Different distributions per control variable

[anor4k]

Hi! I'm trying to understand how I could pass different Prior distributions to each of my control variables. Something like:

prior = md.model.prior_distribution.PriorDistribution(
    gamma_c=[
      tfp.distributions.half_normal.Normal(0, 1),
      tfp.distributions.half_normal.HalfNormal(5),
    ],
)

But the above sample doesn't work, since it expects a single distribution.

I tried using the tfp.distributions.independent_joint_distribution_from_structure function from tensorflow probability but no luck.

[alexwakim]

Hi, we don't currently support mixing distributions for a single parameter.

Can you use tfp.distriubtions.TruncatedNormal to get something close to what you want? You can use different points of truncation.

[anor4k]

Yes, using TruncatedNormal seemed to be the best alternative! Thanks.