How to set a proportional bar width that scales nicely?

[quantoid]

Bar charts have a fairly narrow bar width by default, but if you configure a larger pixel width you get overlapping bars when users resize their browser window. What I need is to set the bar width to say 0.7 of the space between ticks, so they grow and shrink with the chart size and always have a gap between them.

In Vega-lite you can do it like this:

"mark": {"type": "bar", "width": {"band": 0.7}},

However, I can't find any way to do that in Altair, e.g. this throws an exception:

Chart(data).mark_bar(width={"band": 0.7})

How do you specify proportional bar widths?

[mattijn]

Once the next version of Altair is released you can use width=alt.RelativeBandSize(0.7). Using the main repo of altair this gives me currently:

import altair as alt
from vega_datasets import data

source = data.seattle_weather()

alt.Chart(source).mark_bar(
    width=alt.RelativeBandSize(0.7)
).encode(
    x='month(date):O',
    y='mean(precipitation):Q'    
)

[quantoid]

Thanks, is the next release imminent? Or many moons away?

[quantoid]

@jakevdp is there a best-guess timeframe for the next release with this included?

[joelostblom]

In addition to the answer above, you can also change the overall chart width, which will expand the bars. Using a step size indicates how many pixels you want to use for the width of each categorical axis value (ie each bar):

import altair as alt
from vega_datasets import data

source = data.seattle_weather()

alt.Chart(source, width=alt.Step(30)).mark_bar().encode(
    x='month(date):O',
    y='mean(precipitation):Q'    
)

[quantoid]

Thanks, I considered that but unfortunately I'm using streamlit.altair_chart with use_container_width=True so the chart width varies.

[DanielFortuyn]

Was this ever resolved? I use streamlit too. When I update the data of a chart the bars remain the same size. it would be great if they could scale with the amount of x categories. The suggested solution does not work for me at the moment.

[joelostblom]

Does the accepted solution with RelativeBandSize not work for you?