Add support for CdfNorm
Closed this issue · 6 comments
Hi,
Great package, thanks for making it, it's sped up my code considerably.
I'm aware this would probably come under the "if there's enough interest" category, but would it be possible to add support for VML's CdfNorm function? It's not urgent as we can just use the relation to erf() instead, but it'd be interesting to see if using the dedicated function is faster. I would offer to give it a go myself, but I'm not sure I have the chops.
Cheers
Hey, very glad you like this package.
I think that should be doable, you mean this function, correct?
https://www.smcm.iqfr.csic.es/docs/intel/mkl/mkl_manual/vml/functn_vCdfNorm.htm
Oh cool! That's the function I mean, yes.
Sorry that it took me a while, I made an MR and should have it merged by the end of the day. From my (not very rigorous) tests, it seems to be faster than going via erf.
Thanks Steffan that's great, it looks like there was an issue with the MacOS/Julia nightlies which prevented the merge, but I'll give it a go once it's merged and let you know how it performs in my code. Happy to do some comparisons for the README if that's useful - I don't think Base has a cdfnorm() function, but I could run comparisons for various dimensions both against the IVM.erf() -based implementation and against the standard cdf() function applied to a normal distribution from Distributions.jl .
There was an issue with the tagbot, which only got fixed today. Will hopefully merge later today.
Ok, it is done, merged, and registered.