Substring Filtering in Altair

[foufoulides]

Hi all,

I need help with substring filtering in transform_filter() in an Altair chart.

• I have a dataset with data on various locations.
• Each location has its own comparable locations group (so it is not a partition of the locations where each location belongs in one and only one group).
• I want the user to be able to filter by each location's comparable group.
• I looked at Vega string expressions and I think using the the indexof() function in expression indexof(string, substring)>-1 (because it returns -1, when the substring is not found in the string) would do it, if I create a list of strings where each string is the concatenation of the locations in each comparable location group, with a dash separator.
• For some reason though it does not work and I get a black image, though it works when I hardcode in the function one comparable locations group/string.
• Can someone help with what I am doing wrong? A fully coded example is in the image below.

Thank you in advance,
Chris

Code copied from comment below

import pandas as pd
import altair as alt  

# Example DataFrame 
df = pd.DataFrame({'lctn': ['AAA', 'BBB', 'CCC', 'DDD', 'EEE', 'AAA', 'BBB', 'CCC', 'DDD', 'EEE'], 
                   'xval': [10, 4, 1, 9, 7, 1, 3, 4, 2, 8], 
                   'yval': [8, 7, 10, 9, 2, 6, 4, 3, 5, 1]} )

group_dropdown = alt.binding_select(options = sorted(['AAA-BBB', 'BBB-AAA-CCC', 'CCC-BBB-DDD', 'DDD-CCC-EEE', 'EEE-DDD']), labels=list(sorted(df["lctn"].dropna().unique())), name='Group') 
group_selection = alt.selection_single(name='group_sel', fields=['lctn'], bind=group_dropdown)

# This does not work for some reason 
alt.Chart(df).mark_point().encode(
    x=alt.X('xval'),
    y=alt.Y('yval'),
).add_selection(group_selection).transform_filter('indexof(group_sel.lcnt, datum.lctn)>-1').properties(width=80, height=80)

# Hardcoding one comparative groups location string works however 
alt.Chart(df).mark_point().encode(
    x=alt.X('xval'),
    y=alt.Y('yval'),
).add_selection(group_selection).transform_filter('indexof("BBB-AAA-CCC", datum.lctn)>-1').properties(width=80, height=80)

[mjohnson11]

Try writing group_sel.lctn[0] in the expression. This confused me a lot because I was just working on the opposite problem and indexof(datum.lctn, group_sel.lctn)>-1 works without the [0] index - which I think is just because of how indexof works in javascript: 'aaa-bbb'.indexOf(['bbb']) will work and yield 4.

Sorry if it was my question/answer on stack overflow that led you down the wrong path! Will update it!

[foufoulides]

Hi Milo,

Thank you for the prompt reply. I did find your post on stack overflow after not being able to make this work and I noticed that you made it work the other way around. So I could see I was on the right track, but couldn't understand why the opposite wouldn't work! I just tried what you suggested, but unfortunately the issue is still there.

Best,
Chris

[jakevdp]

Can you please copy and paste code as text into your question rather than pasting a screenshot of code?

[mjohnson11]

Just checked and these expression errors don't come through in vega-lite either, so if that improvement is ever made it would be there, not in altair. In general, I think whether and how javascript throws (or doesn't throw) these kinds of errors would probably make figuring out this kind of error handling pretty hard.

[foufoulides]

Hi Milo,

Thanks for spotting the typo! I actually changed the field name before submitting this here and I typed it wrong at that spot. Also fixed the typo above. I just tried your solution again. It takes a while to work for some reason, but it does work. The graph comes a bit "messed up" on the axes as seen below however.

[mjohnson11]

Try including init={'lctn': 'AAA-BBB'} in the selection definition. I'm not sure why it behaves strangely when init is not set.

[foufoulides]

That worked, thank you so much! Now that you mentioned this, I noticed in other instances as well that filters often don't work exactly right when initial/default value is not set, but not having such a big impact on the visual. They just need to be reselected to actually filter.

[mjohnson11]

No problem! I'm also currently learning how to use expressions in altair - powerful but tough to debug! On the selection initialization, the other important thing to pay attention to is probably the empty argument: https://altair-viz.github.io/user_guide/generated/api/altair.selection_single.html