0
$\begingroup$

I have a large set of multi dimensional data.The data points are highly skewed and not smoothly distributed.I want to divide the data set to some finite number of bins.I have approached this problem with some clustering algorithms(like KMeans and Mean shift).But as there are many clustering algorithms I am not sure which one would be most efficient. Again I can apply clustering to multi dimensional data only.But how should I approach when the data set consists of single dimensional data(like a large 1-D array) only?

$\endgroup$
  • $\begingroup$ It would be helpful to know what your goal is. Does your data actually have clustered structure and you want to identify it? Or do you want to do something more like vector quantization (i.e. just bin the data in a multidimensional way)? What was unsatisfying about the approaches you tried, and what does 'efficiency' mean in your application? $\endgroup$ – user20160 May 23 '16 at 11:01
0
$\begingroup$

You should try other distance measures.

For example Canberra and Clark distances often work much better with skewed distributions. So they are worth a try.

Try loading your data into ELKI, and run OPTICSXi with Canberra distance. That has worked very well for me as a first attempt to cluster the data when exploring a new data set.

$\endgroup$

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.