distuv LogCDF

[ljcolling]

What are you trying to do?

Currently all the distributions have a CDF method. However, in addition to this, I think a LogCDF method would be useful. This would necessitate performing the calculations on the log scale, because simply taking the log of the CDF won't have sufficient accuracy.

For example,

dist := distuv.StudentsT{
		Mu:    0.0,
		Sigma: 1,
		Nu:    33,
		Src:   nil,
	}

math.Log(dist.CDF(23)) // returns 0

Compared with the equivalent R code

pt(q = 23, df = 33, ncp = 0, lower.tail = TRUE, log.p = TRUE) 
# [1] -3.43213e-22

Are you able to help contribute the feature?

I'm happy to help contribute to this. I'm currently using translations I've made of R nmath C library, and these translations could be used as the basis for this.

[kortschak]

Can I confirm that the translations you have made are a reimplementation without reference to the nmath source or that the source is compatibly licensed with a BSD license (MIT/BSD etc)? If it is a direct translation of GPL code, I don't think we can accept it.

[ljcolling]

Yeah, unfortunately it's a translation of the gpl source, so you're right that'll probably cause issues with license compatibility 🙁