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.
<kmath>
<AWESoMeStarsERC> 3D printing outreach materials - work done by summer student Dean Patient and PhD student @AdamF_Astro! #scicomm https://t.co/BNqqnToQne
e_14159_ has joined #kspacademia
e_14159 has quit [Ping timeout: 198 seconds]
<rqou>
bofh: ping?
egg|phone|egg has quit [Ping timeout: 186 seconds]
egg|phone|egg has joined #kspacademia
<egg>
rqou: bofh answered in the backlog
<rqou>
yeah i saw that
<rqou>
i wanted to discuss it further
<egg>
rqou: bofh is likely asleep
<egg>
it's 8:11
<rqou>
ah, too early :P
<bofh>
rqou: pong
<rqou>
ah, good morning
<rqou>
anyways, i should probably give a more complete background of what i'm trying to actually achieve
<bofh>
yes, that'd upon further thought prolly be helpful, heh.
<rqou>
the original problem is "perform FPGA cell placement in an analytic manner (i.e. not simulated annealing)"
<rqou>
there are a number of papers describing algorithms to do this, but they all vaguely work like this:
<rqou>
fix some cells at some fixed location and then try to minimize the length of wires connecting the cells using some kind of math
<rqou>
"fix" the answer that the math returns so that the cell locations are all legal
<rqou>
repeat this until some convergence criteria happens
<rqou>
so now more specifically
<rqou>
most algorithms of this type use various ways to approximate the length of wires between cells, and they usually do so in such a way as to lead to a QP problem
<rqou>
the problem with just applying QP is that it tends to cause a very dense cluster of overlapping cells somewhere "in the middle" of the array
<rqou>
so the fix is that they then apply various "spread the cells out" algorithms to this
<rqou>
and then they add pseudo anchors (that are not movable) to the cells at each of the new locations after applying this fixup
<rqou>
and then QP is applied again
<bofh>
so my first thought is: how much faster/more efficient are QPs compared to simulated annealing for this problem? SA has the nice property of eggstremely low per-iteration cost, so it *may* actually be competitive. Assuming it's not...
<bofh>
uhh
<rqou>
it's not clear
<rqou>
supposedly SA is slower for larger designs and chips
<bofh>
this sounds eggstremely hacky; why not add a minimum spread to the original QP objective function?
<rqou>
afaik it's because doing so is hard? :P
<rqou>
one problem is that the final result needs to be in integers
<rqou>
also, various algorithms use the "spreading out" phase to apply extra hacks for additional objective functions :P
<rqou>
such as estimated routing congestion
<bofh>
oh, so you actually have a quadratic IP. yecch.
<rqou>
oh yeah, the overall problem is super-np-hard and people are just coming up with different sets of approximations for it
<rqou>
ok, one paper i read claims that the reason proper constraints aren't actually used in the QP problem is because it makes the problem not convex and/or make the objective function not differentiable
<rqou>
so hacks are used instead :P
<rqou>
bofh: anyways, as for why conjugate gradient: i am not at all familiar with the problem domain and some papers i read said that they did that :P :P :P
<bofh>
well re: super-np-hard: duh, but the right approximation often can have a very good approximation factor
<rqou>
yeah, and that's why i'm reading a bunch of papers and just copying what they did :P
<bofh>
rqou: so conjugate gradient is what is commonly used due to it being eggstremely simple to implement.
<bofh>
(it's kinda like the RK4 of ODE solvers in that regard)
<rqou>
lol i'm not familiar with ODE solvers either :P
<rqou>
i just ask numpy/scipy to give me the answer and hope it's correct :P :P
<rqou>
(this has run into problems before. e.g. sliding mode control (explicitly has a step discontinuity) tends to make the solvers really really sad)
<rqou>
bofh: as to why i was considering fixed-point: i was scared of floating point bugs and figured that the final answer needed to be integers anyways
<rqou>
seems like that's not a good tradeoff
<bofh>
erm, that doesn't help you with actually getting an integer answer from an IP, you have to actually do that carefully (okay you can often use the bruteforce method, i.e. Chvátal-Gomory cuts, but those are slow as hell generally)
<bofh>
it's fairly hard to make conjugate gradient do the wrong thing just due to floating-point bugs, tbh.
<rqou>
what about "dumb" bugs like "is this in X=2 or X=3? why not both?"
<rqou>
as long as i do all the float->int in one place the same way this should not happen right?
<bofh>
okay, so an IP is not what you think it is. integer programming as-is is very NP-hard, so what is done is you solve the *LP/QP relaxation of it* (i.e. tossing out the constraint that your solution be an integer), solve *that* since it's very easy, and then carefully round your solution to *the nearest integer solution that is within the feasible region of your original problem*. wikipedia has a very
<bofh>
simple example why this is a harder problem than ...
<kmath>
<henryfhchan> @celestialweasel @Kazinsal @mjg59 With a total repertoire of 80,000 CJK characters (and growing), many of which sol… https://t.co/QmaRXUv3b6
egg|cell|egg has joined #kspacademia
egg|phone|egg has quit [Read error: -0x1: UNKNOWN ERROR CODE (0001)]
<rqou>
i'm vaguely aware of IVD
<egg>
rqou: that's an interesting thread on the subject ^
<rqou>
nice that they're finally fixing it like a decade+ too late
<egg>
rqou: the thing is fixing it required massively broadening "what is plain text", as usual
<egg>
as the thread says, this used to be a font issue on top of pre-10646/Unicode encodings
<rqou>
yes, i vaguely remember even magic fonts that could "convert" between simplified and traditional
<egg>
but now plain text can come with selectors that say "no here it really matters that you use *this* variant for orthographic reasons"
<kmath>
<whitequark> "Anybody who manufactures [individually wrapped sliced cheese] is familiar" ah, finally, a sentence to match math t… https://t.co/aqKMe31Bom
<egg>
bofh: "Apparently the consumer neither knew nor cared that these so-called slices were not actually sliced off a loaf."
<bofh>
ROFL
<bofh>
I mean, I figured as much, pre-sliced cheese is far too smooth on its edges to have been sliced.
<bofh>
h
tawny has quit [Ping timeout: 198 seconds]
APlayer has joined #kspacademia
<egg>
!meow bofh
* Qboid
meows at bofh
<bofh>
!meow egg
* Qboid
meows at egg
<egg>
!wpn bofh
* Qboid
gives bofh a Galois INTERCAL genome
<egg>
UmbralRaptop: bofh: ANBOcat was chasing a shrew inside, so I moved the shrew outside to avoid having to clean up shrew parts; the cat seems not to have understood that the shrew had been moved though, and was confused
<egg>
!wpn whitequark
* Qboid
gives whitequark a Radon quokka
<bofh>
egg: amused that a shrew got inside in the first place, tbqh.
<egg>
maybe aided by the cat?
<bofh>
oh, that makes sense, prolly trying to hide from it.
<whitequark>
very shrewd
<APlayer>
Can I assume two interplanetary departure trajectories to be reasonably similar (i.e. will arrive at target with not-too-extreme corrections either way) if the hyperbolic excess velocity (and direction) as well as the inclination are the same?
<APlayer>
(@egg in particular)
<APlayer>
What I am looking at exactly is estimating a departure burn as an instantaneous change of velocity, while in reality, it will be carried out as a spiralling slow burn, and the final periapsis may have a different altitude
<UmbralRaptop>
egg: are you sure that ANBOcat wasn't trying to teach you to hunt?
<kmath>
<gvkonsta> Nells also seems to deal with the Helsinki heatwave by being even more of a derp. She’s just hanging out in this po… https://t.co/1OdKJuONUL
pthag has joined #kspacademia
<awang>
Never mind, turns out I still have the frame in my backpack
<awang>
Looks like I forgot to take it out
<awang>
whitequark: Looks like the frame is stainless steel
<awang>
idk what kind specifically
<whitequark>
stainless steel is kind of a pain to solder
<whitequark>
it's probably 304 or 316
<whitequark>
lemme think about it
<whitequark>
I think your best bet is brazing with silver solder and a small torch.
<whitequark>
you don't need flux because silver solder already comes with some sorta flux.
<whitequark>
lemme grab some amazon links
<whitequark>
wtf
<whitequark>
awang: what's your locale?
<whitequark>
can you find an ACR shop nearby?
<whitequark>
amazon doesn't have any decent solder of the kind I have in mind
<whitequark>
nor any of the fluxes I'm looking for
<awang>
whitequark: Ohio
<awang>
ACR?
<awang>
Oh wow
<whitequark>
aircon/refrigeration
<awang>
That sounds quite a bit more involved than I expected
<awang>
Hmmm
<awang>
We have two home improvement stores nearby, but guess those aren't sufficiently specialized to stock what you're thinking of?
<egg>
UmbralRaptop: should I try some more imaging at 40ish % cloud cover, and what should I look at
<UmbralRaptop>
maybe
<UmbralRaptop>
Mars?
tawny has joined #kspacademia
<UmbralRaptop>
!wpn -add:adj 511 keV
<Qboid>
UmbralRaptop: Adjective added!
<UmbralRaptop>
egg: depends on cloud locations. I'm also going to suggest eagle nebula, omega nebula, and wild duck cluster.
<rqou>
wait, my friends and i are not the only one who use the "home despot" joke?
<UmbralRaptop>
rqou: I thought it was ubiquitous in the US.
<SnoopJeDi>
I've not heard it before
* UmbralRaptop
was still somewhat surprised to see colorlessfermion use the term.
<whitequark>
i picked it up from bofh
<awang>
UmbralRaptop: Why 611 keV specifically?
<awang>
s/6/5
<Qboid>
awang meant to say: UmbralRaptop: Why 511 keV specifically?
<UmbralRaptop>
awang: blame the mass of the electron, and the typical energy of a photon from the electron-positron annihilation reaction.
<SnoopJeDi>
e- + e+ -> γ + γ
<UmbralRaptop>
Incidentally, there's "50 Ω" in there for some reason.
<UmbralRaptop>
*there as in Qboid.
<egg>
UmbralRaptop: hm, those might be treed for a bit
<bofh>
egg: I mean Mars is at closest Earth approach until like 2287 tonight & bright as shit; seems like the obvious target imho :p
<iximeow>
it's also a reddish orange dust ball :|
<egg>
UmbralRaptop: how about Cas/And/Lac
<UmbralRaptop>
The dust is fading, right?
<egg>
bofh: what iximeow said, it's kind of boring tbh :-p
<UmbralRaptop>
BL Lac!
<UmbralRaptop>
(actually, don't. It's only interesting for studying astrophysics)
<egg>
what's BL lac
<UmbralRaptop>
Blazar with an infamously featureless spectrum.
<egg>
cat is running around
<UmbralRaptop>
cat!
<egg>
ohai ISS
pthag has quit [Remote host closed the connection]
egg|phone|egg has joined #kspacademia
egg|cell|egg has quit [Read error: Connection reset by peer]
* egg
pokes seeing with a stick
<egg>
UmbralRaptop: jupiter is supposed to be potato-shaped most of the time right,
<UmbralRaptop>
Jupiter, Phobos, close enough.
<egg>
also there are no stars in the field so that's unstackable anyway
* egg
tries deep-sky things instead
UmbralRaptor has joined #kspacademia
UmbralRaptop has quit [Ping timeout: 202 seconds]
<bofh>
UmbralRaptor: yes but presumably ome can't achieve Jovian escape velocity just by tripping over your feet & falling?
* UmbralRaptor
looks around for the nearest mi-go to question.
* egg
stares at seeing
<egg>
UmbralRaptor: so exposures longer than 10 s don't actually eggsist, right? >_>
<bofh>
egg: wouldn't that just completely saturate the sensor?
<egg>
the sky is dim
<egg>
i'ts more that tracking and seeing issues would yield oddly shaped stars :-p
<UmbralRaptor>
egg: depends on target, etc.
<UmbralRaptor>
(er, bofh also)
* UmbralRaptor
wonders why this phone keeps on thinking by "bofh" I mean "bodhisattva"
<SnoopJeDi>
what's the difference?
<SnoopJeDi>
(I just yesterday read Hesse's "Siddhartha")
APlayer has quit [Ping timeout: 383 seconds]
* egg
pokes approaching nebulosity with a stick
<egg>
UmbralRaptor: there is a great nebulosity in andromeda
<UmbralRaptor>
Is the nebulosity metal rich?
<egg>
probably
<egg>
also optically thick
<egg>
also it covers Cas, too,
<egg>
it makes observing the better-known great nebulosity in andromeda a bit tricky,
<egg>
also skycircle is rising
<UmbralRaptor>
Are Her or Cyg clear?
<egg>
Her is underground
<egg>
cyg might be clear but I'm not sure it would be by the time I'm done finding anything up there
<egg>
(I could install the diagonal focuser on the finder but recollimating the focuser would also lose time, and without diagonal looking up is tedious
<egg>
also tbh there are a lot of great nebulosities
<egg>
Mirach is clear again though
<egg>
argh of course now the nebulosity is back on top of the nebulosity as soon as i start eggsposing
<bofh>
this is all sounding eggstremely nebulous, tbh.
<UmbralRaptor>
I'd say that this was why some places got into radio astronomy, but it only explains the UK. Australia, however…
<kmath>
<sjs917> Gosnell: Sub-subgiant stars are stars that fall between the subgiant sequence and the main sequence. They tend to d… https://t.co/ZyUUMfcEnP
* awang
hochos Excel
<awang>
Why would zoom be an undo-able thing >:(
<awang>
And why does "redo" keep redoing past the top of the undo stack
awang has quit [Ping timeout: 182 seconds]
* egg
takes some darks
<egg>
!choose processing time|zzz time
<Qboid>
egg: Your options are: processing time, zzz time. My choice: processing time