Open aloctavodia opened 11 months ago
I also propose to completely get rid of the "mean" line. Nobody understands what it actually means and there are many cases where it's range is much larger than each individual sample ending up in very strange/ugly plots.
Can we keep it, and make it default to False?
maybe if we rename the argument and legend entry?
what name(s) do you have in mind?
I think we should avoid the word "mean" because it makes people think about the mean of the posterior predictive. "combined" seems to be more intuitive in plot_trace, but here we combine over samples instead of over chain only, "aggregate" might also help with avoiding relating that to the actual mean.
Aggregate can be a good choice
Plot_ppc support for discrete values might have to be thought of as well: https://github.com/arviz-devs/arviz/issues/1103
Some more plot_ppc related issues with legacy Arviz plot_ppc:
A bug when zooming in: https://github.com/arviz-devs/arviz/issues/1756
A feature request for adding HDI-style credible intervals for plot_ppc: https://github.com/arviz-devs/arviz/issues/1654
Plot_ppc support for discrete values might have to be thought of as well: https://github.com/arviz-devs/arviz/issues/1103
Discussion related to better support for discrete values should go to #53.
Until explicit indexes are supported it won’t be possible to have the groupby like behaviour