specifying domain/range for color scale puts everything from domain/range in legend even if it's not used

[jowens]

import altair as alt
import pandas as pd

domain = [
    "A",
    "B",
    "C",
]
range = [
    "#4daf4a",
    "#984ea3",
    "#ff7f00",
]

df = pd.DataFrame({"x": [1, 2, 3], "y": [3, 2, 1], "label": ["A", "B", "A"]})

alt.Chart(df).mark_point().encode(
    x="x",
    y="y",
    color=alt.Color("label", scale=alt.Scale(domain=domain, range=range)),
    shape="label",
)

If every label is represented:

df = pd.DataFrame({"x": [1, 2, 3], "y": [3, 2, 1], "label": ["A", "B", "C"]})

alt.Chart(df).mark_point().encode(
    x="x",
    y="y",
    color=alt.Color("label", scale=alt.Scale(domain=domain, range=range)),
    shape="label",
)

I would prefer that

• Labels that aren't in the input data don't appear in the legend
• The color and scale legends are combined into one legend (as they would be without an explicit domain/range)

Thanks for lookin'!

[jakevdp]

The first point is by design: specifying a domain and range is how you make the legend represent points that are not in the data (for example, when dynamically filtering).

The second point is a known issue; see #1785

[jowens]

OK. Other than filtering the domain and range Scale arrays once I see the data, any other way to make it reflect the actual contents of the data? (Might be potentially interesting to make this an option to Scale, legendOnlyReflectsDataBool, that sort of thing.)

(The reason I'm using a domain/range is to get the colors consistent between different plots for the same labels. If there's another way to do this, I'm all ears.)

Thanks.

[jakevdp]

I don't know of any way to make an explicit domain/range respond to the contents of the data. You might ask on the vega-lite lists.

[jowens]

@jakevdp, @kanitw has a vega-lite hack (using "signal") that you might want to peek at, https://github.com/vega/vega-lite/issues/6310#issuecomment-612121376

[joelostblom]

Since this is tracked in https://github.com/vega/vega/issues/2535 and https://github.com/altair-viz/altair/issues/1785, and it needs to be implemented in VL for it to become available in altair I'm closing this issue.