forestplot: support multiple traces

[aloctavodia]

Sometimes is useful to visually compare the results/posterior for different models. One option to achieve this task is to make a forestplot for all variables in all models. This PR modifies forestplot to support either a trace or a list of traces, a new keyword models can be used to pass the names of the models

This is an example of a forestplot for several models:

model m0 does not have a beta parameter, contrary to models m1 and m2. Additionally, the beta parameter for m2has shape=2.

[fonnesbeck]

It would be cool to do something to make the models stand out, such as colorize the labels by model. Conversely, I wonder if you could create shading bands to isolate the same parameters.

[aloctavodia]

I agree, the plot is too flat. I will try to do something to make the models stand out... maybe a color per-model and also the "shading-bands"

[aloctavodia]

Shading bands are used to help visualize groups of parameters (only when more than one trace is passed).

Additionally, by setting colors='cycle', we get a different color per model

Previous plots were done with the style I generally use, i.e. plt.style.use(['seaborn-colorblind', 'seaborn-darkgrid']). This is how the first plot looks with the default matplotlib style.

[junpenglao]

WOW this is really nice!

[fonnesbeck]

Yeah, that's exactly what I had in mind. Nice work.

[aloctavodia]

Should I also add this to the release-notes?

[aloctavodia]

Is this ready to squash and merge or maybe more work is needed?

[junpenglao]

This is really a nitpick but is it possible to adjust the shadow to more in the middle between two errorbars?

[aloctavodia]

No nitpicking at all @junpenglao! This is how it looks with your suggestion.

[junpenglao]

Thanks @aloctavodia !