DEMO_MMGAUSS |
Demo on minimax estimation for Gaussian.
Synopsis:
demo_mmgauss
Description:
demo_mmgauss demonstrates the minimax estimation algorithm
[SH10] for bivariate Gaussian distribution. The training data
is supposed to contain samples which well describing the
probability distribution function (pdf), i.e., which have
high value of pdf. The samples do not have to be i.i.d. in
contrast to the ML estimation.
The estimated model is visualized as an ellipsoid:
shape is influenced by the covariance matrix and the center
corresponds to the mean vector.
The lower (red) and upper (blue) bound on the optimal value
of the optimized minimax criterion is displayed at the bottom
part of the window.
Control:
Epsilon - Stopping condition. The algorithm stops if the
difference between lower and the upper bound
is less then the epsilon.
Iterations - Number of iterations after which the model
is re-displayed.
FIG2EPS - Exports figure to the PostScript file.
Load data - Loads input data sample from file.
Create data - Invokes program for creating data sample.
Reset - Resets the demo.
Play - Runs the algorithm.
Stop - Stops the running algorithm.
Step - Performs one iteration of the algorithm.
Info - Invokes the info box.
Close - Closes the program.
See also MMGAUSS.
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:
2-may-2004, VF
19-sep-2003, VF
3-mar-2003, VF
11-june-2001, V.Franc, comments added.
24. 6.00 V. Hlavac, comments polished.