Hi again,

I've implemented a "proper" custom branching (with branching descriptions etc). I've also run some experiments testing the effect of changing the c-d parameter, and results are not what I expected.

I used two different settings, c-d = 8 and c-d = 1, with both "basic" branching (i.e., only calling the built-in "branch" function, for each of two variable arrays, in the space constructor) and with my custom branching. In *all* cases, using c-d = 8 does more propagation than using c-d = 1! Runtimes do not always correlate with the amount of propagation done, but, for the custom branching strategy at least, using c-d = 1 is faster on all non-trivial problems. (A summary of the experiment results is attached.)

Is there an explanation for this result?

cheers,
                        /[EMAIL PROTECTED]
a) problem #
b) config: (B)asic or (C)ustom branching / c/d parameter
c) runtime (CPU)
d-h) propagations / fails / clones / commits / memory, as
 reported by Gecode branch-and-bound engine.

(a) (b)     (c)        (d)      (e)      (f)      (g)       (h)
p01 B/1     2.71     3834285     3841     3846     7686   13197588
p01 B/8     2.59     3909804     3751     3756    10269    3285648
p01 C/1     0.10      133519       61       66      126    1969396
p01 C/8     0.06      170481       59       64      180     788452

p02 B/1     3.38     4787730     4825     4834     9658   14051876
p02 B/8     3.20     4874971     4736     4745    12947    3483776
p02 C/1     0.12      117311       48       57      104    1378804
p02 C/8     0.08      137777       47       56      138     722868

p03 B/1
p03 B/8
p03 C/1    11.58    39299608     1206     1217     2422   18879388
p03 C/8     5.65    41244181     1197     1208     3132   10685884

p04 B/1  1517.48  1824093538   175530   175535   351064  235268800
p04 B/8  1459.20  1853133644   175292   175297   481871   51137448
p04 C/1     6.29    11958600      733      738     1470   20320524
p04 C/8     3.45    15770960      729      734     1990    9374460

p05 B/1
p05 B/8
p05 C/1
p05 C/8

p06 B/1
p06 B/8
p06 C/1  2153.45   175008251  1132620  1132681  2265300   15474508
p06 C/8  2596.59  1862705194  1132608  1132669  3153370    7281084

p07 B/1
p07 B/8
p07 C/1   180.70   413073514    44267    44298    88564   16783164
p07 C/8   206.05   524941236    44263    44294   121654   11801052

p08 B/1  2386.67  1029822075   248176   248187   496362  422224128
p08 B/8  2343.41  1060040732   247810   247821   680528   79204312
p08 C/1     2.92     4790821      229      240      468   13438012
p08 C/8     3.14     5315465      229      240      571    8980868

p09 B/1
p09 B/8
p09 C/1     5.00     9108983      633      644     1276   11865252
p09 C/8     6.02    11096982      632      643     1683    8850180

p10 B/1
p10 B/8
p10 C/1    30.47    50861990     3237     3262     6498   33494180
p10 C/8    40.12    73212827     3235     3260     8653   20844340

p11 B/1
p11 B/8
p11 C/1    46.54    78493706     4733     4750     9482   25300772
p11 C/8    61.05   112353167     4732     4749    12934   15863396

p12 B/1
p12 B/8
p12 C/1    14.90    20053428      898      915     1812   53482276
p12 C/8    21.99    31074325      896      913     2345   19991500
_______________________________________________
Gecode users mailing list
[EMAIL PROTECTED]
https://www.gecode.org/mailman/listinfo/gecode-users

Reply via email to