support lists for encoding channel

[mattijn]

Dataframes are often categorized as wide-form data:

import pandas as pd
import numpy as np
dates = pd.date_range("20130101", periods=6)
df = pd.DataFrame(np.random.randn(6, 2), index=dates, columns=list("AB"))
df_alt = df.reset_index(names='date')
df_alt.head()

date	A	B
2013-01-01	-0.138047	0.942404
2013-01-02	-1.029359	-1.405696
2013-01-03	0.227675	0.558755
2013-01-04	-0.698109	-1.662299
2013-01-05	0.954903	0.134938

How to draw this as two line series?

There is a long section in the docs on this topic.

One should use melt in pandas, as such:

df_melt = df_alt.melt(
    'date',
    var_name='column_names',
    value_name='values',
    value_vars=['A', 'B']
)

alt.Chart(df_melt).mark_line().encode(
    x='date:T',
    y='values',
    color='column_names'
)

Or apply a transform_fold within altair:

alt.Chart(df_alt).transform_fold(
    as_=['column_names', 'values'], 
    fold=['A', 'B']
).mark_line().encode(
    x='date:T',
    y='values:Q',
    color='column_names:N'
)

I think both approaches are rather complicated and in practice I see people doing this:

c_A = alt.Chart(df_alt).mark_line().encode(
    x='date:T',
    y=alt.Y('A:Q').title('values'),
    color=alt.value('red')
)

c_B = alt.Chart(df_alt).mark_line().encode(
    x='date:T',
    y='B:Q',
    color=alt.value('blue')
)

c_A + c_B

Honestly, I think this approach is more intuitive than melting or folding.
This could already be improved by using alt.datum(<my-color>) to get a legend:

c_A = alt.Chart(df_alt).mark_line().encode(
    x='date:T',
    y=alt.Y('A:Q').title('values'),
    color=alt.datum('A')
)

c_B = alt.Chart(df_alt).mark_line().encode(
    x='date:T',
    y='B:Q',
    color=alt.datum('B')
)

c_A + c_B

Would it be strange to raise if the following could also be supported in the syntax of Altair?

alt.Chart(df_alt).mark_line().encode(
    x='date:T',
    y=['A', 'B']
)

Or is this completely against the grammar? For the tooltip encoding channel we do support a list as input as well.

---

Edit:
One could even do the following:

alt.Chart(df_alt).mark_line().encode(
    x='date:T',
    y=alt.Y(alt.repeat('layer'), type='quantitative').title("values"),
    color=alt.ColorDatum(alt.repeat('layer'))
).repeat(layer=["A", "B"])

[jonmmease]

@nicolaskruchten did something along these lines in plotly.express, so he may have some thoughts on the relationship between wide data and Vega-Lite's grammar.

[mattijn]

Thanks for sharing this link! Interesting to read how similar issues are approached elsewhere.

Melting of the long-form data into wide-form using pandas internally is a possibility (what plx does).

Or a macro approach (either python side or typescript-side), where a long-form data structure is detected and reworked into wide-form using a transform_fold or a repeated layer mechanism in vega-lite.

I think I favor a macro approach, but also no clear thoughts on this.

[joelostblom]

This is an interesting idea! I don't have any strong feelings either way here. I can see how it would be more convenient (and maybe even much more convenient in some cases) to have this functionality. I wonder if it could clash with any new Vega-Lite functionality down the road, although it is hard to imagine what that would look like...