Bootstrap plots

[palday]

coefplot will generate coefplots from bootstrap samples, but maybe there are a few other plots we could do.

Maybe something like coefplot but with distributions/histograms/kde instead of / in addition to the intervals? What's called a "ridge plot" over in the Bayesian world.

Potentially something like a partial effects plot that plots the fitted line + a sample of lines based on the parameters from the bootstrap for each coefficient (I'm thinking of tiling like we do in shrinkageplot!). This would be similar to the posterior predictive distribution used in Bayesian diagnostics and would provide a visualization of the uncertainty around the original fit.

Something showing off the covariance of different parameters?

cc @dmbates

[dmbates]

That is sort-of what I had in mind, although with my background I don't see much in the way of coefplot. I think they are more prominent in your area. I will spend some time tomorrow working on density plots and perhaps 2-dimensional density plots.

We should also consider plots from parameter trajectories and the objective when thin is set to a small value.

[palday]

We should also consider plots from parameter trajectories and the objective when thin is set to a small value.

Most of that code is already in the SMLP2021 notebook I created for the animations. I'll convert it to usable functions when I get a chance. 😄

I agree coefplot is pretty trivial, but a lot of researchers prefer seeing those things as plots instead of tables. 🤷