| |
| |
| |
|
Page: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Comments:
<0> I don't understand their reasoning
<1> japhy: It is so accumulated roundings do not screw your data
<1> Or something
<2> so much fun :)
<3> japhy: I think they're just trying to eliminate bias in-the-large.
<4> isn't IEEE designed for/by people who really care about this?
<3> (law of large numbers and all that)
<0> how? they're CREATING bias!!!
<4> oh, duh! When your host system has a 100 Hz clock, of course qemu can't simulate a 1000Hz one!
<0> they're adding 1 number to the list of numbers that round to 4, and removing one number from the list of numbers that round to 3
<0> how is that "fair" or "unbiased"?
<4> japhy: because there's an infinite amount of each of those...
<0> integral: not when you have a finite set of bits
<2> integral: if could simulate but not in realtime
<4> JohnFlux: yeah, it does simulate, but time ... goes ... really ... really ... slowly
<2> integral: exactly
<4> anyway, I'm happy I've got a reason my code wasn't appearing to work right :-)
<0> integral: any response?
<0> anyway, they might be infinite, but they have the same ordinality.
<4> "finite set of bits" -- but floating point is a wee bit funny in it's ranges since you can make really big numbers like 1e99, and really small ones like 1e-99, and then those which actually get rounded like 1.4
<4> japhy: "ordinality"?
<0> that is, there is a 1-to-1 correlation between x st 1.5 <= x < 2.5 and y st 2.5 <= y < 3.5
<0> integral: so you can't "take one number and move it to the other set" without consequences
<4> umm, notice that one side has a = and the other doesn't?
<4> perhaps they're trying to compensate for the fact that one could setup the system to be 1.5 < x <= 2.5, 2.5 < y <= 3.5 ?
<0> integral: yes. that makes no difference here.
<0> integral: but that's NOT the system.
<0> they're not trying to solve a problem that doesn't exist, they're creating a problem. they are giving bias to rounding to even numbers.
<1> japhy: So what should 2.5 round to?
<0> buu: 3. any integer plus 0.5 rounds up.
<1> So
<1> eval: 2.5 + 3.5 / 2
<5> buu: Return: 4.25
<1> er
<4> I can't believe there's nothing on the web that attempts to cover this :-/
<1> eval: ( 2.5 + 3.5 ) / 2
<5> buu: Return: 3
<0> integral: here's a better example of what I mean when I say ordinality
<4> hmm, I see a "Formal verification of a theoryof IEEE Rounding"
<1> eval: ( POSIX::ceil(2.5) + POSIX::ceil(3.5) ) / 2
<5> buu: Return: 3.5
<0> for every real number X that rounds to 2 (the way I was taught), there is another real number X+1 that rounds to 3.
<4> of course, the IEEE rounding mode *can* be changed...
<6> yes. 5-X. :)
<1> eval: ( (sprintf "%.0f", 2.5) + (sprintf "%.0f", 3.5 ) ) / 2
<5> buu: Return: 3
<1> japhy: Do you see my point?
<0> buu: no, I don't.
<1> Oh ok.
<0> I see your numbers
<0> please make a conclusion
<4> I guess one question that it might be handy to answer would be "Why even and not odd?"
<1> japhy: 2.5+3.5/2 should be 3, not 3.5 =]
<0> you're rounding your numbers prematurely, and that's your own damn fault, buu
<1> japhy: IEEE's rounding method attempts to make this happen.
<4> oh, here's an argument: -0.5 and 0.5 both round to zero
<0> buu: by sacrificing accuracy
<4> but with "round up", one does to 0 and the other to 1
<0> integral: yes.....
<4> so round(-x) = -round(x)
<1> japhy: *shrug*.
<0> that's not an argument. that's a "feature" of changing the rules of rounding
<1> japhy: Not really.
<4> well, my argument is that this property of round is *useful* to have, so you pick a ruleset that favours it
<0> but that property is not taught in mathematics
<4> sure, but IEEE is *engineers* not mathematicians :-)
<7> they're better
<0> well, too ****ing bad. then they should use engine instead of math when they do calculations.
<0> and leave the math to the mathematicians.
<3> japhy: engineering is just applied math. Math for math's sake isn't too useful (except to mathematicians)
<4> pure math ~~ art
<1> japhy: I believe it is also known as "Banker's rounding"
<4> it would be interesting to know how rounding affects algorithms used to compute things like log, sin, etc
<8> math for math's sake is good for teaching
<6> banker's rounding was used in Brewster's Millions.
<6> and Trading Places.
<8> merlyn: Brewster's Millions? That movie with Richard Pryor?
<6> I think that was the one
<8> I loved that movie.
<9> It's been a long time since I've seen Brewster's Millions
<8> and if Trading Places was the one with Eddie Murphy and Dan Ackroyd, that movie kicked *** too.
<6> it's called a "salami attack"
<6> Trading Places - nekkid Jaime Lee Curtis
<8> Randolph and Mortimer
<8> merlyn: *drool*
<6> and Office Space most recentlyh.
<0> wow. my brother just came up with
<0> IEEE = I Extra-Enjoy Evens
<8> oh, wait the rounding thing?'
<8> it was in Hackers too
<8> am I thinking of the right thing?
<4> it was also in Superman ??, another Richard Pryor film
<6> right... that's what I was thinking of instead.
<8> there's a new superman film coming out soon. I haven't seen any of the old ones in a long time.
<1> I never saw the point of superman
<1> I mean, he's superman, what does he fight?
<9> Superer men
<4> let's just blame goethe
<10> can someone tell me what I'm doing wrong here with this RPC::XML client code? http://rafb.net/paste/results/u1x2GP44.html
<11> The paste u1x2GP44 has been moved to http://erxz.com/pb/935
<4> s/goethe/nietzsche/
<8> buu: he fights evil, duh
<8> pryor was in superman 3, btw
<0> ok, my brother points out that the rounding I use rounds away from zero on .5, rather than towards positive infinity
<0> and I don't know which is correct.
<0> I guess it should always be towards positive infinity
<6> and beyond!
<8> *rimshot*
<4> gah, I hate people.
<0> I'll buzz your lightyear
<8> do you hate everyone, integral, or just mostly everyone?
<4> oh, maybe not. but still. it's bloody hard to find the DWARF2 spec.
<8> ah
<4> godling: everyone. I don't like to discriminate.
<9> japhy: You're buzzed from light beer?
<0> PerlStalker: I can't remember the last time I drank a "light beer"
<4> "piss"
<8> Us Americans are good targets for hatred.
<0> US Americans too, godling
<12> howcome?
<6> I know he hates me.
<6> or at least I think he does, from time to time.
<0> ok, time to go home
<8> I've got loads of reasons, I just came from US history cl***.
<13> godling: If only possibly for a lack of using proper grammar. "We Americans ..."
<4> gah, this link turns out to be DWARF for S/390. how wonderously useful.
<12> \
<8> infi: objective vs. nominative?
<10> no RPC::XML folks here, eh? dang
<8> "we had a big bowl of snot." vs. "us had a big bowl of snot."
<8> hrm
<8> infi: thanks
<4> eeep, DWARF2 is complex :-/
<8> integral: did you think it would be otherwise?
<4> *sigh* I guess I should realise that this stuff is thought up by the sick and twisted, by now
<14> TOO QUIET
<15> good day, Yaakov
<14> Hello.
<8> Yaakov: I could start reading your mind again, if you'd like.
<14> godling: No, thank you.
<4> dMnt+'v/um -+ n. largess, gratuity.
<14> Shabbos starts soon and I am sick and uncomfortable.
<14> But I did manage to cook dinner.
<8> what do you do for shabbos?
<14> integral is unicode boi.
<16> GumbyBRAIN, Gimme TreeFiddy
Return to
#perl
or
Go to some related
logs:
Genius GW-7100p linux
#css
zen-m ubuntu
#debian
ubuntu 3dm2 install
Requested Range Not Satisfiable gentoo
how to install limewire in mepis?
relay dfilt
by+Souperman++trivia
#perl
|
|