sort argument doesn't sort categories/colors in stacked bar

[gschivley]

There are several places indicating that the argument sort can be used universally to set a custom ordering by passing an array (list) with the desired order. This works for categories on the x-axis of a grouped bar chart but not the y-axis.

Current versions:

# Name                    Version                   Build  Channel
altair                    4.0.1                      py_0    conda-forge
jupyterlab                1.2.6                      py_0    conda-forge

Example (vega-lite spec):

import pandas as pd
import altair as alt

resource_order = [
    "Onshore Wind",
    "Solar",
    "Battery",
    "NGCC",
]
resource_color_scale = alt.Scale(
    domain=resource_order, 
    range=[
        "#17becf", # onshore wind
        "#d62728",  # solar 
        '#c5b0d5', # battery
        "#bcbd22", # NGCC
    ]
)
resource_colors = alt.Color("Resource Name", scale=resource_color_scale)
    
data = pd.DataFrame(
    {
        "Resource Name": ["Battery", "NGCC", "Onshore Wind", "Solar"] * 2, 
         "Capacity": [2, 20, 10, 15, 3, 10, 30, 25], 
         "Case": ["A", "A", "A", "A", "B", "B", "B", "B"]
    }
)

alt.Chart(data).mark_bar().encode(
    x=alt.X('Case', sort=["B", "A"]), # Default order is alphabetical "A", "B"
    y=alt.Y('Capacity', sort=resource_order),
    color=resource_colors
)

I did, however, find a workaround by ordering according to a new indexing column (vega-lite spec):

resource_order_idx = {
    resource: idx 
    for idx, resource in enumerate(resource_order[::-1]) # Reverse list to align colors
}                                                        # with legend order

# Create "idx" column with integer values indicating order in stacked bar
data["idx"] = data["Resource Name"].map(resource_order_idx)

alt.Chart(data).mark_bar().encode(
    x=alt.X('Case', sort=["B", "A"]),
    y=alt.Y('Capacity'),
    order="idx",
    color=resource_colors
)

I'm not sure if this is a bug in vega-lite, an oversight in the API, or the way that it is supposed to work, but I'm providing this example and my work-around to help anyone else who might be searching for for help with the same issue.

[jakevdp]

The order encoding is the documented way to specify sort order for stacked bars, so I believe this is working as intended.

The docs on the Altair side, as always, could be improved.

[jakevdp]

The y encoding's sort property applies to the axis, not to the marks that appear on the chart. To see why this is not a sensical place to specify stack order, imagine a horizontally-concatenated chart showing two datasets with different stack orders but the same numerical y scale: specifying two different stack orders via one shared axis would not be possible, but each panel can have its own order encoding without any problem.

[gschivley]

To see why this is not a sensical place to specify stack order, imagine a horizontally-concatenated chart showing two datasets with different stack orders but the same numerical y scale: specifying two different stack orders via one shared axis would not be possible, but each panel can have its own order encoding without any problem.

Ok, I think I see what you mean. A simple way to order stacked bars that accepts an array would be more intuitive (and easier) but it seems that's a Vega-Lite limitation.

As always, thanks for all the effort!

[jakevdp]

I believe this issue tracks the functionality you want.

[gilzero]

I came across this 'issue' (perhaps not a issue, but a feature in vega-lite), as suggested above, a work around is to assign a new order column with quantitative values after preprocess with whatever required order in dataframe:

a quick example:

import pandas as pd
import numpy as np

data = [
    {"name": "nike", "share": 61, },
    {"name": "adidias", "share": 30, },
    {"name": "asic", "share": 6, },
    {"name": "reebok", "share": 3, }
]

data = pd.DataFrame(data)

df = pd.DataFrame(data).sort_values(by='share', ascending=False)

df['order'] = np.arange(0, df.shape[0])

df

bars = alt.Chart(data).mark_bar().encode(
    x=alt.X('share:Q', 
            stack='zero', 
           ),
    color=alt.Color('name',
                   sort=df.name.values,
                   ),
    order='order:Q'
)


text = alt.Chart(data).mark_text(dx=-8, dy=3, color='white').encode(
    x=alt.X('share:Q', 
            stack='zero', 
           ),
    text=alt.Text('share:Q'),
    order='order:Q'
)


bars + text

(recall above comments that sort is applying to axis, e.g the legend name follow that specified 'sort=', the actual bar follows 'order=')

further ref:
https://altair-viz.github.io/user_guide/generated/channels/altair.Order.html?highlight=order#altair.Order