2
$\begingroup$

For vectors we can use the covariance matrix which contains the variances per variable and the covariances.

Now I want to compare the variances of multiple point clouds. It is hard to compare the covariance matrices directly, so I would like to condense the information to a single variance scalar. (Actually its deviation, i.e. variance's square root.)

What is the typical/classical value to use?

Use the square root of the trace of the covariance matrix (the sum of its diagonal)? Or maybe not use the covariance matrix at all, e.g. computing the average Euclidean distances of each point to the mean of the point-cloud?

Improve this question
$\endgroup$
7
  • 1
    $\begingroup$ Could you explain why you are comparing these variances? What information are you hoping it will convey? $\endgroup$
    – whuber ♦
    Commented Nov 5, 2015 at 14:17
  • $\begingroup$ @whuber I want to be able to say that point cloud A varies less than B (with respect to the chosen measure of variance) $\endgroup$
    – ben
    Commented Nov 5, 2015 at 15:32
  • 1
    $\begingroup$ But that's the entire point: what aspect of this rich, multivariate complex of points are you trying to describe? One could propose all kinds of single-value measures of variation, but you haven't given us any information about how to select one that is suitable for your objectives. $\endgroup$
    – whuber ♦
    Commented Nov 5, 2015 at 15:34
  • 1
    $\begingroup$ That interpretation would make your question overly broad and unfocused, and perhaps a subjective matter, because there are many ways to do that. Indeed, any question on this site is all about what you need to use. It's not for us to tell you why you're interested in something! $\endgroup$
    – whuber ♦
    Commented Nov 5, 2015 at 16:36
  • 1
    $\begingroup$ There is no one "typical/classical way." Please also consult our help center, which implores you to ask "practical, answerable questions based on actual problems that you face." Requesting something abstractly "classical" would seem to have nothing to do with an actual problem. $\endgroup$
    – whuber ♦
    Commented Nov 5, 2015 at 20:24

0

Reset to default

Your Answer

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

Browse other questions tagged or ask your own question.