Boxplots with filter transform based on selection don't work

[carusov]

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!

[baogianghoangvu]

Hi. I ran your boxplot code and checked the browser console and found this warning: WARN Selection not supported for boxplot yet.. This is probably an issue from Vega-lite and why your boxplot doesn't render.

If you haven't tried this yet, I think an ok alternative to the dropdown list is to have all 3 metrics as columns like so

alt.Chart(metrics).mark_boxplot(size=30).encode(
    x='group',
    y='value',
    color='group',
    column='metric'
).properties(width=100)

[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.

[joelostblom]

@carusov This is being tracked in vega/vega-lite#3702 and can become available in Altair after it is implemented in Vega-Lite.

[devliegherec]

Hi @joelostblom, is there by any chance an update regarding the interactive boxplot issue? I would like to be able to zoom in on my boxplot but the interactive function does not work, also adding a scale does not give the desired results (scale= {"domain": [0,50]})

[joelostblom]

Hi @devliegherec, this will only be available in Altair after the functionality has been added to Vega-Lite, so please follow vega/vega-lite#3702.

Regarding zooming and domain not working, could you create a minimal reproducible example and post your question (ideally on stack overflow tagging it with 'altair') and I will see if there is a workaround?

[joelostblom]

Closing this as there is nothing to do on the Altair side of things, when this function is added to VL, it will be in an upcoming Altair release.

[joelostblom]

Although boxplots still don't support .interactive or direct selections, it is possible to have a dropdown menu that changes the column that is plotted as you asked initially @carusov. This is based on the new example here in the docs:

import altair as alt
from vega_datasets import data


dropdown = alt.binding_select(
    options=['Miles_per_Gallon', 'Displacement', 'Weight_in_lbs', 'Acceleration'],
    name='X-axis column '
)
xcol_param = alt.param(
    value='Miles_per_Gallon',
    bind=dropdown
)

alt.Chart(data.cars.url).mark_boxplot().encode(
    x=alt.X('x:Q', title=''),
    y='Origin:N',
    color='Origin:N'
).transform_calculate(
    x=f'datum[{xcol_param.name}]'
).add_params(
    xcol_param
)

[carusov]

Thanks for the update @joelostblom. I'm excited about the new features in the coming 5.0 release. I'm a long-time and heavy user of Altair, so I'm very grateful to you and all the contributors for continuing to develop and extend Vega-Altair.