ANDRORIG |
Original method to solve the Anderson-Bahadur's task.
Synopsis:
model = androrig(distrib)
model = androrig(distrib,options)
model = androrig(distrib,options,init_model)
Description:
It solves the original Anderson task [Anderson62]. The goal is to
find binary linear classifier which minimizes probability of
misclassification. The class conditional probability distributions
are Gaussians. The a prior probabilities is unknown.
model = androrig( distrib ) solves the original Anderson's task
for given two Gaussians distributions. The structure distrib
contains:
.Mean [dim x 2] Matrix containing mean vectors of the first and
second class distributions.
.Cov [dim x dim x 2]$ Matrix containing covariance matrices of the
first and second distribution.
model = androrig( distrib, options ) allows to specify the maximal
number of iterations options.tmax and the distance to the
optimal solution options.eps defining the stopping condition.
model = androrig( distrib, options, init_model ) allows to specify
the initial point init_model.gamma. The initial value of the
counter of iterations can be specified in options.t.
Input:
distrib [struct] Two Gaussians:
.Mean [ dim x 2] Mean veactors.
.Cov [ dim x dim x 2] Covariance matrices.
options [struct] Defines stopping condition:
.tmax [1x1] Maximal number of iteration.
.eps [1x1] Closeness to the optimal solution. If eps=0 the
algorithm converges to the optimal solution but it does not
have to stop (default 0.001).
init_model [struct] Init model:
.gamma [1x1] Auxciliary variable (default 1).
.t [1x1] (optional) Counter of iterations.
Output:
model [struct] Binary linear classifier:
.W [dim x 1] Normal vector the found hyperplane W'*x+b=0.
.b [1x1] Bias of the hyperplane.
.err [1x1] Probability of misclassification.
.t [1x1] Number of iterations.
.r1 [1x1] Mahalanobis distance of the first Gaussian to the
found hyperplane.
.r2 [1x1] Mahalanobis distance of the second Gaussian to the
found hyperplane. In the optimal solution r1 = r2.
.exitflag [1x1] 0 ... maximal number of iterations tmax exceeded.
1 ... condition delta < eps satisfied.
.delta [1x1] Indicates distance from the optimal solution.
.gamma [1x1] Auxciliary variable.
Example:
data = load('riply_trn');
distrib = mlcgmm(data);
model = androrig(distrib);
figure; pandr( model, distrib );
See also
GANDERS, EANDERS, GGRADANDR, LINCLASS.
About: Statistical Pattern Recognition Toolbox
(C) 1999-2003, Written by Vojtech Franc and Vaclav Hlavac
Czech Technical University Prague
Faculty of Electrical Engineering
Center for Machine Perception
Modifications:
20-may-2004, VF
24-Feb-2003, VF