Highlighting points on a faceted chart

the following is adapted from https://stackoverflow.com/a/54965673/2981639

I want to highlight specific points on a chart. The way I’ve seen this done is to create 2 layers with the first plotting all points and the 2nd plotting a subset in a different style i.e. cluster members colored by cluster and the cluster centers as black points in the linked example.

I also want to facet the chart, i.e. using columns= and I understand from https://github.com/altair-viz/altair/issues/668 that the data should be passed to the facet() function and this gets forwarded to the layers.

But if you want two layers with different data sources and a facet per layer what do you do?

import altair as alt
import pandas as pd
from sklearn.datasets import make_blobs

X, labels = make_blobs(20, random_state=1)
points = pd.DataFrame({
    'x': X[:, 0],
    'y': X[:, 1],
    'labels': labels
})
centers = points.groupby('labels').mean()
data = pd.concat([points , centers.reset_index()])

chart1 = alt.Chart(data).mark_point(filled=True, size=150).encode(
    x='x',
    y='y',
    color='labels:N',
    column='labels:N',
)

chart2 = alt.Chart(centers).mark_point(filled=True, size=50).encode(
    x='x',
    y='y',
    color=alt.value('black'),
    column='labels:N',
)

chart1 + chart2

[jakevdp]

As of Altair 3.1, the information in #668 is out of date: you need not specify the data in the facet method itself, so long as the layers both contain the same dataset.

As to your question: if you want to facet a layered chart, you’ll need to use a single dataset for both layers. This could be done by joining the two datasets and plotting different columns, or by concatenating the datasets and filtering on the appropriate property within each layer, or by using transforms to modify the single base dataset within subcharts. Here I think the third option is probably the cleanest approach. It might look something like this:

import altair as alt
import pandas as pd
from sklearn.datasets import make_blobs

X, labels = make_blobs(20, random_state=1)
points = pd.DataFrame({
    'x': X[:, 0],
    'y': X[:, 1],
    'labels': labels
})

chart1 = alt.Chart(points).mark_point(filled=True, size=150).encode(
    x='x',
    y='y',
    color='labels:N',
)

chart2 = alt.Chart(points).transform_aggregate(
    x='mean(x)',
    y='mean(y)',
    groupby=['labels']
).mark_point(filled=True, size=50).encode(
    x='x:Q',
    y='y:Q',
    color=alt.value('black'),
)

alt.layer(chart1, chart2).facet('labels:N')

[jakevdp]

It is possible, but the datasets need to be joined in the top layer…