Homework 3
MTH 496 { Machine Learning
Due date: March 31, 2018
(4 problems/1 page)
1 Handwritten Homework
Note All problems in this section requires the handwritten answers.
Problem 1.1 (20 pts). Given a data table as in page 19 of \Decision Trees" lecture note.
Write the training data for three di erent trees in the Random Forest.
Problem 1.2 (40pts). How to rank the feature importance using gini and accuracy in
Random Forest? Can Gradient Boosted Trees discussed in the lecture give you the feature
importance ranking and why/why not?
Problem 1.3 (20pts). How does gradient descent relate to the residue in Gradient Boosted
Trees?
2 Programming Homework
Note Write your codes in Azure notebook or similar kinds. Each question in a separate
notebook and submit all of them via a dropbox in D2L.
Problem 2.1 (100pts). Given data set StudySleep.csv. The rst column is the number
of studying hours per day of students, second column is the number of sleeping hours per
day of students, and the last column describes performance of students for the nal test.
Use the rst half of the data as the training set and the latter half as test set.
a) Use SVM with polynomial and radial basis kernels to learn the data and the predict the
test set. For each kernel, explicitly write the predictor and plot the decision boundary
along with training and test data points. Note that use di erent notations or colors to
distinguish test data, training data and di erent classes.
b) Compare Random Forest, Gradient Boosted Trees to the above SVM methods by con-
sidering accuracy, AUC, AP and F1. What is your conclusion?