Request: Add an interactive profiling view (like quak) to explore tabular data

[rluthi]

To explore tabular data, it would be nice to have in viewing panel on the right, a minimalist profiling summary of the selected points vs. the whole population (all the points).

I quite like how @manzt does it in the quak interactive data frame viewer (selected points in blue and whole population in grey):

[koaning]

What might such a view offer that quak currently does not?

[rluthi]

My use-case is exploring a 2-D embedding of my data to identify interesting clusters and describe with a subset of "profiling variables". quak is great when you already have your segments defined. But, in the exploration phase, it requires re-executing the cell with the widget each time a new selection is made.

With quak, there is also a lot of horizontal scrolling to do on larger datasets. I like the idea of having the profiling view in the right pane of the bulk widget with all of my "profiling variables" visible on screen at the same time.

[koaning]

Ah, so in your use-case quak would more be a data viewer, not a selector for other views so to say? In bulk land the UMAP chart is a selector element and the text on the side is a viewer.

[rluthi]

Exactly! I would like to be able to select points in a 2D UMAP scatter plot selector (like in your text dataset demo) and have it synced with a quak-like viewer panel with the profiling panel

[rluthi]

Very roughly:

[koaning]

Would you still not want a table preview at the bottom there? This feels relatively easy to implement. Might have a go at it later today.

[rluthi]

Table preview would be nice if there is enough space. Could be an option...

[koaning]

A lot of it kind of depends on the number of columns. Once you start having a lot of them it because hard to see what the selection is really about.

Just a rough feeling, in your day to day, how many columns do you have? Also, what types? A lot of categoricals?

[rluthi]

8 to 15 columns, mostly categories /boolean (~70% of columns) the rest are numerical