Align labels of Bricks when subplot axes are aligned

[W-L]

Hi ponnhide!
Big fan of your library, very happy that I came across it. I noticed that sometimes labels of Bricks are not quite aligned when the axes of subplots are aligned, but the bounding box of one of the subplots is larger. Here's a simple example:

import numpy as np
import pandas as pd
import plotnine as p9
import patchworklib as pw
np.random.seed(1)


data_a = np.random.normal(loc=1, scale=1, size=1000)
data_b = np.random.normal(loc=1, scale=1, size=1000)
data_c = np.random.normal(loc=1e8, scale=1, size=1000)

df = pd.DataFrame({"data_a": data_a, "data_b": data_b, "data_c": data_c})

plot0 = (p9.ggplot() + p9.geom_point(data=df, mapping=p9.aes(x='data_a', y='data_b')))
plot1 = (p9.ggplot() + p9.geom_point(data=df, mapping=p9.aes(x='data_a', y='data_c')))

plots = [plot0, plot1]
n_plots = len(plots)

m = (2, 2)
bricks = [pw.load_ggplot(p, figsize=m) for p in plots]


letters = ['A', 'B']
for i in range(n_plots):
    bricks[i].set_index(letters[i])


layout = pw.stack(bricks, operator="/")

I was wondering if there was a way to align the labels of the subplots in this case, i.e. align the A and B labels on the left?

As a workaround, I tried to use layout.get_outer_corner()[0] or layout.get_outer_corner()[0] - bricks[i].get_outer_corner()[0] to adjust the x of set_index(x=...), but was not successful. Any tips how that could be done?

Thanks very much!

[W-L]

Hi, thanks for your answer (which I can only see in my emails for some reason), and for implementing a solution.
I installed the development version using pip install git+https://github.com/ponnhide/patchworklib.git and tested the new layout.align_ylabels(), which worked for the simple example I posted above.

However, when I tried it on a slightly more complicated plot, I noticed some new issues that seemed to be caused by this version. More specifically, when using facets weird things happen to the labels of plots, even without using the new align_ylabels() method.

Here's an example:

import pandas as pd
import numpy as np
from plotnine import *
import patchworklib as pw
print(pw.__version__)

data_a = np.random.normal(loc=1, scale=1, size=1000)
data_b = np.random.normal(loc=1, scale=1, size=1000)
data_c = np.random.normal(loc=1e8, scale=1, size=1000)
data_d = np.random.choice(["a", "b", "c", "d"], size=1000)

df = pd.DataFrame({"data_a": data_a, "data_b": data_b, "data_c": data_c, "data_d": data_d})
df['data_d'] = df['data_d'].astype("category")

p0 = (ggplot()
      + geom_line(data=df, mapping=aes(y="data_b", x='data_a', color="data_d"))
      + facet_grid('~data_d', scales="free"))

p1 = (ggplot()
      + geom_line(data=df, mapping=aes(y="data_c", x='data_a', color="data_d"))
      + facet_grid('~data_d', scales="free"))

plots = [p0, p1]
bricks = [pw.load_ggplot(p) for p in plots]

letters = ['A', 'B']
for i in range(len(letters)):
    bricks[i].set_index(letters[i])

layout = pw.stack(bricks, operator="/")

This is the plot with version 0.5.2

and this is the same code with development version 0.6.1

the y-labels are duplicated and the right-most facet gets it's own x-label too.

When using align_ylabels() in version 0.6.1 with this same example, the duplication happens in the same way, but the alignment seems to work

[ponnhide]

Sorry, maybe I solved this problem. Please try same one using "pip install git+https://github.com/ponnhide/patchworklib.git"

[W-L]

Thanks! The issue of duplicated axis labels disappeared again with your update. But unfortunately the alignment also does not work in this example. The plots look the same whether layout.align_ylabels() is used or not. It does however work in the simpler example of my original comment.

[ponnhide]

Sorry, I did not consider for alining y-labels of subplots. It is difficult a little. When I have time, I try to implement this function.
By the way, I decided to quit support plotnine anymore because there were too many internal changes in plotnine v0.12.1.
I'm deeply sorry. If you wanna continue to use plotnine, please use plotnine v0.10.1.

[ponnhide]

In the GitHub version, align_ylabels and align_xlables probably work for the complex example.