How to use one y-label in a facet grid

[mcp292]

Coming here as advised in altair-viz/altair-viz.github.io#28. I will repeat everything here for convenience:

Using one label for an entire facet grid is not documented in Adjusting Axis Labels.

How can I make this plot only use one y-label?

MWE

import pandas as pd
import altair as alt


data = [{"P": "HD", "T": 0,  "C":100, "Max": 1},
        {"P": "HD", "T": 0,  "C":400, "Max": 3},
        {"P": "HD", "T": 0,  "C":800, "Max": 0},
        {"P": "HD", "T": 23, "C":100, "Max": 2},
        {"P": "HD", "T": 23, "C":400, "Max": 1},
        {"P": "HD", "T": 23, "C":800, "Max": 1},
        {"P": "HD", "T": 80, "C":100, "Max": 7},
        {"P": "HD", "T": 80, "C":400, "Max": 2},
        {"P": "HD", "T": 80, "C":800, "Max": 1},
        {"P": "LD", "T": 0,  "C":100, "Max": 1},
        {"P": "LD", "T": 0,  "C":400, "Max": 2},
        {"P": "LD", "T": 0,  "C":800, "Max": 7},
        {"P": "LD", "T": 23, "C":100, "Max": 7},
        {"P": "LD", "T": 23, "C":400, "Max": 1},
        {"P": "LD", "T": 23, "C":800, "Max": 0},
        {"P": "LD", "T": 80, "C":100, "Max": 2},
        {"P": "LD", "T": 80, "C":400, "Max": 0},
        {"P": "LD", "T": 80, "C":800, "Max": 1}]

df = pd.DataFrame(data)

chart = alt.Chart(df).mark_bar().encode(
    x="P:N",
    y=alt.Y("Max:Q", title="Maximum Long Word Observed (units)"),
    color="P:N",
    row="T:N",
    column="C:N"
).properties(width=100, height=100)

     P   T    C  Max
0   HD   0  100    1
1   HD   0  400    3
2   HD   0  800    0
3   HD  23  100    2
4   HD  23  400    1
5   HD  23  800    1
6   HD  80  100    7
7   HD  80  400    2
8   HD  80  800    1
9   LD   0  100    1
10  LD   0  400    2
11  LD   0  800    7
12  LD  23  100    7
13  LD  23  400    1
14  LD  23  800    0
15  LD  80  100    2
16  LD  80  400    0
17  LD  80  800    1

[joelostblom]

You can see a couple of approaches for this here https://stackoverflow.com/questions/71600310/make-collective-title-for-columned-altair-chart

[mcp292]

This is actually very workable thank you! I was hoping to change this on the label level though, not add a title. Is this possible to configure?

[mcp292]

I would still want the T as the outermost label on the y, labeling the 0, 23, 80 in my graph above.

[joelostblom]

You could place a free standing mark text as in one of the answers on the link above. Other than that, it might be possible to add newlines to you T title so that the second line ends up under the 0, 23, 80. Something like ['T', '', '', '', 'Maximum Long Work Hours']. Not sure if that will work

[mcp292]

Wow! It is not documented that you can pass an iterable to either header or title! This seems to work and is a very clever idea! See my answer below.

[mcp292]

Thanks to @joelostblom and his comment, this is one working solution:

chart = alt.Chart(df).mark_bar().encode(
    x="P:N",
    y=alt.Y("Max:Q", title=None),
    color="P:N",
    row=alt.Row("T:N", title=["Maximum Long Word Observed (units)", "", "T"]),
    column="C:N"
).properties(width=100, height=100)

This seems to work, but is not documented. If there becomes a built-in I would love to know. Or a more low-level answer that actually makes the y-axis one, that would be great and more consistent with the other plots in the paper. Therefore, I will leave this answer open unless otherwise advised.

Also, the fact that defining title in altair.Row() overrides the facet header (T) is not documented. Unless "axis/header/legend" is removed, it says this is true in the opposite case:

2. If both field definition’s title and axis, header, or legend title are defined, axis/header/legend title will be used.

Source

[mcp292]

The other problem with this is that the label is not in the correct position, up against the y-axis before the facet header labels, leaving room for misinterpretation.

[mcp292]

The ideal would be to edit this using top-level chart configuration such as chart.configure_header(); after the fact, not within the encoding definition.

[joelostblom]

Regarding having the label closer to the y-axis. You can probably play around with the padding parameters of both the facet header and the facet labels (80, 23, 0) to fix this, but it might take some fiddling.

[mcp292]

Sorry, not that it should be closer as-is, but that it should be interwoven as in the original post: y-axis scale, y-axis label, facet header label, facet header title.