Layered Histogram fails to represent null values

Issue

The method for making layered histograms suggested on altair-viz.github.io seems to fail to take into account null values within the range of the data: bins that should be empty are represented as having 1 observation.

---

Examples

Altair:

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

np.random.seed(42)

# Generating Data
source = pd.DataFrame({
    'Trial A': np.random.normal(0, 0.8, 1000),
    'Trial B': np.random.normal(-2, 1, 1000),
    'Trial C': np.random.normal(3, 2, 1000)
})

alt.Chart(source).transform_fold(
    ['Trial A', 'Trial B', 'Trial C'],
    as_=['Experiment', 'Measurement']
).mark_area(
    opacity=0.3,
    interpolate='step'
).encode(
    alt.X('Measurement:Q', bin=alt.Bin(maxbins=100)),
    alt.Y('count()', stack=None),
    alt.Color('Experiment:N')
)

Seaborn:

import seaborn as sns

source_melt = source.melt(var_name='Experiment', value_name='Measurement')
sns.displot(
    source_melt,
    x='Measurement',
    binwidth=.2,
    hue='Experiment',
    aspect=2,
    element='step'
)

---

Solutions

The easiest solution would be to suggest another method. The one below solves the null values in bins, but is much more verbose, and requires manually setting the x-axis and the legend.

trial_a = alt.Chart(source).mark_bar(
    opacity=0.5, color='#c9d6e5'
).encode(
    alt.X('Trial A:Q', bin=alt.Bin(maxbins=50), axis=None),
    y='count()'
)

trial_b = alt.Chart(source).mark_bar(
    opacity=0.5, color='#fcdab9'
).encode(
    alt.X('Trial B:Q', bin=alt.Bin(maxbins=50), axis=None),
    y='count()'
)

trial_c = alt.Chart(source).mark_bar(
    opacity=0.5, color='#f7cccc'
).encode(
    alt.X('Trial C:Q', bin=alt.Bin(maxbins=100), axis=None),
    y='count()'
)

trial_a + trial_b + trial_c

Another would be to fix the issue. If this is a vega-lite issue, I can move it there, but thought I’d get an opinion here first.

[joelostblom]

@harabat Yup that is the right place. After you have made the changes you can build the documentation locally to make sure that it shows up as you indended with make html from the doc dir. To make the histogram look like an area without the whitelines between the bars, you can set the stroke encoding to the same as the color or maybe even better, set binSpacing=0 inside mark_bar. I have seen a couple of other issues about the step-interpolation not working exactly as a histogram, so I think you can make your suggested change to this example as well https://altair-viz.github.io/gallery/scatter_marginal_hist.html

[joelostblom]

This was resolved in #2419