Introduction to Pattern Recognition: A Matlab Approach

1.7 Parzen Windows 19

[y_est]=mixture_Bayes(m_hat,S,w_hat,P_hat,Z);

[classification_error]=compute_error([ones(1,500) 2*ones(1,500)],y_est);

The computed classification error is equal to 4.20%.

Remark

• In a classificati on t ask, when t he number of summands in the mixture model is not known, the task

is run a number of times with different values of J; the value that results in the lowest classification

error over the test set is adopt ed.

Exercise 1.6.1

Repeat Example 1.6.2 using X1, X2,Z1, Z2 with the following initial parameter values:

• For p

1

(x): Three normal distributions with initial mean estimates m

11,ini

= [5, 5]

T

, m

12,ini

=[5.5, 5.5]

T

,

m

13,ini