logs archiveBotHelp.net / Freenode / #ada / 2015 / August / 14 / 2
charlie5
he is not too chatty here ... he gave #ada a good rap on cla a while back tho
(Action) was awful tempted to chip into prizemoney
... then checked wallet
Visaoni
I don't know, he's seemed pretty chatty while i've been around
charlie5
next time tho ... $100 is a small price to pay for a dozen or so ada videos
heh, maybe it's just me
Visaoni
eh, I think it's probably a timezone thing
charlie5
:) ... that's often the case
(here)
(Action) waiting for gnat_util to build to try to fix some prob reported on cla
soooo ... absolutley no one here uses archlinux ? ... or have fled from the OS due to my pkgs failing miserably ?
Visaoni
I do
it's just on my tiny little chromebook that I avoid writing code on as muc has possible :)
charlie5
(Action) lacks the appropriate emoticon
it's sort of discouraging
Visaoni
if it helps any, once school starts up again I'll probably be spending a lot more time on it
but yes, I can see that
charlie5
you doing ada at uni ?
Visaoni
not right now. if I can slip it in, I will
charlie5
nm, i've probly asked b4 :)
well, if you're trying it of your own accord, then all the better
Visaoni
hopefully my advisor will be open to it
charlie5
heh, tho you seem to have jumped into the 'deep' end
Visaoni
I do tend to do that...
charlie5
mm, well
AayJay
I used Arch before, but got tired of constantly fixing things...
charlie5
gnat_util buitl fine in archy
not sure what i'm sposed to tell the bloke who had prob on cla :/
if i had a few more data points would be simpler :0
:)
atm, my hd is full of test vboxes
(Action) grumbles and groans
AayJay: with ada or in general ?
AayJay
general, mostly input methods and wireless network
charlie5
long ago ?
i had probs with networking a few years back on archlinux ... not so much now
not sure about inout methods
input
AayJay
I switched (back) to Mint about a year ago. My wireless broke a few times when updating kernels.
Oh, and Nvidia Optimus
Not long after upgrading to KDE 5 did everything brake again. Lacking the patience to fix them, I installed LMDE.
Arch is still sitting in my HDD, waiting for me to fix it, someday, maybe.
charlie5
sudo pacman -Syu :)
(if only, eh ;) )
Visaoni
that has caused me occasional problems...
actually, think I've been sitting on a kernal update for a bit now... should probably bite the bullet and see what happens when I reboot
(so far it's all been related to being on a damned chromebook)
charlie5
(Action) has stuck with xfce4
Visaoni
(Action) uses awesome
charlie5: for the subprogram Random (you know, where you your random thing) would you rather pass in a generator to each call or not?
I'd probably fall in favor of it being an argument, but... if you kick the generator up from an argument to a generic parameter... maybe this works with extensive use of tagged types
I mean, you do end up in this weird situation where Random isn't primitive and can't be used with the dot notation... so maybe better to bind it in with the generic
although I might have just been thinking about abstracts totally wrong.
of course I was. I probably shouldn't be doing this at 5:30 in the morning
charlie5
my 1st impression is that 'generator' would be the base class with a primitive operation 'Random'
Visaoni
indeed. the problem is, Random needs to return whatever kind of type
charlie5
but then how to return random 'discrete type'
yes
Visaoni
one of the big issues with the standard random package is that the return type is baked in to the generic package instantiation. if you want random numbers in more than 1 range (even if it is subtypes), you're stuck instantiating multiple packages and generators.
god help you if you want to do the ranges at runtime
if you want to avoid having duplicate stuff all of the place for no good reason, you end up reimplementing the logic in Random yourself again
charlie5
could you make all generators map to a range 0..1 ... then have a generic which maps 0..1 to the discrete type ?
Visaoni
that's just what I was thinking about
and I was wrong about it being wrong - I think
as I mentioned, I don't know why I'm doing this now. but nonetheless
actually, my idea was still wrong. doesn't work. mappign to 0 .. 1 is interesting though
well, hm.
it could work, but you'd be forced to map to 64-bit floats, at which point you might as well just use 64-bit ints, which sort of makes the idea being having variable internal representations useless
at some point I might have to just give something up :P
charlie5
hmm, yes ... it might defaeat the purpose somewhat :0
precision/performance are the main golas i guess
goals
Visaoni
well, one of the nice things about it is that the internal random state thing can be quite small
obviously not particularly useful in anything like a desktop when you're already talking about 64-bits
but if you were to put it on a microcontroller or something, it could be nice to go smaller. whatever your "base" size is, you basically have 3 of them. 1 constant, 2 variables.
charlie5
heh, random internals are mystery to me
1 constant 'seed' + 2 next value vars ?
Visaoni
nah
charlie5
(Action) stabs the dark
(Action) misses
Visaoni
there's an old prng method that basically takes the current value, multiplies by a number, adds a number, and calls it a day. called a linear congruential generator
it's not that great
recently somebody came up with a way to add a twist to make it a lot better
charlie5
mm, there's probly a textbook alone on linear congruential generator :)
(Action) has heard of the twist
Visaoni
not that twist :)
charlie5
mersene ?
Visaoni
yeah, whole other kind of prng :)
charlie5
in what sense is it a twist ?
heh, i should probly google it and read the wiki ;)
Visaoni
basically the more significant bits of lcg output are better. low bits are pretty garbage. the twist (not the mersenne twister :P) is to take the hig hbits and use them to muck with the remaining bits (usually after droppign the lowest ones)
charlie5
mm, sounds very mathy :)
Visaoni
so say, take 64 bits of output. drop the lowest 29. use the top 3 bits to figure out how far to rotate the next 32 bits, and spit that out as the result
charlie5
wow
Visaoni
so, mechanically it's very simple. there's not much state to keep track of, and not a lot of work to do
it's also mind-fu*kingly configurable. what size is your internal state? any power of 2 is fine. how large is the raw output? what twist are you applying? how many bits are you using where?
charlie5
yeah, it seems simple enuff to code ... i expect a lot of time went into researching/proving it tho
Visaoni
and that's just the raw underlying things
it's dead simple
it's been coded up for ages. it's just not useable
« prev 1 2 3 next »