RE: [xsl] Test code and results comparing Muenchian Method with the method suggested by Sergiu Ignat

["Michael Kay" <mike@xxxxxxxxxxxx>]

Measuring a single Saxon run from the command line tells you almost nothing
- you're essentially measuring how long it takes Java to load Saxon into
memory. There are "undocumented" options -3 and -9 to run the transformation
thrice or nince (?) respectively, this gives you an indication of how long
it takes to stabilize. 

Given that, the next interesting thing to do is to take measurements on
different file sizes, and see how they scale. Anything less than 100
elements is rather uninteresting.

There are at least three ways of scaling a grouping problem: (a) keep the
number of groups constant, (b) keep the number of items per group constant,
or (c) generate grouping keys at random from a fixed population. Different
algorithms are going to behave differently depending which you choose - and
they all arise in practical grouping problems.

Michael Kay  

> -----Original Message-----
> From: M. David Peterson [mailto:m.david.x2x2x@xxxxxxxxx] 
> Sent: 17 December 2004 17:33
> To: xsl-list@xxxxxxxxxxxxxxxxxxxxxx
> Subject: [xsl] Test code and results comparing Muenchian 
> Method with the method suggested by Sergiu Ignat
> 
> In an effort to discover what types of gains could be made by using
> the method of grouping suggested yesterday by Sergiu Ignat I wrote the
> equivalent stylesheet using the Muenchian method and then ran a series
> of 5 transforms against both stylesheets using the supplied data.
> 
> No attempt was made to draw any conclusions from the results; I simply
> present them to you to let you determine for yourself what the results
> might suggest.
> 
> The code and description of the process used as well as the data from
> the results can be found here:
> 
> http://www.xsltblog.com/archives/2004/12/new_alternative.html 
> 
> Cheers!
> 
> <M:D/>
> -- 
> :: M. David Peterson ::
> XML & XML Transformations, C#, .NET, and Functional Languages 
> Specialist