Allow to use a non fixed axes formatter

[doronbehar]

Hello,

I am a Python user that came from using plain Matplotlib (not an R user switching to Python). I noticed today that if I take e.g this plot:

And zoom in:

The x & y ticks are fixed! I tried to figure out from reading the code whether I can change this, and I encountered this in plotnine's code:

plotnine/plotnine/_mpl/ticker.py

Lines 6 to 16 in 0665309

	class MyFixedFormatter(FixedFormatter):
	"""
	Override MPL fixedformatter for better formatting
	"""
	def format_data(self, value: float) -> str:
	"""
	Return a formatted string representation of a number.
	"""
	s = locale.format_string("%1.10e", (value,))
	return self.fix_minus(s)

Which is used here:

plotnine/plotnine/facets/facet.py

Lines 351 to 352 in 0665309

	ax.xaxis.set_major_formatter(MyFixedFormatter(panel_params.x.labels))
	ax.yaxis.set_major_formatter(MyFixedFormatter(panel_params.y.labels))

I am not sure whether the graphics grammar allows modifying this behavior, but I won't mind doing it via Matplotlib's native functions. However, that turned out to be hard and frustrating, as plot.draw() returns a figure, but I can't modify the axes of it. Here's something I tried:

from matplotlib.ticker import (
    EngFormatter,
    AutoLocator,
)
for ax in plot.draw(show=False).axes:
    ax.xaxis.set_major_locator(AutoLocator())
    ax.yaxis.set_major_locator(AutoLocator())
    ax.xaxis.set_major_formatter(EngFormatter())
    ax.yaxis.set_major_formatter(EngFormatter())
plot.show() # Also tried `matplotlib.pyplot.show()` here too.

And it doesn't show anything. As a side note, this made me wonder: What is the benefit in returning the figure if it cannot be manipulated afterwards with plain-old Matplotlib? This is related of course to:

• How to draw plotnine figures on a existing matplotlib ax object? #982
• strange things happen when using .draw() #1009

Anyway, I managed to workaround this with the following ugly monkey-patching of ggplot class:

from matplotlib.ticker import (
    EngFormatter,
    AutoLocator,
)
ggplot_original__draw_breaks_and_labels = ggplot._draw_breaks_and_labels
def ggplot__draw_breaks_and_labels_with_my_ax_manipulation(self):
    ggplot_original__draw_breaks_and_labels(self)
    for ax in self.axs:
        ax.xaxis.set_major_formatter(EngFormatter(unit="Hz"))
        ax.yaxis.set_major_formatter(EngFormatter(unit="Hz"))
        ax.xaxis.set_major_locator(AutoLocator())
        ax.yaxis.set_major_locator(AutoLocator())
ggplot._draw_breaks_and_labels = ggplot__draw_breaks_and_labels_with_my_ax_manipulation

Not sure why the ticks are located in fixed positions just due to the major_formatter set above, but for sure the AutoLocator is needed too.

[has2k1]

plotnine uses matplotlib mostly as a canvas and some aspects of MPL are completely side stepped. Everything in matplotlib.ticker is side stepped. The things you can safely do on the figure are to draw extra artists on it.

I'll look into whether there is a way to wire up the zoom to work.

[doronbehar]

That's a bit disappointing :/ What is wrong with MPL's default ticker behavior? Why does it have to be side stepped? I understand that if someone wishes to specify ticker parameters explicitly with the graphics grammar, MPL's default behavior has to be sidestepped, but otherwise I don't see a reason for this at all.