Altair: Wrapping timeseries chart

[brylie]

A Redditor posted a seemingly simple chart design, asking how to produce the design:

Initial responses about how to produce the chart were in three main flavors:

• gantt
• stacked bar
• calendar heatmap

In practice, each of the above chart types present unique challenges for, and may be incapable of rendering, the desired output.

The data structure might be, in effect, an array of two-tuples where each tuple contains a start and end Datetime.

What are some ideas on how this chart could be created with Altair?

[curran]

This could be done with straight D3, but I'm not sure the Altair/Vega abstraction can support it.

[jakevdp]

It depends how your data is represented, but something like this can definitely be done in Altair. For example:

import altair as alt
import pandas as pd

df = pd.DataFrame({
    'start': pd.to_datetime(['2018-09-03T08:00', '2018-09-03T16:00',
                             '2018-09-04T10:00', '2018-09-04T15:00',
                             '2018-09-05T09:00', '2018-09-05T12:30']),
    'end': pd.to_datetime(['2018-09-03T14:00', '2018-09-03T18:00',
                           '2018-09-04T12:00', '2018-09-04T17:00',
                           '2018-09-05T11:00', '2018-09-05T17:30'])
})

alt.Chart(df.reset_index()).mark_bar().encode(
    y='day(start):O',
    x='hoursminutes(start)',
    x2='hoursminutes(end)',
    detail='index'
)

To me, this shows one of the great strengths of the declarative approach: rather than trying to figure out what chart type this maps to, and e.g. calling a gantt() method and digging into the options to figure out how to get it to work, you effectively just write down what you want to see using the grammar: "I want bars from start time to end time, with day on the y axis and detail for every row".

[brylie]

OK, that is elegant!

Thanks @jakevdp :-)

[curran]

That's fantastic!