[r-t] A new Spliced Surprise Major canon

Graham John graham at changeringing.co.uk
Sat Mar 9 10:57:35 UTC 2013

Don wrote:

> I don't think this is a sufficient condition. Unless I'm missing
> something it doesn't necessarily imply there is a partitioning
> of the leads into two atw sets. Or am I, in fact, missing
> something?

and Mark wrote:

> Hmm, I see what you mean. I'm not sure the correct answer at the moment.
> Anyone?

Not sure what is meant by the "correct" answer, as Don's assertion is

My spliced software was developed over 20 years ago, so needs a complete
rewrite, but the iterative tuning functions it contained worked in a similar
way to that you describe. It would take each call to call block of leads and
score all substitutions of that block using truth, atw, com, music, maximum
blocksize (to reduce long blocks of the same method), method balance, and
composition length. It then kept the best and moved on to the next node,
cycling the composition until it reached a steady state (maximum
improvement). I would then make some manual changes to the composition or
the scoring criteria and try again.

Truth is an interesting score, apparently binary. I found that by counting
the false rows, I could substitute a false block, then see if refining the
rest of the composition would improve (reduce) the falseness score. If it
comes back to zero, keep the result.

One of the things to note is that this process is very sensitive to the
scoring you use. I used a simple ranking of the measures - although that
means that one measure always takes priority over another e.g. ATW over
Music, I would rearrange the ranking to see if any further improvement could
be made without detriment to the overall score (i.e. after steady state is
reached again).

A further problem is that compositions typically contain courses that are
split up across the composition. This means that trying to substitute one
block of leads will be limited because the rest of the course is elsewhere
and the calls constrain the structure. One way of overcoming this I found,
was to use NCR (no course revisited) callings to give complete freedom for
substitution. For example, the calling in the 7 method composition below
contains all sixty in-course courses, each one visited once. 


All the work can be achieved by refining a random composition, but the more
methods you add, the harder that gets. Obviously it depends on the methods,
but the maximum I got was 14, without deliberately searching for alternative
methods that would work better. I plan in my software rewrite (when I get
round to it) to identify atw sets i.e. a minimum set (preferably 7 leads in
Major) across different courses that give atw for a method. You then repeat
this for each method, then lock those leads and refine the composition
around them for music and other criteria.


More information about the ringing-theory mailing list