help_me_gui.txt

This Two-Category Classifier Using Discriminant Functions to separeate two classes. The Classifier

     -------------Two-Category Classifier--------------
                               
------------------------------------------------------------
PLEASE NOTE USERS:
THIS IS FOR EDUCATIONAL PURPOSES. SOMEONE CAN ALTER IF FOUND 
APPROPRIATE.
------------------------------------------------------------

This is Two-Category Classifier Using Discriminant Functions to 
separeate two classes. The Classifier is desined on classes which 
has two feature vectors and other case it has one feature vector.
------------------------------------------------------------

--Please click 'UPDATE' button after changing any parameters.
--Click 'RESET' to get defualt settings of parametes.

--Priori Probability has to be between 0.01 to 0.99, otherwise it 
will give warning message  'log of Zero'. Sum of probability of 
both classes will automatically set to 1.

--Select positive integer for No. of samples and it is suggested
that No. of samples are not selected too high, otherwise it may 
take very long time. Because points on decision boundary will 
look like merged together at certain amount of No. of Samples.

--Range for plotting also has to be reasonable in +/- , because 
for larger range density becomes 'zero' farther awy from mean 
and MATLAB will give warning message 'Log of Zero'.
------------------------------------------------------------ 

The classifier itself is simplified in three cases:

CASE 1: 
-------
In this case feature vectors are statistically independent and 
covariance matrix is diagonal. Since, Covariance matrix has to 
be same for both class, it doesn't make sense to allow permission 
to change all elements of Covariance matrix. So, GUI allow changes 
on only first element of Covariance matrix of CLASS 1 and rest 
will automatically updated. Other parameters can be changed to 
see the affect on classification.

For this case, samples fall in equal-size spherical clusters
(see contour plot). If prior prabability is same for both class
optimal decision boundary will bisecting two classes and perpendicular
to the line between two means.

If prior prabability is not same for both class optimal decision 
boundary will move away from more likely mean (class which has higher
priori probability).

Decision boundary will alway be straight line in this case.

CASE 2:
-------
In this case feature vectors are statistically dependent but, 
Covariance matrices are same for both classes. So, GUI allow 
to change only Covariance matrix of CLASS 1 and Covariance 
matrix of CLASS 2 will automatically changed to same.

For this case, samples fall in equal-size ellipsoidal clusters
(see contour plot). If prior prabability is same for both class
optimal decision boundary will be equally away from two means.

If prior prabability is not same for both class optimal decision 
boundary will move away from more likely mean (class which has higher
priori probability), which same as Case 1.

Decision boundary will alway be straight line in this case.

CASE 3:
-------
This case is general and it allows full permission to change 
all parameters within their Max and Min limit.

Optimal decision boundary is quadric. 

For Bivariate (2-D) Class:
--------------------
For 2-D case Mesh and Contour plot can be selected from pop-up 
box To see Decision boundary, it is better viewed in Contour plot.


For Univariate Class:
--------------------
Select Univariate from pop-up field. Change parameters to view the plot.
For this class, Case 1 and Case 2 is same.