In K-NN, is the query time longer than the training time?

• Yes

• No

Answer explanation

Recall K-NN algorithms are lazy learners!

James is trying to decide which machine learning algorithm to use for his project. Which of the following options is true about the k-NN algorithm?

• It can only be used for classification

• It can only be used for regression

• It can be used for both classification and regression

Answer explanation

Regression: considers the VALUE in majority of k neighbours

Classification: considers the CLASS in majority of k neighbours

Which of the following is true about Manhattan distances?

• It can be used for continuous variables

• It can be used for categorical variables

• It can be used for categorical as well as continuous variables

Answer explanation

Manhattan distance is the L1-norm and therefore cannot be used for categorical variables. Hamming distance would be a good choice for categorical variables.

Poppy is working on a machine learning project using k-NN and has a dataset with a lot of noise. What are the appropriate things to do with k-NN in this situation?

• Increase the value of K

• Decrease the value of K

• K does not depend on the noise

• None of these

Answer explanation

Recall when K=1 the decision boundaries perfectly match the data and its noise. This will result in overfitting. Increasing K makes K-NN more resilient to overfitting.

In K-NN, what is the effect of increasing/decreasing the value of K?

• The boundary becomes smoother with increasing values of K

• Smoothness of the boundary does not depend on the value of K

• The boundary becomes smoother with decreasing values of K

• None of these

Answer explanation

Think about what happens when K=1. Increasing K results in K-NN being less effected by noise as a large number of neighbours are considered when making a decision.

For embedded applications (i.e. running on a smartphone), what is the most appropriate family of algorithm?

• Eager learners

• Lazy learners

Answer explanation

Eager learners are desirable as all the heavy computation occurs at training time. Therefore the algorithm is time + compute efficient at inference time.

Given d the distance between a point of the dataset and the query point, which of the following weight functions is appropriate for Distance-Weighted k-NN?

• w = exp( -d )

• w = log ( min ( 0.25 * d, 1 ) )

• w = -d

Answer explanation

Recall that the weight you assign is inversely correlated with the distance metric. Both '-d' and 'log(min(0.25*d, 1)' yield negative distances which does not make sense. exp(-d) acts as a good distance metric as it increases exponentially with distance. This favours points close by and quickly ignores points far away.

Which of the following statements is true for k-NN classifiers?

• The classification accuracy is better with larger values of k

• The decision boundary is linear

• The decision boundary is smoother with smaller values of k

• k-NN does not require an explicit training step

Answer explanation

Too many points to address here. Please refer to lecture slides if confused.

The curse of dimensionality only affects k-NN?

• True

• False

Answer explanation

The curse of dimensionality plagues all machine learning algorithms. Hence the need for feature extractors and engineered priors.

Benjamin is using k-NN to classify types of flowers based on petal measurements. How does the choice of distance metric affect the performance of k-NN?

• It does not affect the performance

• It affects the accuracy of the model

• It only affects the speed of the algorithm

• It affects both accuracy and speed

Answer explanation

The choice of distance metric in k-NN directly influences how distances between data points are calculated, impacting both the accuracy of classifications and the computational speed of the algorithm.