CPSC 330 Lecture 15: K-Means

Andrew Roth (Slides adapted from Varada Kolhatkar and Firas Moosvi)

Announcements

• MT2 registartion open. See Piazza.
• Final exam dates and registration posted. See Piazza.
• MT1 marked. Viewing in CBTF bookings open. See Piazza.
• HW5 is due March 10th (start now!)
• HW6 is posted

(iClicker) Midterm poll

iClicker cloud join link: https://join.iclicker.com/HTRZ

Select all of the following statements which are TRUE.

• 1. I’m happy with my progress and learning in this course.
• 2. I find the course content interesting, but the pace is a bit overwhelming. Balancing this course with other responsibilities is challenging
• 3. I’m doing okay, but I feel stressed and worried about upcoming assessments.
• 4. I’m confused about some concepts and would appreciate more clarification or review sessions.
• 5. I’m struggling to keep up with the material. I am not happy with my learning in this course and my morale is low :(.

Class notebook (recap)

(iClicker) Exercise 14.2

iClicker cloud join link: https://join.iclicker.com/HTRZ

Select all of the following statements which are TRUE.

• 1. L2 regularized logistic regression sets correlated coefficients to zero *
• 2. Lasso regression predicts values using ŷ_i = ∑_jw_jx_ij + b *
• 3. Lasso and Ridge regression optimize different measures of “error” between predicted and observe target values *
• 4. KNN and SVM RBF can be used with SelectFromModel
• 5. I saw way to much math today! *

* Denotes optional details only covered in our section

(iClicker) Exercise 14.3

iClicker cloud join link: https://join.iclicker.com/HTRZ

Select all of the following statements which are TRUE.

• 1. Simple association-based feature selection approaches do not take into account the interaction between features.
• 2. You can carry out feature selection using linear models by pruning the features which have very small weights (i.e., coefficients less than a threshold).
• 3. The order of features removed given by rfe.ranking_ is the same as the order of original feature importances given by the model.
• 4. If you remove 10 features in a single step based on feature importance, the same 10 features would be removed if we performed sequential removal calculating feature importance after removing each feature.
• 5. Forward search is guaranteed to find the best feature set. *

* Denotes optional question

Class notebook

15.1 Select all of the following statements which are True (iClicker)

iClicker cloud join link: https://join.iclicker.com/HTRZ

Select all that are true.

• 1. K-Means algorithm always converges to the same solution.
• 2. K in K-Means should always be ≤ # of features.
• 3. In K-Means, it makes sense to have K ≤ # of examples.
• 4. In K-Means, in some iterations some points may be left unassigned.

15.2 Select all of the following statements which are True (iClicker)

iClicker cloud join link: https://join.iclicker.com/HTRZ

Select all that are true.

• 1. K-Means is sensitive to initialization and the solution may change depending upon the initialization.
• 2. K-means terminates when the number of clusters does not increase between iterations.
• 3. K-means terminates when the centroid locations do not change between iterations.
• 4. K-Means is guaranteed to find the optimal solution.

Class notebook

15.3 Select all of the following statements which are True (iClicker)

iClicker cloud join link: https://join.iclicker.com/HTRZ

Select all that are TRUE.

• 1. If you train K-Means with n_clusters= the number of examples, the inertia value will be 0.
• 2. The elbow plot shows the tradeoff between within cluster distance and the number of clusters.
• 3. Unlike the Elbow method, the Silhouette method is not dependent on the notion of cluster centers.
• 4. The elbow plot is not a reliable method to obtain the optimal number of clusters in all cases.
• 5. The Silhouette scores ranges between -1 and 1 where higher scores indicates better cluster assignments.

Class demo