Name: ———————-
Practice Quiz - 2
CS 591Q/791V - Pattern Recognition
Posted on: April 9, 2009
Please read this first:
This is an in-class, closed-book/notes quiz consisting of 3 questions. ©
You will have to turn in your solutions by 12:15pm.
You are not permitted to engage in any kind of discussion during the quiz.
If a question seems ambiguous, state your assumption and proceed to solve it.
An act of academic dishonesty will fetch you a 0 in the quiz. §
1. Consider a dataset in which every pattern is represented by a set of 15 features. The goal
is to identify a subset consisting of 10 features or less that gives the best performance on
this dataset. How many subsets would each of the following feature selection algorithms
consider before identifying a solution?
(a) Exhaustive search;
(b) SFS;
(c) SBS;
2. Consider the following 2-dimensional labeled training patterns representing two classes C1
and C2. The goal is to derive a linear decision boundary using the perceptron learning
algorithm.
C1(t= +1) C2(t=−1)
x1= (3,4) x4= (1,1)
x2= (3,5) x5= (1,2)
x3= (4,4) x6= (2,1)
(a) Write down the transformed 3-dimensional vectors y1,y2,y3,y4,y5and y6that are
necessary for the perceptron algorithm.
(b) What is the final linear decision boundary obtained using the Batch Perceptron al-
gorithm assuming that the weight vector wis initialized to (−12,8,10)T,η= 1, and
θ= 0? Note that θ= 0 implies that the algorithm terminates once all the patterns are
correctly classified.
3. What is the rationale behind the expression for the Fisher’s criterion?
1