Expanding plotting options of plot_posterior_predictive

[jsnyde0]

Hey! The plot_posterior_predictive has fairly limited options.

In order to see e.g. the posterior mean as well, I wrapped it in my own function that adds more to the returned fig.

I've never contributed to anyone else's code base, but would you be open to a PR where I add some plotting options like that?

[williambdean]

I think that's a good idea. I would love to see the plotting functionality have a bit more flexibility and extensibility.

What changes to do you have in mind?
There are some issues here: https://github.com/pymc-labs/pymc-marketing/issues?q=is%3Aopen+label%3Aplots+label%3AMMM
Feel free to add any more if you feel like

[jsnyde0]

Ok how do you feel about an approach that allows both simple and more advanced usage? We can leave out the function arguments from plot_posterior_predictive if that's your preference.

API

def create_posterior_predictive_plot(ax=None, figsize=(10, 6), **plt_kwargs) -> plt.Axes:
    """Create base posterior predictive plot."""

def add_hdi_to_plot(ax, hdi_prob, color=None, alpha=0.3, **kwargs) -> plt.Axes:
    """Add HDI to existing plot."""

def add_mean_to_plot(ax, color='red', linestyle='-', **kwargs) -> plt.Axes:
    """Add mean prediction to existing plot."""

def add_gradient_to_plot(ax, cmap='Blues', n_percentiles=100, **kwargs) -> plt.Axes:
    """Add gradient representation to existing plot."""

def plot_posterior_predictive(hdi_probs=[0.94, 0.5], show_mean=True, 
                              show_gradient=False, ax=None, **plt_kwargs) -> plt.Axes:
    """Convenience method for common plot configurations."""

Simple usage

ax = mmm.plot_posterior_predictive()
plt.show()

Advanced usage

fig, ax = plt.subplots(figsize=(15, 10))
ax = mmm.create_posterior_predictive_plot(ax)
ax = mmm.add_hdi_to_plot(ax, 0.94, color='blue', alpha=0.2)
ax = mmm.add_mean_to_plot(ax, color='red', linestyle='--')
plt.show()

[juanitorduz]

Hey @jsnyde0, Thanks for the input! What about extending the current function for backward compatibility? We can have individual private functions

def _add_hdi_to_plot(ax, hdi_prob, color=None, alpha=0.3, **kwargs) -> plt.Axes:
    """Add HDI to existing plot."""

def _add_mean_to_plot(ax, color='red', linestyle='-', **kwargs) -> plt.Axes:
    """Add mean prediction to existing plot."""

def _add_gradient_to_plot(ax, cmap='Blues', n_percentiles=100, **kwargs) -> plt.Axes:
    """Add gradient representation to existing plot."""

and then we allow this to happen in the plot_posterior_predictive as for example

mmm.plot_posterior_predictive( . . . , add_mean: bool, ....):
    . . .
    if add_mean:
        _add_mean_to_plot( . . . )
    . . .

WDYT?

[jsnyde0]

Hi @juanitorduz, sounds fine with me. Will seemed to indicate he rather avoids methods with many arguments, which led to my previous proposal.

Going on vacation for a week but I can get started on a draft.

Thanks for your input!

[juanitorduz]

Sure! Take your time! ☀️ 🌴 !

[jsnyde0]

I've implemented this and created an issue for it: #1057 Expanding plotting options of plot_posterior_predictive.

Will continue discussion there, I guess?