chaospy
chaospy copied to clipboard
jonathf
Sobol' and Halton sequences should start with 0
Both sequences don't have (0, ...) as their first point. This is particularly wrong for Sobol' as it requires the number of samples to be a power of 2. Not having the zero will break the balance properties of the sequence when someone will request 2^n points.
I have a current PR about to be merged in scipy (scipy/scipy#10844) which introduce QMC with the zeros. There has been quite some discussions over there regarding the importance of having this point.
Okay, that is an interesting discussion.
I think inclusion is the correct behavior as well. Also, I think this is an awesome addition to scipy, and I will quickly deprecate my own implementation for QMC in favor of yours as soon as it is released.
Am I understanding the conversation corretly, that this is most likely going to be in the scipy release by December?
Am I understanding the conversation corretly, that this is most likely going to be in the scipy release by December?
That's also my understanding 🤞
Glad to here that you will be using this implementation 😃 . In that case, I am not sure that there would be a need to do extra work here and fix the zero in your implementation.
Exactly. Much cleaner to get the new feature with the switching of backends.
FYI, this is merged and will be in scipy 1.7. Let me know if you want some help to do a PR
Thanks for the notice. I've actually followed the discussion for your PR. Interessting to observe the methodologies and the direction it is all taking.
I will make an initial PR as I am going to clean up a few other things when I am touching that part od the code again. But if you don't mind, I'd love for you to take a look when a first draft is ready.
I will make an initial PR as I am going to clean up a few other things when I am touching that part od the code again. But if you don't mind, I'd love for you to take a look when a first draft is ready.
Of course, just ping me when you want me to review it 😃