Theory test: Learning without labels

6 questions. Pass with 5 correct.

1. 01 In the assignment step of k-means, each point is assigned to:

   A random clusterThe nearest centreThe largest clusterThe cluster with the most similar label
2. 02 Which statement about k-means is true?

   It always finds the global minimumIt always converges, but possibly to a local minimumIt can diverge if k is too largeIt needs labelled data
3. 03 The first principal component is:

   The feature with the largest valuesThe direction of maximum variance (the top eigenvector of the covariance matrix)The direction that best separates the classesThe mean of the data
4. 04 The first two components explain 70% and 20% of the variance. Together they explain:

   70%90%14%It can't be determined
5. 05 Why standardise features before PCA when they have different units?

   PCA requires values between 0 and 1Otherwise the features with the largest numerical variance dominate the componentsIt makes PCA supervisedIt isn't needed
6. 06 In a t-SNE plot, two clusters are far apart. You can conclude:

   They are very different in the original spaceNot much: distances between clusters in t-SNE plots aren't reliableThey have the same labelsThe perplexity was too low