Generating colors from data, by RGB channels

[dcbb]

Is it possible to get a color for a mark by specifying three different numerical columns as red, green and blue components of the mark's color?

Something like:

alt.Chart(data).mark_point().encode(
   x="x", 
   y="y",
   color=alt.Color(red="r", green="g", blue="b")
)

...where "r", "g", and "b" are numerical columns in data.

Excessive googling suggests that it does not seem possible, but I'm hoping that I have overlooked something.

Fun fact: I asked this to ChatGPT, and it very confidently suggests that there is an aggregate function in altair named rgb, so you can just go Color("rgb(r,g,b)"). Nice illustration of current AI limits.

[binste]

It is possible using so called expressions, see Expressions for interaction for an introduction. ChatGPT was onto something but it's not quite correct :)

import altair as alt
import pandas as pd

df = pd.DataFrame(
    [[1, 1, 102, 204, 0], [2, 2, 0, 0, 255], [3, 3, 0, 0, 0]],
    columns=["x", "y", "r", "g", "b"],
)

alt.Chart(df).mark_circle(size=200).encode(
    x="x",
    y="y",
    # In Altair 5 you will be able to replace alt.ExprRef with alt.expr
    color=alt.value(alt.ExprRef(alt.expr.rgb(alt.datum.r, alt.datum.g, alt.datum.b))),
    # Or you can write the expression as a string like this
    # color=alt.value(alt.ExprRef("rgb(datum.r, datum.g, datum.b)")),
)

[dcbb]

Works like a charm, thanks a lot! Now I don't have to revert to matplotlib for that plot.

Those "code in string" hacks are creepy but so powerful.