UmbralRaptor 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. | We can haz pdf
<SnoopJeDi>
not sure if murder or murder
<raptop>
Self-gravitating crows
<raptop>
Anyway, how's industry treating you?
<SnoopJeDi>
!wpn -add:wpn crow
<galois>
Entry already exists!
<SnoopJeDi>
raptop, tensorflow is godawful, but the job is cool
<SnoopJeDi>
happily we are moving away from tf because of the way it is
<SnoopJeDi>
but my current task is "here's a pile of research code by the same authors who wrote the net we use for video segmentation, can you evaluate if we should switch, and if so, make the necessary adaptations?"
<SnoopJeDi>
meanwhile, in tensorflow: tensorflow.contrib.slim.python.slim.data.dataset
<SnoopJeDi>
.Dataset
<raptop>
that's a lot of sublevels
<SnoopJeDi>
bonus points: tensorflow.contrib.slim is not the only version of slim this model needs to run, it also depends on a not-released version which isn't in a package
<SnoopJeDi>
so the official Google instructions tell you to clone a 2 GB repo and modify PYTHONPATH to get a 3 MB module 🙀
<raptop>
aaaaaa
<SnoopJeDi>
happily performing the necessary modulectomy was not too hard, but...it's still tensorflow :(
<SnoopJeDi>
just to properly evaluate the thing I've got to rig up some clumsy roundtrip through protobuf
<raptop>
Does that summon the wrath of egg?
<SnoopJeDi>
I'm not sure what he'd make of it
Iskierka has quit [Ping timeout: 190 seconds]
<SnoopJeDi>
the protobuf structures themselves are at least sensible, but I don't see the point of using it for the input, *particularly* if you want to run your nets on arbitrary data and not [so and so dataset] that you need to test on to get published
<SnoopJeDi>
maybe it makes sense if you have streaming input though? idk, it's like drinking from a firehose right now, the API is nowhere intuitive and everywhere stateful
<SnoopJeDi>
raptop, the canonical way to modify variables is to do things like `with tf.Session() as sess: foo = ...; bar = ...; tf.global_variables_initializer()` or something along those lines, where closure state implicitly determines the computation graph :(
Iskierka has joined #kspacademia
<SnoopJeDi>
or semi-implicit I guess. A lot of the cruft is hiding the sausage-making, which seems like a stupid thing to do for a library aimed at people researching how to make better sausages
<SnoopJeDi>
but I think the 1.x API was less designed and more...accreted
<SnoopJeDi>
no idea about 2.x but I don't have high hopes
<raptop>
pebble accretion as applied to APIs
<SnoopJeDi>
raptop, thanks for asking, though! 💙 how are the galeggsies?
<raptop>
erm
<raptop>
The galaxy stuff was undergrad, and I really should do a blogpost at some point about why it didn't work.
<egg|cell|egg>
Aren't they eggsoplanets now
<SnoopJeDi>
oh, right!
<raptop>
As far as eggsoplanets go, cool simulations with intriguing outcomes that I haven't looked into in detail. Also on hold until at least the afternoon of 2020-01-15
<SnoopJeDi>
a blog post is a great format for that kind of post-mortem imo, and more practice writing is always neat
<SnoopJeDi>
raptop, remind me of the core question(s) under investigation, again?
<raptop>
For the galaxies, what we could do to better find interaction features and morphology vs color in SDSS
<raptop>
For planets, survey simulations and estimating yield with different speggtrographa nd telescope combinations
raptop has quit [Quit: connection interrupted by bus]
<egg|cell|egg>
Actual bus?
<SnoopJeDi>
LOL
<SnoopJeDi>
!wpn -add:wpn bus
<galois>
Added wpn 'bus'
e_14159_ has joined #kspacademia
e_14159 has quit [Ping timeout: 378 seconds]
<UmbralRaptor>
Well, catching a bus
<SnoopJeDi>
bus confirmed
<SnoopJeDi>
!wpn -add:adj confirmed
<galois>
Added adj 'confirmed'
<SnoopJeDi>
!wpn -add:wpn debugger
<galois>
Entry already exists!
<UmbralRaptor>
!wpn -add:wpn compiler
<galois>
Entry already exists!
<SnoopJeDi>
!wpn -add:wpn closure
<galois>
Added wpn 'closure'
<UmbralRaptor>
!wpn -add:wpn linter
<galois>
Entry already exists!
<SnoopJeDi>
!wpn -add:wpn modal
<galois>
Added wpn 'modal'
<UmbralRaptor>
!wpn -add:wpn linker
<galois>
Entry already exists!
<SnoopJeDi>
!wpn -add:wpn widget
<galois>
Added wpn 'widget'
<SnoopJeDi>
!wpn -add:wpn slider
<galois>
Added wpn 'slider'
<SnoopJeDi>
!wpn -add:wpn button
<galois>
Added wpn 'button'
<UmbralRaptor>
!wpn -add:wpn toggle
<galois>
Added wpn 'toggle'
<SnoopJeDi>
!wpn -add:adj out-of-bounds
<galois>
Added adj 'out-of-bounds'
<UmbralRaptor>
!wpn -add:wpn dial
<galois>
Added wpn 'dial'
e_14159 has joined #kspacademia
e_14159_ has quit [Ping timeout: 378 seconds]
SilverFox has quit [Read error: Connection reset by peer]
<galois>
title: Shared album - Patrick N - Google Photos
uovo has quit [Ping timeout: 378 seconds]
egg has joined #kspacademia
oeuf has joined #kspacademia
egg has quit [Killed (NickServ (GHOST command used by oeuf))]
oeuf is now known as egg
oeuf has joined #kspacademia
<egg>
whitequark: UmbralRaptor: I've seen ice being used to try to rescue a mayonnaise that had trouble emulsifying
egg|cell|egg has quit [Ping timeout: 202 seconds]
egg|cell|egg has joined #kspacademia
<UmbralRaptor>
hrm
<_whitenotifier-d13c>
[Principia] pleroy opened pull request #2437: Change the ReadFromMessage method of the adaptive-step size integrators to be static - https://git.io/Jvfcc
<_whitenotifier-d13c>
[Principia] pleroy synchronize pull request #2437: Change the ReadFromMessage method of the adaptive-step size integrators to be static - https://git.io/Jvfcc