Learning Julia by an example of a simple Bayesian linear regression

General Usage Performance
, , , ,
4

Thank you, I had the opportunity to try out some of the suggestions.

For example, the resid = Y - X * beta_d; trick reduced the allocations of sigma from 7 to 44, I would watch out for such points in the future.

mine
julia> @time genSigma(Y,X,beta_d,nu0,d0)
  0.000013 seconds (7 allocations: 5.016 KiB)
0.10240049561807199

dot trick:
julia> @time genSigma(Y,X,beta_d,nu0,d0)
  0.000016 seconds (4 allocations: 2.516 KiB)
0.08709325471913029

and simply that reduced the total code by almost a half (and allocations by a lot)!

julia> @time include("MainNewB.jl")
  0.048585 seconds (150.56 k allocations: 51.639 MiB, 9.80% gc time)
vs
julia> @time include("MainNewB.jl")
  0.074861 seconds (171.56 k allocations: 68.732 MiB, 11.86% gc time)

I am at that amount of seconds that I should start using BenchmarkTools

I am unsure how to incorporate the "!" suggestion (just changing the values of beta_d instead of re-allocating it. I tried following the advice in this discussion, more specifically to use x=f(x) as in this post

I tried pre-defining beta_d in gibbs such that

function gibbs(Y,X,BETA0,Sigma0,sig2_d,d0,nu0,n_gibbs,burn)
    beta_d = zeros(5,1)
    ...
    for i = 1:n_gibbs
        beta_d = genBeta(X,Y,BETA0,Sigma0,sig2_d,beta_d)
        ...
    end

and then in genBeta assigning to beta_d’s elements but that doesn’t seem to be it.

beta_d[:,1] = Beta1 + C.L*randn(5,1)

alternatively I tried beta_d .=.

What is the correct way to go about this? Just a reference to the manual can also help, I couldn’t find easily functions with the "!" operator. I looked at the broadcast part and I thought I can follow but obviously I didn’t implement it well as I don’t see any changes to the allocations, nor time.