Hello All!

[cams-data]

Hello fellow Superset enthusiasts!

I'm Cam, I've worked as a data scientist, data engineer and am currently head of data at the company I work for. I have many years experience with multiple BI tools and have begun using Superset for the first time in some personal projects.

I have taken note that Superset seems to be rapidly adopting an extensible platform architecture. That's very exciting. I am keen to contribute some extensions and would also like to contribute to the main code base. I am still familiarising myself with the main codebase, but can see a few small issues to cut my teeth on.

I see there has been some recent discussion on semantic models, I have my own thoughts, but am still developing them.

I have deployed Superset 6.0.0 using the 0.15.2 helm chart and a custom Dockerfile, all seems to be working well so far. My main connections are to ClickHouse and Postgres.

I am wanting to do some visualizations with deck.gl in Superset. I don't think my exact use-case is covered, so a custom deck.gl visual may be my first attempt at an extension. I have a Tegola server that serves map tiles, the map tiles are the outlines of buildings. I intend to use deck.gl to extrude the buildings in the vertical axis as a form of bar chart, across the city. The extrusion height will come from the datasets I have in Superset -> each record will have an id that maps to a building in the tileset. This interplay between a custom tileset and a dataset is what I do not think is supported in any of the base deck.gl visuals?

Anyway, that's my rambling over. Hope you are all well!

[herve-lg]

Why not explore this solution indeed... Nevertheless, it seems to me that the same result can be achieved simply with Superset (without Tegola)?

If you use Deck.gl Polygon with

• "Polygon Column" = column of the geometry of the footprint of your buildings in GeoJSON (Feature ::Text);
• "Elevation" > "Based on a metric" = column of the height of your buildings.

That said, I like the idea of being able to integrate map services into the Deck.gl charts as in the Cadiodiagram (WMS, WFS, and XYZ). Is that your idea?

[cams-data]

I was aware of that method, I am just afraid it won't be performant enough at many different zoom levels. Using a proper tile server let's me stream in only the geometry being rendered and at different LODs. I'll try the direct geometry method out, but I suspect it won't give me the results I need.

I haven't fully fleshed my idea out yet, but it is essentially to build a visual that can use metrics from the dataset to affect the rendering of a deck.gl chart whose map/geometry data comes from a tile server. That has wide reaching applications and will serve as a good base. The existing charts can do much of this, but where they fall short is enforcing that the geometry must live with the dataset, a la polygons and multipolygons in Postgis.

[herve-lg]

@cams-data - You’re right about that, "it won't be performant enough at many different zoom levels". Indeed, the column GeoJSON::Text does not allow to have a geometric index like GIST in PostGIS. Therefore, all data in the column must be loaded before it is displayed. In my case, to display 60,000 buildings, it takes 30 seconds. Afterwards, movement through the window is smooth.

Given the importance of PostGIS in the field of cartography, I personally dream of being able to directly connect my PostGIS geometry column (indexed with GIST) into Superset instead of having to transform it into GeoJSON::Text. But I’m also interested in your solution, but I use Geoserver.

[cams-data]

@herve-lg Importantly, besides spacial indexing for more efficient access, a tile server allows you to pre-compute polygon simplifications and serve them at different tile sizes, giving you LODs. When paired with a GIST index you get fast data access and a chart that looks great at any zoom level. I intend on using >1m buildings in a single chart, so these techniques will all be needed.

Before I begin though I need to keep familiarising myself with both deck.gl and Superset.

[new5r]

nice