Nice. Note that d doesn't need to be global.
The 63 bit (or 31 for older hardware) limit shouldn't be a problem in
practice?!
Wondering if it was possible to avoid boxing, I've found this version -
not faster,
though more parsimonious in space, so probably not for serious use:
combbitub =: 4 : 0
assert y<:<:##:_1 (32 b.) 1
lshift =.33 b.
or =.23 b.
k =. (y->:ix =. i.>:d=.y-x)lshift 1
z =. ,0
l =. 1#~ #k NB. "box-sizes" as successively 1s, i.#k, triangular nums etc.
for. i.x do.
nz =. _1 lshift z
z =. nz or {.k
for_j. }. i.#k do. NB. tack on the smaller "boxes"
z =. z, (j{k) or (-j{l) {. nz
end.
l =. +/\. l NB. bump up the box sizes
end.
#: z
)
3 (combbit -: combbitub) 5
1
Cheers,
Mike
10/11/2017 11:46, Erling Hellenäs wrote:
Hi all!
Bitmap version of comb. Unfortunately handles sets of max 63 items.
I tried to make a general version, but hit the problem with #: .
ts'5 combbit 10'
3.37831e_5 20864
ts'5 combbool 10'
3.89148e_5 22400
ts'5 combREBoss 10'
5.51649e_5 44416
ts'5 comb 10'
3.76319e_5 50944
ts'13 combbit 26'
0.44835 8.03039e8
ts'13 combbool 26'
0.880373 1.4314e9
ts'13 combREBoss 26'
2.39319 4.22297e9
ts'13 comb 26'
2.59969 4.41693e9
ts'60 combbit 63'
0.00416474 5.26221e6
ts'60 combbool 63'
0.0534253 1.73281e7
ts'60 combREBoss 63'
0.0344776 6.81943e7
ts'60 comb 63'
0.310994 1.38554e8
combbit=: 4 : 0
assert y<:<:##:_1 (32 b.) 1
lshift=.33 b.
or=.23 b.
k=.<"0 (y->:i.>:d=:y-x)lshift 1
z=. (d$<i.0),<0
for. i.x do. z=. k (or)&.> ,&.>/\. (_1&lshift)&.> z end.
#: ;z
)
Cheers,
Erling Hellenäs
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
---
This email has been checked for viruses by Avast antivirus software.
https://www.avast.com/antivirus
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
