I have almost finished Richard McElreath’s online course called “Statistical Rethinking” which is pretty good for Bayes beginners like myself. (You can find it by searching youtube for the playlist code PLDcUM9US4XdNM4Edgs7weiyIguLSToZRI which I cannot post a direct link to here because of the forum’s spam-filter.)

The latter part of the course uses a lot of multi-variate normal distributions, which are fairly simple and elegant to make in R. See for example the homework examples for that course:

But MV Normal distributions seem to be really complicated in pymc3. See for example the translation of McElreath’s book from R into pymc3:

I have searched this forum and other blogs etc., and this seems to be the correct way of using MV Normal distributions in pymc3. But it is an awful lot of code-lines that I frankly don’t understand, because I’m not familiar with the underlying math and algorithms to improve efficiency and numerical stability or whatever.

Since this is fairly simple in R, I wonder why it has to be so complicated in pymc3? MV Normal distributions are quite common, so is this something you might consider simplifying in pymc3 in the future?

Thanks!

PS: Please disable the forum’s spam-filters, it is really annoying that it rejects posts with more than 2 links etc.