1 True/False [20 points]
1.1 [2 points] KNN is an unsupervised learning algorithm.
SOLUTION: False - the input is (x; y) pairs = supervised learning
1.2 [2 points] A classifier is less likely to overfit when trained o
...
1 True/False [20 points]
1.1 [2 points] KNN is an unsupervised learning algorithm.
SOLUTION: False - the input is (x; y) pairs = supervised learning
1.2 [2 points] A classifier is less likely to overfit when trained on less training data.
SOLUTION: False - overfitting is usually the result of variance which decreases with more data
1.3 [2 points] Given a housing dataset, we want to predict the housing prices given the attributes of the
house. Logistic regression could be an appropriate model for the task.
SOLUTION: False { logistic regression is a classifier and we want a continuous real value
1.4 [2 points] It usually takes a longer time to train a KNN than a SVM.
SOLUTION: False { knn essentially requires no training whereas SVM = convex opt.
1.5 [2 points] Linear discriminant analysis models the feature vectors of each class as a multivariate Gaussian,
sharing the covariance matrix between classes.
SOLUTION: True { its the definition of LDA
1.6 [2 points] Moving in the direction of the gradient always decreases the function’s value.
SOLUTION: False { the gradient points in the direction of stepest ascent
1.7 [2 points] Linear discriminant analysis is an example of a discriminative classifier.
SOLUTION: False { it models P(XjY ) as a gaussian and hence is generative
1.8 [2 points] The algorithm we use to fit logistic regression explicitly minimizes the number of errors on the
training set.
SOLUTION: False { we are minimizing an upper bound on the training error
1.9 [2 points] The theory for agnostic learning only applies if the hypothesis space contains the true labeling
function (aka "concept").
SOLUTION: False { PAC learning makes this assumption, but agnostic learning removes it
1.10 [2 points] When we fit a random forest, each decision tree is allowed to consider all available features.
SOLUTION: False { each tree uses a random subset of the feature
[Show More]