Check that all plot text is correct (e.g. titles, axis labels, etc.) about bayesplot HOT 12 CLOSED

I have checked all pp_check plots and only have three minor suggestion:

1. In ppc_resid_binned, the y-axis label says "Average Residual" but wouldn't "Average Residuals" be more consistent?
2. In ppc_ts* and ppc_vs_x* one might want to add a legend indicating that y is displayed by the lines / dots similar to how it is done in ppc_violin_grouped.
3. In the facet titles of ppc_resid, yrep is written in italic while y is not.

from bayesplot.

Text in ppc plots

• Fix y-axis label in ppc_resid_binned
• Fix facet labels in ppc_resid
• Add legend to ppc_ts* and ppc_vs_x*

Text in mcmc plots

still needs checking

from bayesplot.

I have checked the mcmc_plots and have a few suggestions:

1. mcmc_hist_by_chain uses the same y axis for all parameters, which may lead to strange looking densities (try it with example_mcmc_draws()). Similarily, mcmc_areas uses the same x-axis, which again may look strange. (for mcmc_intervals using the same x-axis not a problem from my perspective)
2. The visualization of the point estimate in mcmc_areas only goes to the middle of the density. I think it looks better to display it completely from the bottom to the top of the density.
3. In mcmc_neff_hist the colors overlay each other at the borders.
4. In mcmc_nuts_* you write accept_stat__ and treedepth__. Are those many underscores necessary when displayed in the plot?
5. mcmc_nuts_energy fails with Aesthetics must be either length 1 or the same as the data (1): fill, colour, x

from bayesplot.

Thanks Paul, I'll try to get to these today. Much appreciated.

On Monday, August 29, 2016, Paul-Christian Bürkner [email protected]
wrote:

I have checked the mcmc_plots and have a few suggestions:

1. mcmc_hist_by_chain uses the same y axis for all parameters, which
   may lead to strange looking densities (try it with example_mcmc_draws()).
   Similarily, mcmc_areas uses the same x-axis, which again may look
   strange. (for mcmc_intervals using the same x-axis not a problem from
   my perspective)
2. The visualization of the point estimate in mcmc_areas only goes to
   the middle of the density. I think it looks better to display it completely
   from the bottom to the top of the density.
3. In mcmc_neff_hist the colors overlay each other at the borders.
4. In mcmc_nuts_* you write accept_stat__ and treedepth__. Are those
   many underscores necessary when displayed in the plot?
5. mcmc_nuts_energy fails with Aesthetics must be either length 1 or
   the same as the data (1): fill, colour, x

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#24 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AHb4Q8srcY3Fkmej3jnp_uCxf9chpwu8ks5qkto6gaJpZM4Jo3Le
.

from bayesplot.

For (1):

It's not clear to me how to deal with the issue you raise about mcmc_hist_by_chain. Try this:

x <- example_mcmc_draws()
data <- reshape2::melt(x, value.name = "Value")
graph <- ggplot(data, aes(x = Value, y = ..density..)) +
  geom_histogram()

graph + facet_grid(Chain ~ Parameter, scales = "fixed")
graph + facet_grid(Chain ~ Parameter, scales = "free")
graph + facet_grid(Chain ~ Parameter, scales = "free_y")
graph + facet_grid(Chain ~ Parameter, scales = "free_x")

None of the different scales settings creates a plot that is equally readable for all parameters.

from bayesplot.

For (5):

I'm not seeing that failure with mcmc_nuts_energy. Do you have an example?

from bayesplot.

For (2):

should be taken care of by 3971db4

from bayesplot.

For (3):

I'm not sure why this is happening. You? It might be that unless the binwidth is really tiny there is going to be some overlap at the cutoff points for where the color changes.

from bayesplot.

For (4):

No, we don't need the underscores, I was just trying to be consistent with how we usually write them. I'm undecided on whether to keep them or remove them.

from bayesplot.

(1). You are right. We won't be able to achieve this with facet_grid, since rows are indicating chains not parameters. Accordingly, we will have to leave it as it is I guess.

(3) I have no idea and it is only a minor issue anyway and likely nothing that can be changed outside of ggplot...

(4) I was seeing that you write log-posterior instead of lp__ so I thought, you might also want to change the other names. I am fine with both of course.

(5) After updating all of my packages the error went away.

from bayesplot.

(1) Even if we swapped it so that rows were parameters and columns were chains (i.e. Parameter ~ Chain in the facet formula) it wouldn't work either. So I guess we either leave it as, get rid of the plot entirely (I'm not even sure how useful it is), or put in the documentation that it doesn't work well if you show parameters with large uncertainties and small uncertainties in the same plot. Preference?

(3) If it's confusing to have that overlap I could make the histogram all the same color and just put vertical lines at the points where the color currently changes. Do think it's preferable to do that or leave as is?

(4) Good point. Although it may actually make the most sense to just change log-posterior back to lp__ because it's almost always not really the log-posterior (just up to constant)

(5) Ok great

from bayesplot.

(1) I would leave the plot as is but put a little note in the documentation.

(3) It's not confusing it is just not perfect. I prefer colored bars over vertical lines though, so leave it as is for now.

(4) As long as you explain all the technical parameter names in the doc, using lp__ will be more consistent I guess. And technical names are ok for technical plots I guess.

from bayesplot.