Logarithmic scale not showing any results about vega-lite

[kgoodrick-uu]

By default, the bar chart draws a bar from 0 to the y encoding value. It's not possible to include 0 in a logarithmic scale, so it does not draw anything. You can fix this by giving vega-lite an alternative starting location for the bar that can be represented on a logarithmic scale. You can do this by providing a constant "y2" encoding e.g.

"y2": {"datum": 0.1}

Working example here.

[mscolnick]

Thanks for explaining. But in my example, there is no datum with a y value of 0. Am I missing something?

[kgoodrick-uu]

Hi @mscolnick, the 0 is implied with bar mark which is generally desirable as almost all bar charts go from 0 to the value. Since this is not possible with a logarithmic scale you have to tell vega-lite what you want it to do by manually setting a datum that can be represented on a log scale.

[mscolnick]

Got it - thanks for clarifying.

What's weird is that when Altair renders this, it works, but the "Open in Vega Editor" still does not work.
(p.s. I will file this oddity in the Altair github)

For example, here is a reproduction: https://marimo.app/l/nhwpi8

import altair as alt
import pandas as pd
data = {"x": range(1, 11), "y": [5, 10, 2, 8, 3, 15, 1, 4, 9, 6]}
df = pd.DataFrame(data)

(
    alt.Chart(df)
    .mark_bar(stroke="black", strokeWidth=0.5)
    .encode(
        x=alt.X(
            "x",
        ),
        y=alt.Y(
            "y",
            scale=alt.Scale(type="log"),
        ),
    )
)

But the produced vega playground, that can be found here, still does not render (due to the y=0 issue mentioned above).

[PBI-David]

For some reason, the same spec in the online editor introduces a stack transform in the compiled Vega which is what is causing the issue. Maybe one of the devs will know why.

[mscolnick]

Thank you - I filed an issue in the Altair repo: vega/altair#3447

EDIT: They believe this (vega-lite) is the is the correct place for the issue

[domoritz]

What solution do you propose? Not implying automatic zero for bars when the scale is log?

[mscolnick]

@domoritz - i'm not familiar enough with vega-lite to propose a solution, but is this not a regression? is anything wrong with the previous implementation?