Vega-lite: Include other useful transforms from Vega
Vega-Lite include a number of useful basic transforms. However, there are a number of transforms in Vega that can be migrated to the Vega-Lite level.
For each of this transform, we need to design syntax for the transform in Vega-Lite, which would not have type like in Vega, but always contains the transform name as one property of the transform (e.g., aggregate, filter, etc.)
Definitely useful
- [x] [fold](https://github.com/vega/vega-lite/issues/1538) https://github.com/vega/vega-lite/issues/1538
- [x] sample
[x] [flatten](https://vega.github.io/vega/docs/transforms/flatten/)
[x] impute
Data source
- Graticule (#3337)
- Sequence -- not sure if it should be a transform though -- perhaps more like a different kind of data source (#3864)
Probably useful?
- [x] stack -- this should be a good one to start, since we already have stack node, but just have no way to specify it outside of
encoding-- https://github.com/vega/vega-lite/issues/3311 [ ] pivot?
[ ] project ??
[ ] density & contour -- may need different syntax
[ ] countpattern
[ ]
https://vega.github.io/vega/docs/transforms/identifier/~~[ ] Collect (https://github.com/vega/vega-lite/issues/7017)
kanitw
All 11 comments
Sequence would be amazing for creating examples of mathematical functions. However, we have to decide whether it should go in data or transforms.
domoritz
on 23 Jan 2018
To me, sequence feels more natural as a special type of data source.
kanitw
on 23 Jan 2018
Yes, I think so too. It's just that Vega has it as a transform and I wanted to point out that it's not the most natural place and we can change that in Vega-Lite. Yet another reason to not just support all Vega transforms.
domoritz
on 23 Jan 2018
Sequence is covered in https://github.com/vega/vega-lite/issues/3864
domoritz
on 12 Dec 2018
Graticule is in https://github.com/vega/vega-lite/issues/3337
domoritz
on 12 Dec 2018
I don't think we need pivot in Vega-Lite, do we?
domoritz
on 8 Apr 2019
It's actually very useful. For example, with pivot, we can create "Multi Series Line Chart with Tooltip" with the stocks data (as the stocks data has a long format, but we need a pivoted format for "Multi Series Line Chart with Tooltip".
kanitw
on 8 Apr 2019
@kanitw I'm interested in submitting a PR to add pivot, but from a quick look at the code I'm not sure how to implement producedFields since they're generated from the data values. Should it just return an empty set, or would that cause issues?
pjsier
on 13 May 2019
@pjsier Thanks for looking into this. Figuring out the correct producedFields is super tricky here. I think we can only return null or {} in this case but we need to update the checks that use this field so that when we check for intersections and we have a pivot transform, we need to always say that there is an intersection. This will make it difficult to optimize the dataflow when there is a pivot transform.
domoritz
on 15 May 2019
@domoritz sounds good, thanks! I鈥檒l try to put together a PR with that setup
pjsier
on 15 May 2019
null or {} would seems like "nothing". I wonder if we should use a specific string literal "unknown" to make it explicit that we don't know at compile time.
kanitw
on 15 May 2019
Related issues
rmales
路
3Comments
kanitw
路
3Comments
kanitw
路
3Comments
fuglede
路
4Comments
domoritz
路
3Comments
Most helpful comment
It's actually very useful. For example, with
pivot, we can create "Multi Series Line Chart with Tooltip" with the stocks data (as the stocks data has a long format, but we need a pivoted format for "Multi Series Line Chart with Tooltip".