6
$\begingroup$

I have such points

pts = Uncompress[FromCharacterCode[
    Flatten[ImageData[Import["http://i.stack.imgur.com/zEP05.png"],"Byte"]]]];
ListPlot[pts]

If I use the normal method to calculate the center point, it will be in the red area.

ListPlot[pts, Epilog -> {Red, PointSize[.02], Point[Mean[pts]]}]

Now I want to calculate the specific "center" point. When the point in the denser area, it has greater weight. I think the "center" point I'm after should be at the bottom of that red point. But how to calculate it?

$\endgroup$
10
$\begingroup$

The median is known to be more robust against outliers than the mean:

ListPlot[pts, 
 Epilog -> {Red, PointSize[.02], Point[Mean[pts]], Green, 
   Point[SpatialMedian[pts]]}]

enter image description here

have also a look on the other "Location Statistics" in this guide on descriptive statistics.

$\endgroup$
3
  • 1
    $\begingroup$ Note that SpatialMedian requires you to load the MultivariateStatistics package in versions of Mathematica before 11.1. Also note that the SpatialMedian does not calculate the "median" as the term is usually used. (I agree it's probably the simplest thing to get at what the OP is asking for, though.) $\endgroup$ – Michael Seifert May 21 '18 at 13:41
  • $\begingroup$ Median also work but it takes the median componentwise. This is however not rotationindependent. (But if the different axes have different scaling, this might not be a problem at all...) $\endgroup$ – Henrik Schumacher May 21 '18 at 13:51
  • 1
    $\begingroup$ +1 It should be mentioned that there are many many different types of medians for multivariate data because the univariate median doesn't generalize to higher dimensions. I only see "Median" and "SpatialMedian" implemented in MMA but there are others such as Oja's Simplex Median, Beach and the Halfspace Median, Simplical Depth Median, and Convex Hull Stripping Median. Each one has its own advantages, disadvantages, robustness, computational complexity, etc. $\endgroup$ – Fixed Point May 21 '18 at 16:04
7
$\begingroup$

One possible way to operationalize the requirement "When the point in the denser area, it has greater weight" is to weight each point by the number of neighbors within a specified distance:

nf = Nearest[pts];
radius = 50;
weightedData = WeightedData[pts, Length[nf[#, {All, radius}]] & /@ pts];
center = Mean[weightedData];
ListPlot[pts, Epilog -> {Red, PointSize[.02], Point[Mean[pts]], Green, Point @ center}]

enter image description here

With radius = 100 we get

enter image description here

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.