GLM model from statistical rethinking video series

Specific Domains Probabilistic programming
1

Hello, i have an issue with the Turing code related to this model in the Statistical Rethinking video series. This code is not in the book so i didn’t find the solution in the good ressourcesTuringModels.jl or SR2TuringPluto.jl.

To reproduce the error here is the code of the data (included in the StatisticalRethinking package).

using MLLabelUtils, Turing, Distributions, CSV, DataFrames, StatisticalRethinking

df = CSV.read(sr_datadir("UCBadmit.csv"), DataFrame)
df.gender = ifelse.(df.gender .== "female",1,2)
df.dept = convertlabel(LabelEnc.Indices{Int}, categorical(df.dept))


@model function direct_effet_binomial(x,n, a)
    alphas ~ filldist(Normal(0, 1), length(levels(x[:,1])),length(levels(x[:,2])))

    p = logistic.(alphas[x[:,1],x[:,2]])
    @. a ~ Binomial(n,p)
end

mGD = direct_effet_binomial(df[:,[:dept, :gender]], df.applications, df.admit)
sample_mgd = sample(mGD, NUTS(), 3000)

precis((DataFrame(sample_mgd)))

Here is the model from the video

m2 <- ulam(
    alist(
        A ~ bernoulli(p),
        logit(p) <- a[G,D],
        matrix[G,D]:a ~ normal(0,1)
    ), data=dat_sim , chains=4 , cores=4 
)

The model works and give me some results but they are not the ones that i’m supposed to fet grom the course

good_output
good_output627×689 347 KB

I must have made a mistake somewhere and i am still new to the field so i’m sorry if it is an obvious bug.