Boxplots with filter transform based on selection don't work

I’m have a series of metrics that I’ve computed for different sample groups in my data, and I want to compare the distributions of a given metric with a boxplot for each sample group, using a dropdown list to choose the metric being compared. But when I do the following, I get a JavaScript Error:

import altair as alt, pandas as pd
import numpy as np

np.random.seed(431)

metrics = pd.DataFrame({
    'metric1': np.random.normal(30, 20, 100),
    'metric2': np.random.lognormal(2, 1, 100),
    'metric3': np.random.uniform(0, 80, 100),
    'group': np.random.choice(['A', 'B'], 100)
})

metrics = pd.melt(
    metrics,
    id_vars='group',
    var_name='metric',
    value_name='value'
)

group_dropdown = alt.binding_select(options=['metric1', 'metric2', 'metric3'])
group_selection = alt.selection_single(
    fields=['metric'], 
    bind=group_dropdown,
    name='Group ',
    init={'metric': 'metric1'}
)

distribution_boxes = alt.Chart(metrics).mark_boxplot(size=30).encode(
    x='group',
    y=alt.Y('value', title='metric1'),
    color='group',
).properties(
    width=150,
    height=300
).transform_filter(
    group_selection
).add_selection(
    group_selection
)

distribution_boxes

When I change the mark type to mark_tick(), the plot renders correctly:

distribution_ticks = alt.Chart(metrics).mark_tick(size=30).encode(
    x='group',
    y=alt.Y('value', title='metric1'),
    color='group',
).properties(
    width=150,
    height=300
).transform_filter(
    group_selection
).add_selection(
    group_selection
)

distribution_ticks

Is this an issue with Altair, or with vega-lite?

Thanks for your help!

[joelostblom]

@carusov This is being tracked in https://github.com/vega/vega-lite/issues/3702 and can become available in Altair after it is implemented in Vega-Lite.

[carusov]

Hi @baogianghoangvu, thanks for looking into this. I figured it probably wasn’t implemented in vega-lite yet, but I don’t know much about JS and I didn’t think to check the browser console- I’ll remember that next time I have an issue.

Your suggestion to use the column channel is a good one, but in my actual use case I have another variable that I’m faceting via columns. I guess I could use the row channel too, but I was trying to keep the chart from getting too busy.