Gaussian

Edge weight that weights each edge using a gaussian.

d^2    = sum_i (a_i - b_i)^2
weight = exp( -d^2 / (2 sigma^2) )

Closer neighbours weigh more. The weight lies in (0, 1] and is 1 at zero distance.

sigma sets the distance at which neighbours stop mattering, in the units of cols. The weight is about 0.61 at one sigma, 0.14 at two, and 0.01 at three. Choose it near the typical distance to a node’s neighbours.

Configuration

Selected as name: gaussian.

Option	Description	Type	Default
cols	Logical feature column groups the endpoint distance is taken over.	list[str]	required
sigma	Width of the kernel, in the units of cols. Must be positive.	float	required

components:
  edges:
    name: knn
    kwargs:
      k: 10
      neighbor_cols: [ dom_x, dom_y, dom_z ]
      weights:
        - name: gaussian
          kwargs:
            cols: [ dom_x, dom_y, dom_z ]
            sigma: 50.0

The attribute produced has shape [E, 1].