Virtual Labs

Q1. What is the effect of correlation between two features on the feature space representation?

a: Correlation can help in identifying the relationship between features. Explanation

Explanation

b: Highly correlated features may provide redundant information. Explanation

Explanation

c: Correlation can affect the shape and orientation of the feature space. Explanation

Explanation

d: All of the above Explanation

Explanation

Q2. Is it possible that the addition of a new feature decrease the accuracy of the NN classifier? How?

a: Yes, due to the curse of dimensionality. Explanation

Explanation

b: No, more features always improve accuracy. Explanation

Explanation

c: Yes, because new features can add noise. Explanation

Explanation

d: No, the accuracy remains the same. Explanation

Explanation

Q3. Is it correct that for nearest neighbor classifier, it is always best to have all features scaled to the same range? Show why or give a counter example.

a: Yes, scaling features helps to ensure each feature contributes equally. Explanation

Explanation

b: No, scaling features does not impact NN classifiers. Explanation

Explanation

c: Yes, because unscaled features can dominate the distance calculation. Explanation

Explanation

d: No, sometimes unscaled features provide better results. Explanation

Explanation

Q4. Can the nearest neighbor classifier be applied to feature vectors of different length? Is it possible to do so if you have a function that takes in two feature vectors (possibly of different lengths), and outputs the similarity/distance between them? Suggest a possible function of this nature, and a classification problem, where it can be applied.

a: Yes, by using a custom distance function like Dynamic Time Warping (DTW). Explanation

Explanation

b: No, all feature vectors must be of the same length. Explanation

Explanation

c: Yes, by padding shorter vectors with zeros. Explanation

Explanation

d: No, using vectors of different lengths would lead to errors. Explanation

Explanation

Feature Representation