Rank 3 temperament search

I can do these now that I have a wedgie complexity method. The library code's in

http://x31eq.com/temper/regular_wedgie.py

It requires Numeric Python to do the matrix operations for RMS optimization, among other things. It also needs my own wedge product code to do the invariants and complexity. The wedge products are currently slowing it down. I'm sure there are better ways of calculating them but this is what I have that works.

I did a rank 3 temperament search in about 2 minutes. Probably I need more seed ETs, but the search time is cubic in these so I'll leave it there for now. Here are the ratios for the 7-limit ones:

4375:4374
2401:2400
184528125:184473632
52734375:52706752
200120949:200000000
5250987:5242880
3136:3125
225:224
6144:6125
5120:5103
1029:1024
10976:10935
16875:16807
126:125
245:243
19683:19600
15625:15552
235298:234375
1600000:1594323
4096000:4084101
81:80
823543:820125
26873856:26796875
1728:1715
64:63
283435200:282475249
321489:320000
156250000:155649627
50:49
4000:3969
100442349:100000000

Obviously there are easier ways of getting 7-limit commas, but at least this shows the search is making some sense. The ET-search gets to be easier than a comma-search in the 13-limit, so here's a 13-limit planar temperament:

[ 6.71952672 11.99319487 16.44702813] cent steps

wedgie:
(0, 6, -6, 0, 5, -5, 0, -1, -14, 14, -6, 6, 0, 36, 0, 0, 29, -14, 14, 84)

mapping by steps:
[[ 15 24 35 42 52 56]
[ 19 30 44 53 66 70]
[ 53 84 123 149 183 196]]

wedgie based complexity: 2.407
RMS weighted error: 0.000232
max weighted error: 0.000347

It's the first one in the list that has a smaller RMS error than mystery.

I can do rank 4 searches as well, as high as you like.

Graham