Simpler way to achieve labelled bar graph

[mcp292]

Is this still the simplest way to achieve a labelled bar graph?

https://altair-viz.github.io/gallery/bar_chart_with_labels.html

Is there a way to do this without layering? Possibly an encoding I am not aware of?

[joelostblom]

You will need to layer two marks to have both bars and text in the same chart. Having that said, the syntax of the example you linked could be more succinct:

bars = alt.Chart(source).mark_bar().encode(
    x='wheat',
    y="year:O",
    text='wheat'
)

bars + bars.mark_text(align='left', dx=1)

This looks pretty readable to me, but I'm curious to hear what you had in mind for the syntax and if you have encountered issues with the layering approach.

[mcp292]

I appreciate it but you guys go ahead and take the credit on the PR. I would love to contribute but I'm pretty swamped with work at the moment, which is how all these questions are arising.

for myself I think about the columns that I want to plot first, and then which plot I would like to do. E.g. in the example above, my first thought would be something like "I want to see how the wheat yield changes over time" and then I would think of which mark type to use: bar, points etc.

This is a great point actually, I guess I do think of the data encoding first and then the visualization follows naturally after I have established that. The one time this was different was yesterday when I plotted a horizontal bar graph, and the quantity I wanted to observe was on the x channel.

You know, I think I know why I said that, I usually map everything out on paper first, so by the time I go to code, I know I want a bar chart, and then figure what goes where. So the difference between us might be where that thinking happens. Yours might happen in the editor.

[joelostblom]

OK PR here #2991

I usually map everything out on paper first

I think this is a great practice and can be very helpful! I do it sometimes myself, mainly when planning for charts and datasets that I am less familiar with, or to get an overview early on in a project. Or when I get stuck on some detail of a visualization and need to take a step back. Definitely helps a lot, and I do start writing out the columns/encodings on the xy-axis on the paper before adding the marks, so my process is similar here to (but as you said, when transitioning to coding from paper it might differ)

[mattijn]

I can understand you from a theoretical point of view, but my practical experience differs a bit. I think I start from the mark. Like, I need a map that shows x, or I need a roseplot that shows y.

And then I often start with a verbose approach, especially for a layered charts, so that I can visualize each sublayer individually.
Only in the last part I start merging and beautify things together, like what is suggested here.

[joelostblom]

Thanks for sharing that @mattijn. I got more curious and tried to look back in the history of the repo a bit. It seems like the syntax actually used to be Chart().encode().mark_*() in the early days, so there is probably some good reason that it changed. I couldn't find the specific PR that changed it, but this one include changing some of the examples https://github.com/altair-viz/altair/pull/107/files and this one clearly shows the old syntax #92. Maybe I will look into this more in the future, but I'll leave it here for now.

[mcp292]

It seems like the syntax actually used to be Chart().encode().mark_*() in the early days, so there is probably some good reason that it changed.

Fascinating! Nice digging!!