How to define polygons in Grakn?

Question

I want to model physical objects in Grakn. It's quite straightforward to model physical objects as a point or a rectangle:

define

length sub attribute,
    value double;

position sub attribute,
    value double;

lon sub position;
lat sub position;
elevation sub position;
width sub length;
depth sub length;
height sub length;


physical_object sub entity,
    has lon,
    has lat,
    has elevation,
    has width,
    has depth,
    has height;

For a point, simple leave width, depth and height empty.

However, I'm struggling with how to model a physical object that is a polygon, i.e. a list of connected lines. How can I model this elegantly in Grakn?

Answer 1

Given that not every physical_object in your world has width, depth and height, maybe we can remove these attributes from the high-level definition?

physical_object sub entity,
    has lon,
    has lat,
    has elevation;

And rectangle can be

rect sub physical_object,
    has width,
    has depth,
    has height;

Then you can define point like that

point sub physical_object,
    plays edge-point_point;

and an edge would be 2 points

edge-point sub relation,
    relates edge-point_point,
    relates edge-point_edge;

edge sub entity,
    has length,
    plays edge-point_edge;

after that polygon can be represented as

polygon sub physical_object,
    has edge; // here it can be multiple edges, we don't have cardinality constraints yet

But in that way polygon can be a set of not connected edges, you can define polygon differently depending on how you want to use it

Answer 2

Most spatial graph databases (and CAD programs) store their data as arrays of primitives to facilitate describing any possible shape. These databases also allow one to natively perform intersection/unions on such arrays. You might want to follow that pattern if only to be able to import/export from/into these systems. Here is one example: https://docs.oracle.com/database/121/SPATL/simple-example-inserting-indexing-and-querying-spatial-data.htm#SPATL486

So perhaps the question to the GRAKN community should be: how to best model arrays of primitives?

P.S. adding temporality into the mix would make it even more interesting/challenging!