Learning Julia by an example of a simple Bayesian linear regression

[gdalle]

Now onto some of your specific questions:

@Boris

I am not sure I understand the output of @code_warntype for the gibbs function. Highlights union of nothing and Tuple at some point. Is this something to look out for?

Usually this denotes iteration, and you’re fine. Anything that shows up blue in @code_warntype is not a cause for worry. If you want an output that more closely matches your source code (instead of the low level representation that is harder to decipher), check out GitHub - JuliaDebug/Cthulhu.jl: The slow descent into madness. A bit scary at first but remarkably easy to get a handle on. Basically a @code_warntype on steroids with arbitrary depth.

@Boris

If I have one-liners for specific tasks (such as drawing sigma), is putting it in a function still a good idea, is there overhead associated with it? If I understand the performance tip function barriers this is always a good idea.

You also have to take into account clarity here. Separate functions are usually a good thing… up to a point. Not sure what that one-liner is, but there are many situations where sigma = rand(Normal(0, 1)) would be clearer that sigma = draw_sigma() (dumb example). So if it doesn’t affect your code speed, you shouldn’t always bother.

@Boris

Is there a way for module NewB to inherit using LinearAlgebra or is this not a good idea? What happens if I have a module needing another module and vice versa in the same time?

What do you mean by “inherit”? You already put using LinearAlgebra in your NewB module, and if you need it in your main file then put it there again. The module is designed to be reusable by someone else, who may not import LinearAlgebra in their main file after all.