Specify axis ordering for specific facet in multi-facet plot

[tr8dr]

Correct me if I am wrong, but it is not possible(?) to create a multi-facet plot with continuous facets where one of the facets reverses the x (y) axis in a particular facet. For example if I have:

df = pd.concat([
    pd.DataFrame({'x': [1, 2, 3, 4, 5, 6, 7, 8], 'level': [2, 4, 9, 25, 36, 49, 64], 'pane': 'left'}),
    pd.DataFrame({'x': [1, 2, 3, 4, 5, 6, 7, 8], 'level': [1, 3, 8, 24, 30, 40, 55], 'pane': 'right'})
])

(ggplot() + 
    geom_area(aes(x='x', y='level', fill='pane'), data=df) +
    facet_grid(". ~ pane"))

Supposing that I want the 'left' pane to be ordered from 8 downto 1 and the 'right' ordered from 1 to 8, is it possible to accomplish this in the same ggplot()?

[has2k1]

You can do it in one of two ways.

1. Drop into matplotlib and reverse the limits of the 2nd facet
2. Create your own coordinate system that reverses the limits of the facet.

import pandas as pd
from plotnine import *

df = pd.concat([
    pd.DataFrame({'x': [1, 2, 3, 4, 5, 6, 7], 'level': [2, 4, 9, 25, 36, 49, 64], 'pane': 'left'}),
    pd.DataFrame({'x': [1, 2, 3, 4, 5, 6, 7], 'level': [1, 3, 8, 24, 30, 40, 55], 'pane': 'right'})
])

# Option 1
p = (ggplot()
 + geom_area(aes(x='x', y='level', fill='pane'), df)
 + facet_wrap("pane", scales='free_y')
 + theme(subplots_adjust={'wspace': 0.20})
)

fig = p.draw()
lax, rax = fig.get_axes()
rax.set_ylim(rax.get_ylim()[::-1])



# Option 2
def alt_sign_gen():
    while True:
        yield 1
        yield -1

alt_sign = alt_sign_gen()

class coord_alt_reverse(coord_cartesian):
        def setup_panel_params(self, scale_x, scale_y):
            r = super().setup_panel_params(scale_x, scale_y)
            sign = next(alt_sign)
            r.y.limits = r.y.limits[::sign]
            r.y.range = r.y.range[::sign]
            return r

(ggplot()
 + geom_area(aes(x='x', y='level', fill='pane'), df)
 + facet_wrap("pane", scales='free_y')
 + coord_alt_reverse()
 + theme(subplots_adjust={'wspace': 0.20})
)

[tr8dr]

Thanks! that is useful. I "solved" previously by rendering 2 ggplots and pasting together with a multi-panel figure. I wonder whether it makes sense to allow for for thematic or coordinate changes on a per facet basis as a feature in plotnine or perhaps this is too niche.

[has2k1]

Yes it is niche, but the solution is to make it easy to combine multiple ggplots.