i have a matrix data. Perhaps some data in one cluster and another in some cluster.
data scale is between [0-1000](just example). and i want to normalize into [0-1] and good in mean and variance. it means, i want to have a Good visibility in mean and variance.
for more description,mean and variance Be chosen in perfect place that Consider all data, and not ignore noisy data.
Foe example one way is zi=xi−min(x)max(x)−min(x), but this way if Most of the data in one cluster and few data in another cluster, mean and variance be chosen between large cluster and not attend to noisy data.
Please introduce me a paper(if newest is better) or a way to solve this problem.
Thank u.
$\begingroup$
$\endgroup$
1
-
$\begingroup$ No clue what u wrote. $\endgroup$– user88Commented Dec 2, 2013 at 21:48
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
If you know or you think that there are two clusters in the data, then fit a mixture of two normal distributions, one for each cluster. then for every point, choose the set of points belonging to that cluster and then perform the transformation you think is right. when reporting or using each point, report the distribution you have assigned the point to as well as the transformed value. that way you get to preserve the natural structure of data, yet change it to a form you want it to.
