Manuscript Error: Get `NotImplementedError: Sorry, pickling not yet supported.` when trying to run `Listing 5.5`

[paw-lu]

Description

When trying to run Listing 5.5:

with pm.Model() as splines:
    τ = pm.HalfCauchy("τ", 1) 
    β = pm.Normal("β", mu=0, sd=τ, shape=B.shape[1])
    μ = pm.Deterministic("μ", pm.math.dot(B, β))
    σ = pm.HalfNormal("σ", 1)
    c = pm.Normal("c", μ, σ, observed=data["count_normalized"].values)
    idata_s = pm.sample(1000, return_inferencedata=True)

I get

NotImplementedError: Sorry, pickling not yet supported. See https://github.com/pydata/patsy/issues/26 if you want to help.

As seen by the traceback, this seems to be due to using Patsy's design matrix in a model (variable B in the snippet above). You can reduce the above example to just:

with pm.Model() as model:
    β = pm.Normal("β", mu=0, sd=3, shape=b.shape[1])
    μ = pm.Deterministic("μ", var=pm.math.dot(B, β))
    idata_s = pm.sample(1000, return_inferencedata=True)

And still get the same exception thrown.

Reference

https://bayesiancomputationbook.com/markdown/chp_05.html#id18

Possible solution

We can just convert the DesignMatrix into an ndarray to avoid the patsy issues.

-     μ = pm.Deterministic("μ", var=pm.math.dot(B, β))
+     μ = pm.Deterministic("μ", var=pm.math.dot(np.asarray(B), β)) 

As always, happy to provide PR here if it helps!

Author Section

Do not close until

• Added to Errata
• Fixed in Open Access
• Fixed in latex source