lballabio
update JoeKuoD6initializers
In fact, JoeKuoD5initializers and JoeKuoD7initializers can contain 21200 direction numbers now, just as JoeKuoD6initializers. See https://web.maths.unsw.edu.au/~fkuo/sobol/
BTW,
- the comments (
// maxTabulated=1799) forJoeKuoD6is wrong, it should be// maxTabulated=21200. - maybe,
JoeKuoD6is a better choice thanJaeckelas the default argument forSobolRsg.
It seems that, the direction numbers used by Sobol sequences in NAG, SciPy and Julia are same as JoeKuoD6initializers. So, JoeKuoD6 is a better choice.
See,
- https://www.nag.com/numeric/cl/nagdoc_cl24/html/G05/g05ylc.html
- https://docs.scipy.org/doc/scipy/reference/reference/generated/scipy.stats.qmc.Sobol.html
- https://www.juliapackages.com/p/sobol
I agree to change the default — also, the D6 numbers should be updated since the version we have is pre-June 2010.
I'm not sure about extending D5 and D7; it adds a lot of size and they're probably unused.
I agree to change the default — also, the D6 numbers should be updated since the version we have is pre-June 2010.
I'm not sure about extending D5 and D7; it adds a lot of size and they're probably unused.
JoeKuoD7 is used by SobolBrownianBridgeRsg as default argument.
JoeKuoD7 is used by SobolBrownianBridgeRsg as default argument.
Do you think we should move it to D6, too?
Sorry, I don't know, it is beyond my knowledge. But professor Kuo says that D6 is the recommended choice.
Hmm, I gave it a try and moving to D6 as the default caused some figures in the tests to get worse. This might need some investigation.
On the other hand, I stand corrected — we already have the correct post-2010 version of the D6 numbers.