It might be useful to try to see how to investigate the relationship of this project to b2, by @yifanwu.
It lets you interactively create cross filtering charts from a notebook that pulls from your kernel: https://github.com/yifanwu/b2#b2
It's at a slightly higher level than this project, and more constrained in the types of charts it produces. However, it has the advantage of being really easy to use! Not requiring any knowledge of Ibis or Pandas...
I think it would be a great complement to this project, to help with the data exploration step.
We chatted a bit about how they could possibly work together. One way could be this:
Right now b2 produces many vega lite charts and manually synchronizes them. Instead we could have it synthesize one large vega lite chart with the crosfiltering and use ibis-vega-transform to do the execution of the queries. Yifan said this might be a bit complicated, because currently she is doing the the calculation from the crossfiltering in relational algebra to create SQL directly. And in this case, she would instead have to use it to create Vega Lite, or Vega, charts.
We could provide a way to get access to the ibis expression that corresponds to the currently created crossfilter, allowing you to move between interaction and back to the notebook. Possibly also some way to serialize it in the notebook? So that it get's preserved when you reload?
We could also create some sort of side UI in jupyterlab, or use an existing extension, to show you all datasets/tables that you have loaded, so you can operate more freely, not stuck to a cell.
It might be useful to try to see how to investigate the relationship of this project to b2, by @yifanwu.
It lets you interactively create cross filtering charts from a notebook that pulls from your kernel: https://github.com/yifanwu/b2#b2
It's at a slightly higher level than this project, and more constrained in the types of charts it produces. However, it has the advantage of being really easy to use! Not requiring any knowledge of Ibis or Pandas...
I think it would be a great complement to this project, to help with the data exploration step.
We chatted a bit about how they could possibly work together. One way could be this: