| PRTools contents |
GENDATGAUSS
A = GENDATGAUSS(N,U,G,LABTYPE)
INPUT (in case of generation a 1-class dataset in K dimensions) N Number of objects to be generated (default 50).
U Desired mean (vector of length K).
G K x K covariance matrix. Default eye(K).
LABTYPE Label type (default 'crisp')
INPUT (in case of generation a C-class dataset in K dimensions) N Vector of length C with numbers of objects per class.
U C x K matrix with class means, or
Dataset with means, labels and priors of classes
(default: zeros(C,K))
G K x K x C covariance matrix of right size.
Default eye(K);
LABTYPE Label type (default 'crisp')
| Output | |
| A | Dataset containing multivariate Gaussian data |
Generation of N K-dimensional Gaussian distributed samples for C classes. The covariance matrices should be specified in G (size K*K*C) and the means, labels and prior probabilities can be defined by the dataset U with size (C*K). If U is not a dataset, it should be a C*K matrix and A will be a dataset with C classes.
If N is a vector, exactly N(I) objects are generated for class I, I = 1..C.
1. Generation of 100 points in 2D with mean [1 1] and default covariance
matrix:
GENDATGAUSS(100,[0 0])
2. Generation of 50 points for each of two 1-dimensional distributions with mean -1 and 1 and with variances 1 and 2:
GENDATGAUSS([50 50],[-1;1],CAT(3,1,2))
Note that the two 1-dimensional class means should be given as a column
vector [1;-1], as [1 -1] defines a single 2-dimensional mean. Note that
the 1-dimensional covariance matrices degenerate to scalar variances,
but have still to be combined into a collection of square matrices using
the CAT(3,....) function.
3. Generation of 300 points for 3 classes with means [0 0], [0 1] and [1 1] and covariance matrices [2 1; 1 4], EYE(2) and EYE(2):
GENDATGAUSS(300,[0 0; 0 1; 1 1]*3,CAT(3,[2 1; 1 4],EYE(2),EYE(2)))
| PRTools contents |