0
$\begingroup$

I have a set of unlabeled data $x \in \mathbb R^n$, and I want to visualize how scattered those points are using python. In dimension 2, we can just use plt.scatter which is quite straightforward. I wonder whether there is any approach to deal with this in higher dimensions.

$\endgroup$
5
  • 1
    $\begingroup$ What properties of the result interest you? Clearly a monitor can't display data in more than 2 dimensions, so we will, in some way, be forced to make a compromise. Do you wish for the result to be orthogonal? Or preserve pairwise distances? Or something else? $\endgroup$
    – Sycorax
    Commented Jul 24, 2020 at 18:39
  • $\begingroup$ I am hoping to visualize whether all the points are sufficiently close to a given fixed point with respect to the Euclidean distances. A good indicator apparently is to directly calculate the average Euclidean distance. However, I do not quite have a good feeling about what the scale should be. So I wonder whether it is possible to have some sort of visualization. $\endgroup$
    – Adam
    Commented Jul 24, 2020 at 18:46
  • 1
    $\begingroup$ What, then, do you mean by "sufficiently" close? That might help answer the question. $\endgroup$
    – whuber
    Commented Jul 24, 2020 at 19:04
  • $\begingroup$ I am not sure about what threshold to pick, and that is exactly why I need to visualize a bit. $\endgroup$
    – Adam
    Commented Jul 24, 2020 at 19:06
  • 3
    $\begingroup$ Hm. If your main interest is the Euclidean distance from a given point, then a simple histogram of these distances may be a good start. Alternatively, you may want to look into multidimensional scaling, which attempts to project your points into two dimensions while preserving distances as much as possible. $\endgroup$ Commented Jul 24, 2020 at 21:08

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.