| 00:06.00 | Notify | 03BRL-CAD:starseeker * 69209 (brlcad/trunk/include/bn/rand.h brlcad/trunk/src/libbn/tests/sobolseq.c): Add some comments. bn_sobol_urand doesn't need to be public - we need it only for the test app. |
| 00:07.31 | Notify | 03BRL-CAD:starseeker * 69210 brlcad/trunk/include/bn/rand.h: clarify |
| 00:41.29 | *** join/#brlcad LordOfBikes (~armin@dslb-088-066-136-232.088.066.pools.vodafone-ip.de) |
| 00:45.40 | *** join/#brlcad teepee (~teepee@unaffiliated/teepee) |
| 00:47.11 | *** join/#brlcad LordOfBikes (~armin@dslb-092-075-149-164.092.075.pools.vodafone-ip.de) |
| 00:55.03 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 01:00.57 | *** join/#brlcad LordOfBikes (~armin@dslb-088-066-148-121.088.066.pools.vodafone-ip.de) |
| 01:01.56 | *** join/#brlcad KimK (~Kim__@ip68-102-85-68.ks.ok.cox.net) |
| 01:03.16 | *** join/#brlcad fayjvpopdltvucpb (~armin@dslc-082-083-184-149.pools.arcor-ip.net) |
| 01:45.39 | *** join/#brlcad LordOfBikes (~armin@dslb-092-074-230-072.092.074.pools.vodafone-ip.de) |
| 02:20.47 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 05:53.47 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 06:02.09 | *** join/#brlcad amarjeet (~amarjeet@202.164.53.117) |
| 07:56.21 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 10:15.32 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 10:41.42 | *** join/#brlcad teepee] (bc5c2133@gateway/web/freenode/ip.188.92.33.51) |
| 10:59.28 | *** join/#brlcad yorik (~yorik@2804:431:f721:4f81:290:f5ff:fedc:3bb2) |
| 11:16.27 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 14:03.45 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 14:17.28 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 15:18.33 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 15:19.52 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 16:19.28 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 17:20.32 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 17:46.09 | starseeker | shakes his head - sobol isn't enough by itself for this. We probably need a scrambled sobol such as http://www.dtic.mil/docs/citations/ADA510216 |
| 17:46.44 | starseeker | pity SPRNG is GPL... |
| 17:47.34 | starseeker | brlcad: would it be a reasonable student project to implement the FSU sobol scramble? |
| 17:47.48 | starseeker | (the one documented in the ADA510216 report) |
| 17:52.25 | Notify | 03BRL-CAD:starseeker * 69211 (brlcad/trunk/include/bn/rand.h brlcad/trunk/src/libbn/sobolseq.c): simplify slightly |
| 17:55.46 | Notify | 03BRL-CAD:starseeker * 69212 brlcad/trunk/src/libbn/randsph.c: Preserve for the moment the code that uses sobol sequences instead of pseudorandom numbers. Sobol will not work by itself - we need a scramble to avoid patterning. |
| 18:00.15 | brlcad | starseeker: would it make sense for the new sobol functions to be in the/a private header? |
| 18:01.24 | brlcad | starseeker: I highly doubt we need anything nearly so complciated for quasirandom sampling |
| 18:01.43 | brlcad | we could even start with a non-random tessellation and get better results |
| 18:01.59 | brlcad | his. We probably need a |
| 18:04.50 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 18:08.03 | starseeker | brlcad: I'm yanking |
| 18:08.51 | starseeker | or rather, unhooking it from the build - will yank if it looks like it's ultimately not useful |
| 18:09.37 | Notify | 03BRL-CAD:starseeker * 69213 (brlcad/trunk/include/bn/rand.h brlcad/trunk/src/libbn/CMakeLists.txt and 4 others): shelve Sobol unless/until it looks like it will be useful. |
| 18:12.08 | Notify | 03BRL-CAD:starseeker * 69214 brlcad/trunk/src/libbn/randsph.c: note other paper |
| 18:17.22 | starseeker | brlcad: was looking for minimizing chances of duplicate numbers and faster convergence behavior - I'll concede it's not clear yet if those are needed |
| 18:17.30 | *** join/#brlcad Lord_of_Codes (~Lord_of_C@122.163.244.145) |
| 18:25.41 | *** join/#brlcad Lord_of_Codes (~Lord_of_C@122.163.244.145) |
| 18:52.29 | ries | ping brlcad |
| 19:50.18 | brlcad | starseeker: better than gqa is what I meant if unclear .. even pure random using our existing generators should work. what quasirandom offers is indeed faster convergence, but don't have that quantified and it's technically an optimization (so might as well do what is simpler and works first) |
| 19:51.46 | brlcad | starseeker: as for the scrambling, could we use another random number generator (with a seed) to seed it? |
| 19:51.52 | brlcad | ries: pong! |
| 19:52.19 | ries | brlcad: got a sec for a pm? |
| 19:52.27 | brlcad | sure |
| 19:57.41 | brlcad | starseeker: and my comment wasn't to yank ... working wins over complicated (and my assertion was not proven true, just a doubt) |
| 20:00.47 | brlcad | comment was more regarding bn public api, that it doesn't need to be first class functions ... just another random pattern generator where (whatever) is the implementation detail |
| 22:51.26 | *** join/#brlcad teepee_ (~teepee@unaffiliated/teepee) |