These simulation data are described on the revised page 384-5 of the book. See the errata file for a revised version if your book is not the fourth or later printing. There are 50 x four datasets - training and test data for the four dimensional problem, and training and test data for the 10 dimensional problem. These come in four files. For example, orange4.train is a 5000 x 5 matrix consisting of fifty datasets of size 100 x 5. The first column is the -1/1 output, the remaining four are the inputs. orange4.test is 50000 x 5, since for each training set of size 100, there is a test set of size 1000. Hence it has 50 test sets.