[r-t] Similar compositions

[Graham John]

On 23 January 2018 at 11:56, Alan Burlison <alan.burlison at gmail.com> wrote:
> As I understand it, the order of the rows was taken into account, from
> John's description he says he generates 6 hashes which are used for
> different purposes.

Yes, but it is not clear to me how the row hashes help in John's
method. I would have thought that hashing leadends and leadheads where
there are calls (plus the order of these) would be better on its own,
since it would catch the same composition rung to different methods.
However, while good for methods with Plain Bob leadheads, it would not
catch WHW rung to Minor methods without Plain Bob leadheads. This is
why I think it is necessary to focus on the calling.

>> Secondly, hashes for the same composition applied to a different
>> method would not match, nor would rotations or reversals using the
>> same method.

> Would that not count as a different composition anyway?

Definitely not. Middleton's Cambridge has been rung to lots of
methods, but it is still considered to be Middleton's composition. As
I recall it was first used for London. Similarly rotations and
reversals. There are often good reasons for separately recording some
reversals and rotations, as they have different music properties and
different coursing orders/part heads that can make them easier to
conduct.

> Rotations and reversals seem tricky, are there any standard ways of
> 'normalising' compositions?

I very much doubt it. However, it doesn't really matter how they are
normalised as long as it is done consistently. Hashing all the rotated
and reversed callings and using the lowest hash would work fine.

>> Trivial variation checking is rather trickier. It could need
>> significantly more hashes to be stored using a range of additional
>> factors, which can then be compared using a scoring system, in a
>> similar way to spam checkers. This will require some experimentation
>> to refine, but is nevertheless feasible.

> That's why I pointed to the various string similarity algorithms. I think
> the hard part will be getting good distance metrics.

Yes, they might work with a consistently presented calling string, but
as John pointed out, you have decide a boundary for what is considered
trivial vs distinct.

Graham