This is a collection of data sets for the halfspace depth. We are making a thorough computing of the depth value of very point in the data sets. We will keep updating the results.

You may need to see the explanation of the format of the files.

Randomly generated data sets grabbed from David Bremner's website

A set of 120 points in dimension 10

The data set and the results

Data sets from the UCI Machine Learning Repository

Pima Indians Diabetes Database: the last column of the original data set is removed since they are some binary values. Details about the original data set is here.

The data set.

Some results of this data set.

Upper bounds with random hyperplane method. Point p is not included in the set of d points. For every point we tried 10000 hyperplanes (we did the same for all the following tests), and recorded the best result. the running time on whole data set is given at the end of this file.

Upper bounds with random hyperplane method. For any point p, the set of d points consists of p and d - 1 other randomly selected points.

Upper bounds with random hyperplane method. This time, no points are selceted, and the hyperplanes are randomly generated.

Boston Housing Data: the fourth attribute of the original data set is removed since they are some binary values. Details about the original data set is here.

The data set.

Some results of this data set.

Upper bounds with random hyperplane method. Point p is not included in the set of d points.

Upper bounds with random hyperplane method. For any point p, the set of d points consists of p and d - 1 other randomly selected points.

Upper bounds with random hyperplane method. This time, no points are selceted, and the hyperplanes are randomly generated.

BUPA liver disorders. Details about the original data set is here.

The data set.

Some results of this data set.

Upper bounds with random hyperplane method. Point p is not included in the set of d points.

Upper bounds with random hyperplane method. For any point p, the set of d points consists of p and d - 1 other randomly selected points.

Upper bounds with random hyperplane method. This time, no points are selceted, and the hyperplanes are randomly generated.

Johns Hopkins University Ionosphere database: the first two and the last attributes are removed from the original data set. Details about the original data set is here.

The data set.

Some results of this data set.

Upper bounds with random hyperplane method. Point p is not included in the set of d points.

Upper bounds with random hyperplane method. For any point p, the set of d points consists of p and d - 1 other randomly selected points.

Upper bounds with random hyperplane method. This time, no points are selceted, and the hyperplanes are randomly generated.