These are chat archives for bluescarni/pagmo_reborn

31st
May 2016
Francesco Biscani
@bluescarni
May 31 2016 08:39
I merged the de_works branch :+1:
Dario Izzo
@darioizzo
May 31 2016 08:39
Yeah ...
Question, since you are there:
/home/dario/Documents/PaGMOreborn/tests/rastrigin.cpp(29): error: in "rastrigin_test": check (ras1.gradient(x1) == vector_double{2.}) has failed
/home/dario/Documents/PaGMOreborn/tests/rastrigin.cpp(30): error: in "rastrigin_test": check (ras5.gradient(x5) == vector_double{2.,2.,2.,2.,2.}) has failed
[2] false
[2, 2, 2, 2, 2]
Francesco Biscani
@bluescarni
May 31 2016 08:39
I need to do stupid fucking shit today, I wish I could work on pagmo instead :(
Dario Izzo
@darioizzo
May 31 2016 08:40
sorry....
then I will not bother you ...
Francesco Biscani
@bluescarni
May 31 2016 08:40
no it's ok... what about that?
Dario Izzo
@darioizzo
May 31 2016 08:40
well, the test fails, but ras1.gradient(x1) is [2] if i print it
Francesco Biscani
@bluescarni
May 31 2016 08:40
isn't it floating point comparisons doing bullshit?
Dario Izzo
@darioizzo
May 31 2016 08:41
but I thought that 2 is represented correctly in floating point no?
when I print on screen 2.000000000000000000000000001 is going to be 2 ?
Francesco Biscani
@bluescarni
May 31 2016 08:41
2 as a mathematical entity can be represented exactly, the printing is implementation-defined
Dario Izzo
@darioizzo
May 31 2016 08:41
must be that then ... thanks
last question can I log in eve and start a training without the eve client?
Francesco Biscani
@bluescarni
May 31 2016 08:42
I'd say it's that... I mean, if you do float f(2) then the result is exactly the mathematical 2, but it does not mean that if the result of an operation is 2 then it has to be exactly 2 in floating point
if it makes any sense :)
no you need the client
Dario Izzo
@darioizzo
May 31 2016 08:43
fuck ... loosing 1 day then :(
Francesco Biscani
@bluescarni
May 31 2016 08:43
itsaulgoodman
Dario Izzo
@darioizzo
May 31 2016 09:52
static_cast<double>(n)
This is safe right if n is an unsigned ... ?
We risk loosing nothing relevant correct?
Francesco Biscani
@bluescarni
May 31 2016 09:55
sure, conversion from integrals to floating point is fine... on 64bit archs a double will not be able to represent exactly ints > 2^53 or so, but I don't think anybody cares :)
Dario Izzo
@darioizzo
May 31 2016 09:55
thought so.
Marcus Märtens
@CoolRunning
May 31 2016 10:32
We absolutely don't need one, this is just a penis-waving contest here.
Francesco Biscani
@bluescarni
May 31 2016 10:49
eve in a nutshell
Marcus Märtens
@CoolRunning
May 31 2016 11:00
Sounds like it
Dario Izzo
@darioizzo
May 31 2016 12:48
Merge request more_problems is finished. Ackley, Rastrigin (with grad and hessians), Griewank and Schwefel are added.
Please review and merge