| 00:13.12 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 00:21.19 | *** join/#brlcad ries (~ries@D979C7EF.cm-3-2d.dynamic.ziggo.nl) |
| 00:41.34 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 00:51.34 | *** join/#brlcad teepee (~teepee@unaffiliated/teepee) |
| 00:55.08 | *** join/#brlcad ries (~ries@D979C7EF.cm-3-2d.dynamic.ziggo.nl) |
| 01:33.18 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 01:43.18 | *** join/#brlcad usybtveujxrllreu (~armin@dslb-088-065-186-000.088.065.pools.vodafone-ip.de) |
| 05:43.18 | *** join/#brlcad devi_ (73f8f961@gateway/web/freenode/ip.115.248.249.97) |
| 05:43.50 | starseeker | brlcad: I had first implemented the sph sampling with our exiting random number generator - that worked, but it looked as if there was some "clumping" so I investigated methods for making things a bit more evenly spread out |
| 05:44.14 | brlcad | starseeker: which one? |
| 05:44.15 | starseeker | that led to the sobol, but once I got that hooked up I could see the pattern in the generated points on the sphere |
| 05:44.38 | starseeker | randmt |
| 05:45.51 | starseeker | so looking for ways to fix that led to the Florida State report, but I don't have a license compatible implementation of that in C ready to hand |
| 05:46.10 | starseeker | so sobol in its current form is fairly useless |
| 05:46.42 | starseeker | but as you say, the randmt method should be "good enough" and is readily available |
| 05:47.32 | starseeker | the next step is to try hooking it up to the raytracer rather than fiddling with improving sph point generation... |
| 05:47.53 | starseeker | unfortunately I got interested ;-) |
| 05:49.26 | brlcad | I think it "can" be good enough for a first step, it'll probably just take 10x longer to converge a confidence interval below some tolerance |
| 05:51.18 | brlcad | so cool, that's awesome progress |
| 05:51.34 | brlcad | was it just the pattern that made sobol useless?? |
| 05:51.39 | starseeker | nods |
| 05:51.55 | brlcad | did it support seeding? |
| 05:51.56 | starseeker | it looks like its some sort of projection |
| 05:52.12 | starseeker | yes, but it was the way the points were generated that was the problem not the specific values |
| 05:52.26 | starseeker | visually it looked like a grid wrapped around a sphere |
| 05:52.35 | starseeker | not precisely, but something along those lines |
| 05:52.43 | brlcad | maybe in polor coordinates? |
| 05:52.50 | starseeker | possibly |
| 05:53.30 | starseeker | there was one point where a lot of points were clumping, like the parametric edge collapsed to a near-point there |
| 05:54.12 | starseeker | it's possible I did something wrong - I know there are a lot of subtle points to this |
| 05:54.33 | brlcad | sounds like something wrong |
| 05:54.47 | brlcad | here's something you might have better luck with that saucier wrote |
| 05:54.50 | brlcad | http://ftp.arl.mil/random/random.pdf |
| 05:55.50 | brlcad | it's a c++ class that should be easily wrapped |
| 06:03.12 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 06:10.40 | starseeker | brlcad: what I've got so far is in libbn/randsph.c and libbn/tests/randsph.c if you want to see - the sobol bit wouldn't be hard to turn back on, but at the moment it's all commented out |
| 06:11.06 | starseeker | I need a better test viewer - right now I'm dumping in commands for spheres, which doesn't scale well |
| 06:11.28 | starseeker | 50,000 in cmds for spheres makes MGED sad, apparently... |
| 06:13.06 | starseeker | calls it a night |
| 06:17.43 | brlcad | starseeker: nifty snippet I had stashed: http://brlcad.org/~sean/tmp/randompoints.html |
| 06:18.08 | brlcad | basic quasirandom in like 10 lines of code |
| 06:24.37 | brlcad | and the way the avoidance works, it will work incrementally (it'll just get slower/harder to get more points as the density increases) |
| 06:26.32 | brlcad | for testing, should use the pnts primitive... perfect for this |
| 06:34.57 | brlcad | hm... documenting the points primitive will make for a good gci task... :) |
| 06:40.08 | *** join/#brlcad amarjeet (~amarjeet@202.164.53.117) |
| 06:42.42 | *** join/#brlcad Lord_of_Codes (~Lord_of_C@122.163.244.145) |
| 06:55.05 | *** join/#brlcad teepee_ (~teepee@unaffiliated/teepee) |
| 07:54.52 | *** join/#brlcad caen23 (~caen23@79.112.95.83) |
| 10:02.40 | *** join/#brlcad teepee] (bc5c2133@gateway/web/freenode/ip.188.92.33.51) |
| 11:31.19 | *** join/#brlcad Lord_of_Codes (~Lord_of_C@122.163.244.145) |
| 11:33.08 | *** join/#brlcad Caterpillar (~caterpill@unaffiliated/caterpillar) |
| 12:15.43 | *** join/#brlcad yorik (~yorik@2804:431:f721:47e1:290:f5ff:fedc:3bb2) |
| 12:34.52 | *** join/#brlcad MikeH (~Mike@188.175.158.32) |
| 12:40.03 | *** join/#brlcad Lord_of_Codes (~Lord_of_C@122.163.244.145) |
| 13:22.33 | *** join/#brlcad Lord_of_Codes (~Lord_of_C@122.163.244.145) |
| 13:25.51 | starseeker | brlcad: the bit about getting slower with more points added was one of the things I was trying to avoid - my intuitive expectation is that these methods may require a *lot* of points to converge... |
| 13:26.43 | starseeker | pnts primitive would work execept IIRC it doesn't have an in cmd implementation... maybe I'm misremembering |
| 14:42.00 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 14:57.49 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 15:16.41 | Notify | 03BRL-CAD:starseeker * 69215 brlcad/trunk/include/bn/rand.h: start working on a design for a generic API that can subsume the various possibilities for number sequence generation. |
| 15:17.53 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 15:21.09 | starseeker | brlcad: comments appreciated on r69215 - those are my initial thoughts after the initial quasi-random sph pnt work on the "desired" API for libbn |
| 15:26.30 | Notify | 03BRL-CAD:starseeker * 69216 brlcad/trunk/include/bn/rand.h: Make some rework/TODO notes for this function |
| 16:24.50 | *** join/#brlcad kintel (~kintel@unaffiliated/kintel) |
| 16:55.45 | *** join/#brlcad Caterpillar (~caterpill@unaffiliated/caterpillar) |
| 17:47.31 | *** join/#brlcad Lord_of_Codes (~Lord_of_C@122.163.244.145) |
| 21:09.42 | Notify | 03BRL-CAD:starseeker * 69217 (brlcad/trunk/include/bg/CMakeLists.txt brlcad/trunk/include/bn/CMakeLists.txt and 3 others): start stubbing in bits of a proposed generic pseudo/quasi random number sequence API. |
| 22:21.18 | *** join/#brlcad teepee (~teepee@unaffiliated/teepee) |