1. Supervised Learning: A classification example
This assignment is about a classification problem. Given are training data in file TrainData.mat with three features (first three collumns) and class labels in the fourth collumn. Your task is to make predisctions for the feature values given in file TestData.mat.
You should use the LIBSVM implementation of Support Vector Machines (SVMs) for this task. There are two functions that you need to call after implelmentation on is svmtrain(y,x,'parameters'), the other is svmpredict(y,x,model), where model is the pointer to the trained object returned by the training function. There are some choices you can make for parameters, and you should specify which choices you made in your submission. Also, note tht the svmpredict function ask for labels for the test data that you don't have. So just supply random labels for the test data and ignore the evaluation feedback.
You can also use, in addition, a method of your choice. You get bonus points if you can beet the SVM solutions.
2. Inference in Bayesian network: A diagnostics example
Given is the following diagnostic problem: There is a probability of P(d)=0.01 of having a particular disease. A test for this disease comes back positive in 90 perent of cases when the patient has the disease, and the test is negative in 80 percent of the cases when the person does not has the disease. Suppose a person received two tests and both tests came back positive. What is the probability of having the disease? Also, what is the probability of having the disease when one test comes back positive and one test comes back negative.
a: Show how you can calculate the answers analytically.
b: Use the Bayes net toolbox by Kevin Murphy to calculate your answers.
Submit your answers (.mat file for label predictions in question 1, written anwsers in pdf format, and the matlab code for 2b) to prof6905@cs.dal.ca with subject line A5 before 6pm on Thursday , Nov. 3.