Tooltip example doesn't work with data aggregate transforms

I was trying to add a scatterplot tooltip based on the Multi-Line Tooltip example and think I ran across a bug. The example works fine when data are plotted directly but exhibits weird behavior when an aggregation function is applied.

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

# Create data and labels
a = np.random.randint(-5, 5, 200)
b = a * np.random.random(200)
s = np.random.choice(['first', 'second', 'third', 'fourth'], 200)
data = pd.DataFrame.from_dict({'a': a, 'b': b, 's': s}, orient='columns')

# Create a selection that chooses the nearest point & selects based on x-value
nearest = alt.selection_single(name='nearest', nearest=True, on='mouseover',  empty='none')

# Plot the mean of a against the mean of b
points = alt.Chart().mark_point(filled=True, size=100).encode(
    x='mean(a):Q',
    y='mean(b):Q',
    color='s:N'
)

# Transparent selectors across the chart. This is what tells us
# the x-value of the cursor
selectors = alt.Chart().mark_point().encode(
    x='mean(a):Q',
    y='mean(b):Q',
    opacity=alt.value(0),
).properties(
    selection=nearest
)

# Draw text labels near the points, and highlight based on selection
text = points.mark_text(align='left', dx=5, dy=-5).encode(
    text=alt.condition(nearest, 's:N', alt.value(' '))
)

alt.layer(points, selectors, text, data=data)

[jakevdp]

If your goal is to show tooltips, you could use the tooltip encoding channel. It’s not fully operational yet (there is some current work in the renderers that will make the tooltip implementation much better in the near future) but it works using the HTML title attribute now. For example:

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

# Create data and labels
a = np.random.randint(-5, 5, 200)
b = a * np.random.random(200)
s = np.random.choice(['first', 'second', 'third', 'fourth'], 200)
data = pd.DataFrame.from_dict({'a': a, 'b': b, 's': s}, orient='columns')

# Plot the mean of a against the mean of b
alt.Chart(data).mark_point(filled=True, size=100).encode(
    x='mean(a):Q',
    y='mean(b):Q',
    color='s:N',
    tooltip='s:N'
)

The whole thing about using selections and layers should only be used if you want to do something more sophisticated.

[gschivley]

Additional documentation for anyone who comes across this.

Another thing about your example: there’s no reason to add transparent selectors that are identical to your points. Just tie the selection to the points themselves.

The reason that the example you refer to uses transparent selectors is that we want to select on just the x-value, not the x and y value. There’s an example of a simpler nearest selection here: https://altair-viz.github.io/gallery/us_state_capitals.html

From vega-lite documentation it looks like the transparent selectors are used because the nearest transform isn’t supported for continuous mark types (lines and area).