Altair: Limit number of ordered facets

Question

This example code show 6 facets, ordered by their means. How could I display only e.g. 2 top or bottom facets, which are ordered this way? I tried playing around with the transform_window rank option with no luck, it limits the bar contents of all the facets instead.

import altair as alt
from vega_datasets import data

source = data.barley()

alt.Chart(source).mark_bar().encode(
    x='yield:Q',
    y='year:O',
    color='year:N',
    facet=alt.Facet('site',
        sort={
            'field': 'yield',
            'op': 'mean',
            'order': 'descending'
        },
        columns=1
    )
)

Answer 1

You could use a joinaggregate transform to calculate the site means, then rank them using a window transform, and finally filter the ranks using the filter transform (similar to this example in the docs):

import altair as alt
from vega_datasets import data


source = data.barley()

n_top_sites = 3

alt.Chart(source).mark_bar().encode(
    x='mean_yield_per_site:Q',
    y='year:O',
    color='year:N',
).facet(
    alt.Facet('site', sort=['mean_yield_per_site'], title=''),
    columns=1,
).transform_joinaggregate(
    mean_yield_per_site='mean(yield)',
    groupby=['site']
).transform_window(
    rank='dense_rank()',  # Don't skip any rank numbers, https://vega.github.io/vega-lite/docs/window.html#ops
    sort=[alt.SortField('mean_yield_per_site', order='descending')]
).transform_filter(
    f'datum.rank <= {n_top_sites}'
)

The same result could be achieved by doing the computation of the top site index labels with pandas and then filter for membership in that list via altair's filter transform:

n_top_sites = 2
top_sites = source.groupby('site').mean()['yield'].nlargest(n_top_sites).index.tolist()

alt.Chart(source).mark_bar().encode(
    x='yield:Q',
    y='year:O',
    color='year:N',
    facet=alt.Facet('site',
        sort={
            'field': 'yield',
            'op': 'mean',
            'order': 'descending'
        },
        columns=1
    )
).transform_filter(
    f'indexof({top_sites}, datum.site) != -1'
)