The KDE plot is strange when extent is not specified in transform_density

[j4ggr]

When I use the given example for a KDE plot, I also get the expected result:

import altair as alt
from vega_datasets import data

source = data.iris()

alt.Chart(source).transform_fold(
    ['petalWidth',
     'petalLength',
     'sepalWidth',
     'sepalLength'],
    as_ = ['Measurement_type', 'value']
).transform_density(
    density='value',
    bandwidth=0.3,
    groupby=['Measurement_type'],
    extent=[0, 8]
).mark_area().encode(
    alt.X('value:Q'),
    alt.Y('density:Q'),
    alt.Color('Measurement_type:N')
)

But as soon as I remove the optional extent specification, the chart looks pretty bad:

import altair as alt
from vega_datasets import data

source = data.iris()

alt.Chart(source).transform_fold(
    ['petalWidth',
     'petalLength',
     'sepalWidth',
     'sepalLength'],
    as_ = ['Measurement_type', 'value']
).transform_density(
    density='value',
    bandwidth=0.3,
    groupby=['Measurement_type']
).mark_area().encode(
    alt.X('value:Q'),
    alt.Y('density:Q'),
    alt.Color('Measurement_type:N')
)

Versions

Python: 3.11.1
Altair: 5.1.1

[jonmmease]

Thanks for the report @j4ggr, I think this likely has the same root cause as vega/vega-lite#9078. The fix for this has been merged into Vega-Lite in vega/vega-lite#9106, but we need to wait for a Vega-Lite release in order to update Altair with the fix.

[joelostblom]

This is working as expected in altair 5.1.2 so closing this issue

[marianokamp]

Still broken for me @joelostblom. How did you verify that it does not break anymore? My steps to reproduce:

• Install 5.1.2 (or 5.2.0 or 5.3.0)
• Restart Jupyter Kernel
• Execute 2nd code snippet.

[joelostblom]

@marianokamp If you recently installed the new version of ALtair, make sure you clear all outputs in the notebook and then restart jupyterlab (not just the kernel). Otherwise an old version of VegaLite might still be used in the background.

[marianokamp]

@joelostblom Re-installed 5.3, then restarted jupyterlab (not just the kernel): Now the 2nd code example, without extent, works. Thank you.