating a ruleset that will endeavour to make the same decisions in
similar situations. This is the approach taken by spamassassin. In
scenarios where you can create a set of base rules which accurately
distinguish between the different situations, it works quite well.

The other approach is to look at each tool, and try to qualify what
scenarios it works in, and build a pattern matching algorithm based on
these qualifications. This works when there are clear advantages to
each tool, and these can be encoded as pattern matches.

-- 
  .''`.  ** Debian GNU/Linux ** | Andrew Suffield
 : :' :  http://www.debian.org/ |
 `. `'                          |
   `-             -><-          |

signature.asc
Description: Digital signature