Feature Request: Expose a method for prior predictive checks

[markjrieke]

It would be useful to expose a method for running prior predictive checks / plotting the results (similar to the posterior predictive checks that are available via visualizer.ModelFit(mmm).plot_model_fit(). This would help assess reasonable-ness of custom priors ahead of sampling.

[cpulavarthi]

Hello @markjrieke,

Thank you for your request for the feature you’d like to see in Meridian. We shall take it into consideration for future releases. You can track the previous changes to Meridian and unreleased feature updates in our Change Log.

Thank you

Google Meridian Support Team

[markjrieke]

I've just done some digging around in the source code, and it appears that the plot_model_fit() method actually plots the distribution of the posterior mean rather than the distribution of the posterior response (i.e., plot_model_fit() plots the distribution of $\mu_t$, rather than the distribution of $\hat{y}_t$). Ideally, the prior predictive check method requested here would plot the prior response distribution, rather than the prior mean distribution.

I'll open up a separate issue for a request for changes to plot_model_fit() to plot the posterior predictive distribution

[markjrieke]

@lukewang-aofl