This preview shows page 1. Sign up to view the full content.
Unformatted text preview: Assignment 3 (Computer Exercise)
Faculty of Computers and Information Faculty Department of Computer Science Pattern Recognition A First Simulation Example on Designing and Assessing Classiﬁers
• Generate two small (10 observations per class) training data sets both from binormal distribution with diﬀerent mean vectors and identity covariance matrix (case I in text). • Design a Linear Discriminant Classiﬁer (LDA) using the training sets, which is another name for the estimated Bayes Classiﬁer with equal covariance matrices. • Generate two large (1000 observations per class) testing data sets to assess your classiﬁer (in real life application we do not have this luxury). • Assume equal priors and costs and calculate the two types of errors of your trained classiﬁer. Compare its performance to the Bayes classiﬁer, i.e., the actual Bayes classiﬁer that results from using the mean vector and covariance matrix directly without estimation. • Repeat the above 10 more times with diﬀerent training-set sizes, e.g., 20, 40, 80, 100, 200, 300, 400, 500, 700, 1000. Plot the error vs. the training set size. What do you observe? ...
View Full Document
- Spring '10
- Computer Science