add guidance on sparsity

[jaime-benavides]

Can we add here that as a general rule we consider optimal a 1% of valus flagged as extreme for EH data? and that more than 5% might be too much?

pcpr/vignettes/pcp-applied.Rmd

Line 289 in 316e977

`r 100 - round(gs$summary_stats$S_sparsity[1] * 100, 1)`% of values in the data were flagged as extreme,

[lawrence-chillrud]

Hmm, we describe how to interpret sparsity measurements in the theory crash course here as well as in the documentation for the grid_search_cv() function.

I do not want to prescribe a one-size-fits-all rule for "good" vs. "bad" output sparsities. Additionally, measuring sparsity is subjective in the sense that it relies on setting a threshold thresh determining which observations to consider "practically zero" and count in a sparsity calculation, and which to consider as "non-zero" extreme events.

As such, I'm inclined to leave this section as is. I can add a little line that refers users to the grid_search_cv() documentation though.