# Bayesian Model Initial Parameters

**URL:** <https://discourse.pumas.ai/t/bayesian-model-initial-parameters/726>\
**Category:** bayesian\
**Created:** [July 15, 2022, 7:22pm UTC](https://discourse.pumas.ai/t/bayesian-model-initial-parameters/726 "2022-07-15T19:22:55Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Rahulub3r](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pumas.ai/rahulub3r/32/182_2.png) [@Rahulub3r](https://discourse.pumas.ai/u/Rahulub3r)\
**Post date:** [July 15, 2022, 7:22pm UTC](https://discourse.pumas.ai/t/bayesian-model-initial-parameters/726/1 "2022-07-15T19:22:55Z")

</div>

How does Pumas calculate initial parameter of the omegas in a Bayesian model for Inverse Gamma distribution ?

For e.g. if I define an omega like this: `ω²SCR0 ~ InverseGamma(10/2, 0.145*(10 + 1 + 1)/2)` and run `init_params(model)` I see `ω²SCR0 = 0.18625925314116212`. I want to know where the 0.186… number is coming from?  
Also, is there some underlying meaning in writing 10/2 as opposed to 5, or (10+1+1)/2 instead of 6 in the specification of the distribution ?

---

<div class="post-metadata">

**Author:** ![storopoli](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pumas.ai/storopoli/32/190_2.png) [@storopoli](https://discourse.pumas.ai/u/storopoli)\
**Post date:** [July 20, 2022, 7:47am UTC](https://discourse.pumas.ai/t/bayesian-model-initial-parameters/726/2 "2022-07-20T07:47:47Z")

</div>

> [@Rahulub3r](#):
>
> ω²SCR0 ~ InverseGamma(10/2, 0.145\*(10 + 1 + 1)/2)

My view is that if you do an `init_params` in a Bayesian model it will sample prom the prior.

```julia
julia> using Distributions

julia> ω²SCR0 = InverseGamma(10/2, 0.145*(10 + 1 + 1)/2)
InverseGamma{Float64}(
invd: Gamma{Float64}(α=5.0, θ=1.149425287356322)
θ: 0.8699999999999999
)

julia> rand(ω²SCR0)
0.2632303021306265

julia> rand(ω²SCR0)
0.5991101635236422

julia> rand(ω²SCR0)
0.270555800869863

julia> rand(ω²SCR0)
0.11468427269228272

julia> rand(ω²SCR0)
0.09238162211517542

julia> rand(ω²SCR0)
0.29396951024526896

```

---

<div class="post-metadata">

**Author:** ![mohamed82008](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pumas.ai/mohamed82008/32/127_2.png) [@mohamed82008](https://discourse.pumas.ai/u/mohamed82008)\
**Post date:** [July 20, 2022, 9:52am UTC](https://discourse.pumas.ai/t/bayesian-model-initial-parameters/726/3 "2022-07-20T09:52:17Z")

</div>

> [@storopoli](#):
>
> My view is that if you do an `init_params` in a Bayesian model it will sample prom the prior.

It’s actually the median.

```julia
julia> median(InverseGamma(10/2, 0.145*(10 + 1 + 1)/2))
0.18625925314116207

```