|
I have a data set that its features are binary data(0,1). What is the best covariance function to use?? I played around with the RBF and it gives relative good results. Thanks |
|
There are many measures of association for nominal variables, but I usually reach for mutual information (typically scaled as a fraction of joint entropy). See Cover and Thomas for details: |
|
If you interpret the binary data as indicator values, then any data item can be seen as a set consisting of the features whose values are 1. Then you might use the Jaccard coefficient to quantify the similarity of two items. |
