point-2d-smoothing

Estimate 2D smoothing for 2d coordinates.

Motivation

This library is to estimate values (smoothing or density) for 2D discret points. It can be geographical coordinates or x,y pairs. However, currently it only supports Euclidean distance, which means that you will get distorted images for latitude and longitude. It would be useful if you have bunch of points and want to get some estimations, such as KDE or a heatmap.

Although this library can achieve reasonable performance through KD tree, Gaussian Blur might be a better alternative if you just want some blurring effect.

Example

import Smoothing from 'point-2d-smoothing';
// only have two peaks
const data = [{
  x: 0,
  y: 0,
  value: 1,
}, {
  x: 1,
  y: 1,
  value: 1,
}];
const mySmoothing = new Smoothing({
  data,
  algorithm: 'IDW',
  options: {
    radius: 0.6, // required
  },
});
const point = [0.5, 0.5];
// Inverse Distance Weighting
const idwResult = mySmoothing.estimate(point);

Currently supported kernels in KDE:

• gaussian
• tophat (aka uniform or "rectangular window")
• epanechnikov (aka parabolic)
• linear (aka triangular)
• cosine

One parameter is needed: bandwidth (default value is 1).

As gaussian is not bandwidth-bounded, you can provide a option "searchRatio" to expand the search radius of neighbors. By default, this value is 3, which means that the algorithm will use all points within the circle of radius 3*bandwidth will be included in calculation.

Things on the road

• Inverse Distance Weight (IDW) algorithm
• Test for IDW
• KDE with Epanechnikov kernel
• KDE with Gaussian kernel
• Test for KDE
• More distance function support, such as great circle distance
• Documents