--- Log opened Mon Oct 03 00:00:36 2022
05:12 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::1909] has joined #hplusroadmap
05:19 < juri_> docl: at the moment, i can't predict the size of the circle required based on the noise in question.
06:33 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
07:41 < kanzure> video about ASML https://www.youtube.com/watch?v=zQu_TMgHO98
07:41 < Muaddib> [zQu_TMgHO98] ASML's Secret: An exclusive view from inside the global semiconductor giant | VPRO Documentary (53:09)
07:49 < kanzure> https://github.com/Cryptogenic/PS5-4.03-Kernel-Exploit
07:51 < kanzure> "Protein structure generation via folding diffusion" https://arxiv.org/abs/2209.15611
07:53 < kanzure> hashcash https://blog.lopp.net/protect-contact-forms-from-spam-with-proof-of-work/
07:59 < kanzure> metal seedplate flamethrower testing (for data archival) https://www.youtube.com/watch?v=XWfTyGpNXxM
07:59 < Muaddib> [XWfTyGpNXxM] A Treatise on Metal Bitcoin Seed Storage Design (29:37)
08:58 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has joined #hplusroadmap
09:01 < docl> juri_: is this a failure case? https://enkimute.github.io/ganja.js/examples/coffeeshop.html#bf5qID6Vx
09:36 -!- spaceangel [~spaceange@ip-78-102-216-202.bb.vodafone.cz] has joined #hplusroadmap
09:58 < juri_> that's the one from my posting, yes? all of the cases that are output by the test program are failure cases.
09:58 < juri_> ... i'm still working on getting successes out of the program.
10:20 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has quit [Ping timeout: 265 seconds]
11:17 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
11:18 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
11:39 < kanzure> https://patrickcollison.com/fast
13:20 < docl> juri_: it looks like d,e,f are 3 points with some error. you can make a circle from any 2 points by making that the diameter. so with 3 possible combinations, you have 3 possible circles. you could then loop through them and pick the first one that also contains the remaining point
13:23 < docl> given the situation where one point is far from the other two, it's going to be a bit closer to one than the other. a circle that includes the more distant one should include the closer one
13:30 < docl> actually I think that just boils down to picking the longest line and making a circle with that diameter. no way for a point closer to either of the other two to not land inside
13:38 < docl> I think also the longest line needs to include the radius of the error circle for each of the two points
13:39 < docl> so you make 3 lines from the points, add the radius of the error circle to each, pick the longest one, then draw the final circle around that
14:10 < juri_> yeah, but this should NOT work in other cases. the only reason this is true is because this is a triangle. squares work, other polygins work, but irregular ones don't.
14:10 < juri_> I've pulled reports for the passing ones now, and am working on getting those into gnuplot, to analyse.
14:15 < muurkha> squares are kind of the best case
14:15 < juri_> yep, but i'm pretty sure they'll fail, due to the same problem.
14:16 < muurkha> probably, please forgive me for not having attempted to read Haskell
14:17 < juri_> actually, no, it won't, because i should detect it as two lines, which by definition will always intersect in just one place.
14:17 < juri_> but that's a hack. a regular pentagon will fall apart.
14:18 < juri_> i'll turn this into a dataset, and dig further.
14:20 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Quit: night vision in the city is like water breathing in the desert | ncog libertas H. domesticus sedentaris]
14:49 -!- spaceangel [~spaceange@ip-78-102-216-202.bb.vodafone.cz] has quit [Remote host closed the connection]
14:55 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
16:44 < docl> juri_: I'll have to think more about how to apply it to other polygons. triangles are a bit special :)
16:46 < docl> ideally one might translate the polygon to a set of triangles. but we don't just want a big circle around the polygon
17:58 < juri_> so, keep in mind, this works for things that are not closed, as well. as long as the bisectors of the corners reach the same place, i can draw a straight skeleton with a node at that place.
17:58 < juri_> i need a math solution for this, not a geometric one. :D
17:59 < muurkha> what do you think about interval arithmetic?
18:02 < juri_> i'm sortof doing something like that, by doing ULP tracking. its not exactly interval arithmatic, but it gets me no closer to the question of what i do with that interval.
18:03 < muurkha> hmm, I guess I thought that would solve your problem, so I guess I don't actually understand the problem
18:04 < muurkha> which is, you know, maybe my issue and not yours
18:05 < juri_> so the problem is, what do you do with the interval when you're done? i've tried the dumb thing, of just summing them, and using that to decide on the size of the error-circle. that works in a lot of cases, but falls apart about 1/30th of the time.
18:05 < muurkha> but I thought you wanted "the error range of really steep line intersections" or in more detail "how big a circle i have to draw around the intersection of two lines, to guarantee that the real answer is somewhere in the circle" and that's almost exactly what interval arithmetic gives you
18:05 < muurkha> it sounds like you're calculating the interval in a way that is not guaranteed to be conservative
18:06 < muurkha> that is, you intended to calculate a conservative approximation to the true error bound, but sometimes the real error is greater than the error you calculated
18:06 < juri_> take that, and multiply it by the sin of the angle of the intersection times 10,000? still wrong (once every few tens of thousands of times). and its clear that there should be a formula here, not a hard number.
18:07 < juri_> i think the problem is my error is not linear, when you're supposed to interpret it. the steeper the angle of the intersection, the more wrong it is.
18:07 < muurkha> indeed
18:07 < muurkha> adding two errors together does not give you the right error, it's true
18:08 < juri_> i don't know why it's not linear, and i don't know what i should be doing yet... so i'm taking a dump of all of the successes, and pulling them into gnuplot.
18:08 < juri_> i'm betting there's some curve i can't see in the raw numbers.
18:09 < muurkha> I linked you to some code I wrote in 02014 that has formulas that would give you the right answer if calculated with the proper rounding modes
18:10 < juri_> yeah, i still don't know what to do with those numbers, tho.
18:10 < muurkha> well, if they are correctly a conservative approximation of the true answer, you can just use the smallest circle that encloses the paraxial box they outline
18:10 < muurkha> (though if you look at the git log you'll see I wrote the first few versions while logged into Darius's account on my netbook!)
18:10 < juri_> i'm relatively sure my implementation of interval logic, while not perfect (it assumes 1ULP when the cpu makes 0.5ULP guarantees), i think it's doing the job.
18:11 < muurkha> yeah, I don't think the problem is that you're estimating the rounding errors that arise in each calculation incorretly
18:11 < muurkha> *incorrectly
18:11 < muurkha> it's that you're propagating rounding errors from previous calculation steps incorrectly, from what you say
18:12 < juri_> actually, i'm preserving each step's error in one big structure, so i have the error of each calculation available, when drawing my error circles.
18:13 < muurkha> I am trying to understand the code but unfortunately for me it is in Haskell
18:14 < muurkha> (which is not to say that Haskell is a bad tool for this kind of thing of course, but you might need someone smarter than me to help you as a result)
18:15 < juri_> it is tearse. :)
18:15 < juri_> part of why i write it. my hands aren't what they used to be.
18:15 < muurkha> yeah :(
18:16 < muurkha> "having the error of each calculation available" sounds like affine arithmetic, which is an approach that can give you tighter error bounds than regular interval arithmetic at much greater computational cost
18:17 < muurkha> amusingly I already have a clone of HSlice from more than a year ago
18:17 < juri_> it's changed quite a bit. ;)
18:18 < muurkha> etags doesn't do anything useful with Haskell files it seems
18:19 < juri_> ok, bedtime for me.
18:19 < juri_> thanks for looking. :D
18:20 < muurkha> I finally found the definition of PLine2Err!
18:21 < muurkha> sorry I couldn't be more help
18:24 < muurkha> I was expecting something recognizable as x and y coordinates, or x,y,z, and maybe a representation of a line as a start point and a direction, or a start point and two normal vectors, or something
18:25 < muurkha> I haven't been able to find any of those things
18:26 < juri_> its X, Y, and T for lines, and X, Y, and S for points. this is all projective geometric algebra. see: siggraph2019 geometric algebra talk.
18:26 < muurkha> hmm, it looks like maybe lines are represented as their x-intercept and their y-intercept?
18:27 < muurkha> hmm, thanks, I'll tryto find the talk
18:39 < muurkha> .t https://arxiv.org/abs/2002.04509
18:39 < saxo> [2002.04509] Course notes Geometric Algebra for Computer Graphics, SIGGRAPH 2019
18:44 < muurkha> I uh
18:44 < muurkha> think it's going to take me longer than a couple of hours to assimilate this
18:47 < muurkha> it looks pretty freaking sweet though
18:48 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::1909] has quit [Quit: Leaving]
18:52 < docl> I'll uh, see what I can get out of it as well... :)
18:59 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 265 seconds]
19:03 < muurkha> I like algebraic approaches to things, and this is an algebra that embraces (in 3-D) scalars, points, lines, planes, and something called "4-vectors", and defines addition, multiplication, meet, join, integer powers (defined from multiplication), the exponential function (defined by its power series), and I think some kind of multiplicative inverse operation
19:07 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 268 seconds]
19:13 < kanzure> david wood tries to rebrand transhumanism https://futuresurge.org/
19:23 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
19:34 < muurkha> this poster looks pretty nice, docl: https://projectivegeometricalgebra.org/projgeomalg.pdf
19:34 < docl> thanks!
19:47 < muurkha> although... the version in https://arxiv.org/abs/2002.04509 is a lot simpler
19:59 < docl> I'm sort of flipping between the two. interesting vocabulary for thinking about points/objects in space
20:02 < docl> I'll probably have to read through multiple times, but I'm anticipating it being highly productive for me
20:21 < muurkha> this might be a little faster read than the usual sort of math paper, which usually goes at about a page a day
20:21 < muurkha> this might go several pages a day
21:09 < lsneff> well, I just an offer to stay here instead of finishing school
21:10 < lsneff> goddammit, *just got an offer to stay here
21:10 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Read error: Connection reset by peer]
21:10 < nmz787> lol
21:10 < nmz787> will they pay for your classes?
21:11 < nmz787> one at a time until you're done
21:11 < nmz787> Intel does that as long a you get at least a B
21:32 < muurkha> lsneff: I kind of regret having done that but also I wasn't at SpaceX
21:34 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
21:59 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Read error: Connection reset by peer]
22:18 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
23:58 -!- codaraxis___ [~codaraxis@user/codaraxis] has joined #hplusroadmap
--- Log closed Tue Oct 04 00:00:37 2022