quiz4_2009

# Gabriel van helsing one of the two founders once

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: e remembers Professor Winston's Vampire Identification Tree lecture. He assigns you the task of creating a superior classifier for vampires by using boosting on the following data. Vampires: ID Name Vampire Evil Emo Transforms Sparkly # Romantic Interests ID 1 Dracula Y Y N Y N 5 1 2 Angel Y N Y Y N 5 2 3 Edward Cullen Y N Y N Y 1 3 4 Saya Otonashi Y N Y N N 3 4 5 Lestat de Lioncourt Y N Y N N 5 5 6 Bianca St. Claire Y Y N Y N 5 6 7 Mircalla Karnstein Y Y N Y N 5 7 8 Sailor Moon N N N Y Y 1 8 9 Squall Leonhart N N Y N N 1 9 N N N N N 5 10 10 Circe 7 Part A (10 points) To make it easier for yourself, you first determine for each possible classifier (including the classifiers True and False which mean that everyone is a Vampire or not a Vampire, respectively), which of the data points would be misclassified. For example, for the first classifier, the one that says a person is a vampire if he is Evil (and not a vampire if not evil) the classifier is wrong on sample 2, 3, 4, and 5. We have given each classifier a name to make it easier for you to refer to them. Classifier A B C D E F G H I J K L M N Test Evil Emo Transforms Sparkly # of Romantic interests # of Romantic interests TRUE Evil Emo Transforms Sparkly # of Romantic interests # of Romantic interests FALSE Value Y Y Y Y >2 >4 Misclassified 2, 3, 4, 5 N N N N <2 <4 Part B (5 points) Again to help yourself later, only 6 of these classifiers would ever be used because the other 8 make all the same errors as one of the other classifiers and then make additional errors. Circle the 6 above that you would consider using. If you are sure you have the right 6, you don't have to waste your time with any of the others in Part C. 8 Part C (15 points) Now you are ready to perform Boosting on the collected Vampire dataset. Fill in the following blanks for the weights, classifiers, errors, and alphas of the first three rounds of Boosting. If there is ever a tie, break it by choosing the classifier that is higher on th...
View Full Document

## This document was uploaded on 03/17/2014 for the course EECS 6.034 at MIT.

Ask a homework question - tutors are online