Bayesian Structural Time Series - Code adjusments #207

Chapter 06,
Section 6.4.3 Bayesian Structural Time Series

I was trying to run some code available on https://bayesiancomputationbook.com/markdown/chp_06.html#bayesian-structural-time-series, but it was giving me some errors. Then, I checked the jupyter notebook on this repo and found that the code on site has some differences from the jupyter notebook.

I've listed below some differences that I've found, maybe it is just updating the site to reflect the code on ch06.ipynb

The line below is on ch06.ipynb, but isn't on the site version.

birth_model_jd = birth_model.joint_distribution(observed_time_series=observed)

The code block 6.27, below, has been updated on ch06.ipynb, but isn't on the site version.

Site:

# Using a subset of posterior samples.
parameter_samples = [x[-100:, 0, ...] for x in mcmc_samples]

Ch06.ipynb

# Using a subset of posterior samples.
parameter_samples = tf.nest.map_structure(lambda x: x[-100:, 0, ...], mcmc_samples)

Also, one important line to run the code isn't on the site. Maybe it is good to show it in some code block on the site too.

mcmc_samples, sampler_stats = run_mcmc(
    1000, birth_model_jd, n_chains=4, num_adaptation_steps=1000,
    seed=tf.constant([745678, 562345], dtype=tf.int32))

Thanks!