egg|nomz|egg changed the topic of #kspacademia to: https://gist.github.com/pdn4kd/164b9b85435d87afbec0c3a7e69d3e6d | Dogs are cats. Spiders are cat interferometers. | Космизм сегодня! | Document well, for tomorrow you may get mauled by a ネコバス. | <UmbralRaptor> egg|nomz|egg: generally if your eyes are dewing over, that's not the weather. | <ferram4> I shall beat my problems to death with an engineer.
e14159 has quit [Killed (NickServ (GHOST command used by e_14159!~e14159@dslb-094-216-202-099.094.216.pools.vodafone-ip.de))]
e_14159 has joined #kspacademia
e_14159 is now known as e14159
<egg|zzz|egg>
!wpn whitequark
* Qboid
gives whitequark a first/follow tea
<egg|zzz|egg>
!pet whitequark
* Qboid
pets whitequark
<egg|zzz|egg>
!choose sleep|df
<Qboid>
egg|zzz|egg: Your options are: sleep, df. My choice: sleep
<egg|zzz|egg>
;choose sleep|df
<kmath>
egg|zzz|egg: sleep
* egg|zzz|egg
zzz
<egg|zzz|egg>
!wpn bofh
* Qboid
gives bofh an ISO 9899 kukri
<egg|zzz|egg>
!wpn the котяchrome kitten
* Qboid
gives the котяchrome kitten a plasma scale
awang_ has quit [Ping timeout: 194 seconds]
<bofh>
https://doi.org/10.1093/jac/dky164 WELL THAT AIN'T TERRIFYING AT ALL ("Emergence of an XDR and carbapenemase-producing hypervirulent Klebsiella pneumoniae strain in Taiwan")
awang has joined #kspacademia
<UmbralRaptor>
Fun.
e14159 has quit [Killed (NickServ (GHOST command used by e_14159!~e14159@dslb-188-098-205-090.188.098.pools.vodafone-ip.de))]
<kmath>
<✔stuartcantrill> Today has been brought to you by the letter 'C'. C for clusterfuck.
<UmbralRaptor>
egg|zzz|egg: Random TRAPPIST-1 observation: getting to the nearest neighbors of the habitable ones (or really any) isn't much higher ΔV than getting from 🜨 to ♂.
<UmbralRaptor>
Though more widely separated ones can get, uh, interesting.
* UmbralRaptor
assumes that these numbers are correct.
<UmbralRaptor>
Also, they have tiny SOIs (4-20x the planet's radius).
ferram4_ has joined #kspacademia
<UmbralRaptor>
!acr -add:TNTC Too Numerous To Count
<Qboid>
UmbralRaptor: I added the explanation for this acronym.
ferram4 has quit [Ping timeout: 202 seconds]
awang has quit [Ping timeout: 202 seconds]
awang has joined #kspacademia
awang has quit [Ping timeout: 194 seconds]
<egg|zzz|egg>
UmbralRaptor: yeah, Дv is pretty big, you'd need RealSolarSystem to play there decently
<egg|zzz|egg>
um, that should be a Δ
<egg|zzz|egg>
UmbralRaptor: iirc there was a similar thing when we considered dreary kerbin with regex
<egg|zzz|egg>
UmbralRaptor: (that was same diametres as in stock KSP, realistic densities)
<egg|zzz|egg>
!wpn whitequark
* Qboid
gives whitequark a dye range
awang has joined #kspacademia
awang has quit [Ping timeout: 182 seconds]
<egg|zzz|egg>
bofh: okay but what language was the argument in
<egg|zzz|egg>
it tweets halfway between principia releases :-p
tawny has quit [Ping timeout: 194 seconds]
APlayer has joined #kspacademia
awang has joined #kspacademia
<egg|zzz|egg>
bofh: okay but how do you do simulated annealing on orbital elements to match transit timings anyway
<egg|zzz|egg>
bofh: what is your distribution for perturbing the system
awang has quit [Ping timeout: 198 seconds]
<bofh>
egg|zzz|egg: hrmmm. that's actually a damn good question, I was expecting it to be easy to acquire from the МCMC ensemble means but I don't think so anymore.
<egg|zzz|egg>
bofh: I don't know how MCMC works at all anyway >_>
<egg|zzz|egg>
bofh: nice video on crystals though ^
<egg|zzz|egg>
!wpn bofh
* Qboid
gives bofh a meow
<egg|zzz|egg>
is Qboid a cat
* bofh
meows back at Qboid
* UmbralRaptor
mrows at Qboid.
<UmbralRaptor>
"In the event that Fairfax County Government Offices close due to inclement weather, absentee voting will be cancelled."
<UmbralRaptor>
uh
<B787_300>
UmbralRaptor: ... how does the office being closed affect absentee ballots?
<UmbralRaptor>
B787_300: additional checking suggests that what they're calling absentee is actually early voting. So closed office means no in-person early voting that day.
* egg|zzz|egg
meows at UmbralRaptor
<B787_300>
Ah. That isn't absentee ballots tho
<B787_300>
Also you see that some politicians are complaining about FBs new political ad rules because they didn't follow the instructions that were sent out a month ago?
<bofh>
I literally just sat back down at a computer, I need to think of what the ensemble statistics are like here for a bit.
<bofh>
I'm used to fairly natural systems where you just have a Hamiltonian, but I don't think one for TRAPPIST-1 would be linear since it's classical
* bofh
takes the boundary of the polenta
* egg|zzz|egg
eats the interior
<egg|zzz|egg>
bofh: up to now we're just using that genetic algorithm to minimize the sum of squares of residuals on transit timings, with no regard to the error bars on the transit timng observations
<egg|zzz|egg>
bofh: and the L2 norm being the L2 norm, we end up with 5 min but a max transit timing residual of 2 h or something similarly daft
<bofh>
Like I feel like L^∞ is the norm you want here
<bofh>
Mind you, optimizing wrt. L^∞-norm is nasty, particularly when your problem doesn't have differential structure (OTOH, it feels like something like this *should* have enough for at least a Jacobian to be sensible, perhaps even a Hessian).
<egg|zzz|egg>
bofh: well, I'm not sure the uniform norm is appropriate since the observations actually have errors
<egg|zzz|egg>
so you'd have to take into account the error bars somehow
<bofh>
oh *yuck*
<egg|zzz|egg>
bofh: it's astronomy, not numerics :-p
<egg|zzz|egg>
bofh: maybe the probability that the transits on the computed transits fall into independent normal distributions corresponding to the observations?
<egg|zzz|egg>
is that how likelihoods work
<egg|zzz|egg>
am I going to summon bayesian demons
<egg|zzz|egg>
bofh: okay I managed to underflow things with my probabilities :-p
<bofh>
um
<bofh>
how? :P
<egg|zzz|egg>
bofh: so I'm trying things out, just using a 0.001 day blanket σ for now (data entry is not fun), and I get residuals to the tune of thousands of seconds on the initial guess, which is rather unlikely; multiply that by itself over the number of transits and it gets silly
<bofh>
Yeaaaaaaaaaaaaah that's going to be dynamically unstable
<egg|zzz|egg>
bofh: that's going to be 0, mostly :-p
<egg|zzz|egg>
bofh: is there a way to compute log erfc
<egg|zzz|egg>
has something gone horribly wrong if i'm trying to compute log erfc,
<bofh>
Not really? Done that before.
APlayer has quit [Ping timeout: 182 seconds]
<egg|zzz|egg>
bofh: okay but how do i do it
<egg|zzz|egg>
bofh: if this turns into spending another month writing an elementary function with exactly one usage I'll be slightly annoyed :-p
<bofh>
log(erfc(x)), both are standard functions in libm and the result *is* numerically stable
<bofh>
yeah, ideally you have erfc(x) = e^{-x*x}*erfcx(x), and the log of that winds up being log(erfcx(x)) - x*x, and erfcx(x) can be nicely implemented via Padé approximants (boost::math does this, you can just pull the P and Q polys out of the header and then feed them to Estrin). alternatively, you can use erfc(x) = 1-erf(x), erf(-x) = -erf(x), so erfc(-x) = 1-erf(-x) = 1+erf(x), so you compute
<bofh>
log1p(erf(-x)) if your erf(x) is defined correctly for ...
<bofh>
(Yes I know that file path from memory, don't ask >_>)
<egg|zzz|egg>
bofh: tbh I can probably get away with log erfc directly
<bofh>
You *should*, but you were complaining about underflow so I numerical analysis'd! :P
<egg|zzz|egg>
:D
<bofh>
(That reminds me, I need to actually get back to figuring out why those optimal guesses for your n'th root functions scarily well fit a linear relation, with weird oscillations in the values and relative error over the order).
<bofh>
(aaaaaaaaaaaaaaaaaaaaaaaaaaa so much to do)
<egg|zzz|egg>
bofh: also the Nice 69th root
<egg|zzz|egg>
(what are 69th roots for anyway)
<egg|zzz|egg>
bofh: also doubledoubledouble errors
* egg|zzz|egg
stares
<egg|zzz|egg>
lɒl it *does* underflow
<egg|zzz|egg>
okay let's try that log1p erf -x
<bofh>
right, that too, I meant to look at how to do error analysis for that and then conferences and then paper deadlines+uveitis combo
<bofh>
!u lɒl
<Qboid>
U+006C LATIN SMALL LETTER L (l)
<Qboid>
U+0252 LATIN SMALL LETTER TURNED ALPHA (ɒ)
<Qboid>
U+006C LATIN SMALL LETTER L (l)
<bofh>
rofl
<bofh>
Approval.
<egg|zzz|egg>
oh so there's a backstory to that
<egg|zzz|egg>
if you look at the banlist you'll see "ian"
<egg|zzz|egg>
there was somebody known as ian2
<egg|zzz|egg>
who was a general annoyance in #kspmodders back when that was a thing
<egg|zzz|egg>
and who terminated sentences by ". lol."
<X>
Omg. Lol
<egg|zzz|egg>
(not sure about that final full stop actually, might have been ". lol")
<X>
I wonder if he is still alive
<egg|zzz|egg>
bofh: anyway, at some point majiir (snoozee) got annoyed and bade one of his bots (maybe eye of jeb?) kick anybody who lolterminated a sentence, or maybe simply said lɒl
<egg|zzz|egg>
bofh: hence the workaround, my client has a macro to turn it to the appropriate IPA :D
<bofh>
NICE.
<bofh>
Also there's a bot named Eye of Jeb? lmao
<X>
O_o you are not the real bofh.
<egg|zzz|egg>
bofh: there used to be an Eye_of_Jeb
<egg|zzz|egg>
X: wait what
<X>
Heresy
<egg|zzz|egg>
who is the real bofh then
* egg|zzz|egg
is confused
<egg|zzz|egg>
am I the real egg
<X>
Le google
<X>
It is a thing. On the earlier internet.
<X>
It involves certain devices.
<X>
Some which have strong magnets.
<bofh>
I am not Simon Travaglia, no.
<egg|zzz|egg>
bofh: Eye_of_Jeb would stare at those who said mechjeb
<bofh>
I'm curious why someone *would* say mechjeb.
<egg|zzz|egg>
bofh: lamont works on doing a fancy guidance module in there, which may interface with principia in the future
<egg|zzz|egg>
Atlas-Centaur PEG, but I think he has moved on to much fancier algorithms
<egg|zzz|egg>
bofh: wait the log1p thing will underflow just as much, right?
<egg|zzz|egg>
since it will be log1p -1
<egg|zzz|egg>
(well not underflow, but still be useless)
<bofh>
Well no it won't, since there's no subtraction, assuming a numerically stable implementation of erf(x).
<bofh>
<shrug>
<bofh>
It's worth a quick try.
<egg|zzz|egg>
bofh: it won't underflow, it's just that the result of erf will be -1
<egg|zzz|egg>
with all other bits being far into the distance
<bofh>
Ahh.
<egg|zzz|egg>
bofh: okay how do i write a log erfc that's correct to a couple digits, I'm doing a genetic algorithm here, I don't need precision :-p
<egg|zzz|egg>
-2/√π (x+x²) :-p
<bofh>
I mean if you want more precision, http://csclub.uwaterloo.ca/~pbarfuss/erff.c look at the compute_erfcf path and split out the exponential appropriately, and use a power series or just log(erfc(x)) for the x < 1.5 case.
<bofh>
but honestly for a GA -2/√π (x+x²) should actually suffie