mikejacktzen
mikejacktzen
I have some familiarity with this topic, it's very tricky. First, I'm assuming this discussion is all about 'probability' == 'sampling' == 'scale up' == 'representation' weights? If unit i...
Thanks for the thoughtful reply. I understand to your agreeable points. Those numerical stability checks are always a fun surprise. I think I rushed my original post (excited about the...
That sounds right. Is something similar to the quasi likelihood already what lightgbm uses? If so, that's even better, time / coding wise. I'm hoping this request does not have...
I think i should have started my hunt using lightgbm examples! Looks like lightgbm supports cross-entropy allowing for [0,1] labels with a loss similar to binomial(N=1,p) ~= quasi likelihood https://lightgbm.readthedocs.io/en/latest/Parameters.html#core-parameters...
Hi Brian and axsk, I think this is a good opportunity to check in. Awhile ago, I wanted to implement a custom many to one function to use as a...
@brian-j-smith I have confirmed that updating julia to 0.4 and Mamba to 0.7 has broken my original 'maybe-working' solution of a customized density that computes on a vector of scalar...
@brian-j-smith Reporting back! the multivariate solution with the newer signature restrictions via Mamba 0.7 work! I think the new mamba signature restrictions is clearer. In previous mamba pre 0.7, i...
Hi @brian-j-smith and @axsk , I'm following up this thread by trying to generalize my successful example of a vector containing univariate outcome with logpdf evaluation of a scalar parameter...
@bdeonovic Thanks for the wishart.jl example I was under the impression that in pure julia, everything is an array, even if 'matrices' are operated in matrix fashion. But the wishart.jl...
@bdeonovic Just to report back, my particular case was resolved with the correct typing of AbstractArray{T,2} @axsk I like the idea of a abstract density distribtuion where the user could...