Statistics 503 Homework 4 Due: in class on Wednesday, Mar 30 2005 Background: Neural networks and support vector machines are computationally intensive methods for fitting classifiers to data. This homework explores how these classifiers work with the music data. It is to be done individually, with help from Dr Cook as necessary. Purpose: To learn how to fit neural networks and support vector machines to data, and how to assess the fit. Exercises: 1. Use your previous subsetting of the music data into 2/3 training, 1/3 test data. List your row numbers for training and test sets. 2. Compute a new variable LFreq, which is the median of LFreq1-LFreq15. 3. Fit a neural network classifier to your training set using variables LVar, LAve, LMax, LFEner, LFreq. (a) To begin use size=0, so that you are fitting a regression model. Report the model, the training error and the test error. (b) Next fit several models, size=2, 5, 10. Run each size 10 times, and tabulate the results. Report the final value, training error and test error for each of the fits, in a table like the following. Trial Ave SD Size 1 2 3 4 5 6 7 8 9 10 2 final value train error test error 5 final value train error test error 10 final value train error test error 4. Build an SVM classifier for the training data. Write down the classification rule prescribed by SVM. Report the training and test error. Report which tracks are unbounded and bounded support vectors.