Go over the rest of the plotting functions so they obey the mpl color scheme

continuing #3112 , get rid of all the shocking blue in visualization.rst

scatter_plot still uses the evil blue, for one.
Also, grid defaults.

[jreback]

@cpcloud if you are interested

or maybe cc @jorisvandenbossche as you seem to like these kinds of things!

[jorisvandenbossche]

I think the only remaining 'evil blues' on visualisation.html are for scatter_matrix and lag_plot. They both use the scatter method from matplotlib. Some observations:

• The reason is because c='b' is hardcoded in the matplotlib scatter function (see here), and thus not following the color in the settings (rcParams['patch.facecolor']). So changing this settings does not affect the scatter plot.
• I think this can be seen as a bug/inconvenience in matplotlib => maybe issue there
• We could try to circumvent this in pandas:
  • explicitly set color or facecolor equal to to matplotlib.rcParams['patch.facecolor']
  • however, I don't directly see how to do that without getting into problems when the user also sets a color in the functions (scatter_matrix(..., c= ..) because this is then leads to 'keyword argument repeated'

[jorisvandenbossche]

Ah, for the last point, I suppose we could simply check if the user included c in de **kwds, and then using that color.

[jtratner]

@jorisvandenbossche sounds like a job for dict.setdefault!

[cpcloud]

also we should use color and not c

[jorisvandenbossche]

@jtratner you mean just kwds.setdefault('c', matplotlib.rcParams['patch.facecolor'])?

@cpcloud Would you add a color keyword? Because at the moment, there is no color keyword in scatter_matrix. If you want to change the color of the markers, you just have to use the passing of c to the matplotlib scatter via **kwds.
But that's a seperate issue I think, because als eg bootstrap_plot, lag_plot, autocorrelation_plot, have no color keyword, but it depends on the underlying matplotlib function if setting color would work or not.

[cpcloud]

@jorisvandenbossche c or color is accepted by scatter, it's actually already "there" since kwds is passed to scatter, so you can use either c or color when you're calling scatter_matrix. The problem is that you break that "feature" of calling scatter if you choose either c or color so you need to account for the case where someone would pass both (pathological, but possible). I would enforce the color keyword parameter.

[jorisvandenbossche]

@cpcloud Ah, OK, I didn't know that you could use color in scatter (not really clear from the docstring or code).

Isn't it then enough to just check both and set to default value if necessary? 'color' is already given preference by matplotlib if both are specified

kwds.setdefault('c', matplotlib.rcParams['patch.facecolor'])
kwds.setdefault('color', matplotlib.rcParams['patch.facecolor'])

[jreback]

@jorisvandenbossche is this still an issue?

[jorisvandenbossche]

@jreback yes it still is, but I think it is an easy fix (if what i said above is true), I will do a PR.