A=GAUSS([M],[S],H) returns a gauss object initialized with mean M,
std S and hyperparameters H. Can also be called with GAUSS(H)
Training will try to fit a single Gaussian to the data.
Testing will return the probability estimate, or if passed
an empty dataset will generate new data according to the
density learnt.
Hyperparameters:
l=50 -- number of data points to generate if asked to generate
assume=[] -- can make assumptions: {'diag_cov','equal_cov'}
where the matrix is diagonal ('diag_cov'),
or diagonal with all elements equal ('equal_cov')
Model:
mean=0 -- centre of gauss distbn
cov=1 -- cov matrix, if single value assumes diagonal
of matrix all has same values, if vector assumes it
is the diagonal of the matrix, with 0s everywhere else
Methods:
train, test, generate
Examples:
gen(gauss) generate data
gen(gauss('l=5;mean=[1 1];cov=0.01;')) generate data
d=test(gauss([1 5],[1 -0.4 ; -0.4 1],'l=300'));
[d2 a]=train(gauss('l=200'),d);
d2=test(a); train a gauss on some (gauss) data and then
try to generate similar data
hold off; plot(d.X(:,1),d.X(:,2),'o')
hold on; plot(d2.X(:,1),d2.X(:,2),'rx')
| |
Reference : chapter 2 (Richard O. Duda and Peter E. Hart) Bayesian Decision Theory |
| |
Author : Richard O. Duda , Peter E. Hart |