Temperament finder failing for strange "limits"

I'm searching for "consonant class scales" (see the wiki article I wrote), and a good way to search might be to run temperament finder searches in "limits" like 2.7/6.6/5.5/4 . A comma can only lead to a consonant class scale if its monzo entries for the consonances all have the same sign. This is easy to see at a glance in the 2.7/6.6/5.5/4 basis, but requires some computation to see in the 2.3.5.7 basis.

However, usually when I run temperament finder searches for such "limits", I get a stack trace ending in MemoryError. This seems strange because on the face of it I can't see why these searches would take substantially more memory than the usual ones.

Any idea why this happens, Graham?

Keenan

--- In tuning-math@yahoogroups.com, "Keenan Pepper" <keenanpepper@...> wrote:
>
> I'm searching for "consonant class scales" (see the wiki article I wrote), and a good way to search might be to run temperament finder searches in "limits" like 2.7/6.6/5.5/4 . A comma can only lead to a consonant class scale if its monzo entries for the consonances all have the same sign. This is easy to see at a glance in the 2.7/6.6/5.5/4 basis, but requires some computation to see in the 2.3.5.7 basis.
>
> However, usually when I run temperament finder searches for such "limits", I get a stack trace ending in MemoryError. This seems strange because on the face of it I can't see why these searches would take substantially more memory than the usual ones.

Seems like I can work around this bug by simply inverting all the consonances. For example, 2.8/7.7/6.6/5 consistently fails, but 2.5/3.12/7.7/4 consistently works.

Hope this gives you some clue,
Keenan

"Keenan Pepper" <keenanpepper@gmail.com> wrote:
> I'm searching for "consonant class scales" (see the wiki
> article I wrote), and a good way to search might be to
> run temperament finder searches in "limits" like
> 2.7/6.6/5.5/4 . A comma can only lead to a consonant
> class scale if its monzo entries for the consonances all
> have the same sign. This is easy to see at a glance in
> the 2.7/6.6/5.5/4 basis, but requires some computation to
> see in the 2.3.5.7 basis.
>
> However, usually when I run temperament finder searches
> for such "limits", I get a stack trace ending in
> MemoryError. This seems strange because on the face of it
> I can't see why these searches would take substantially
> more memory than the usual ones.
>
> Any idea why this happens, Graham?

Right, what happens is that 1p or <1 0 0 0] is coming up in
the results. To decide if it's ambiguous, we need to know
if there are any mappings with a lower error. There happen
to be so many that the interpreter runs out of memory
trying to store them all. So, it's better to give up
listing them once we know there are a lot, which is what it
does now.

The interesting thing is that, whereas 1p has a TE error of
1039 cent/oct and 1qrs or <1 1 1 1] has an error of "only"
467 cent/oct, 1p is the only mapping of 1-equal that ends
up in the shortlist. It looks like the rule of thumb that
complexity is proportional to the number of steps to the
octave doesn't hold and <1 0 0 0] has lower badness because
of its lower complexity.

Graham