AIML Rounds

QUIZ

Other classifiers

5 questions on Other classifiers, from 4 lessons. Pick an answer in your head, then open it.

On the roadmap for: ML Engineering, Data Science.

  1. Question 1 of 5

    Of all the lines that separate two classes, which does a support vector machine pick?

    1. A. The first line the algorithm finds
    2. B. A random line among the ones that work
    3. C. The one with the widest margin to each class
    4. D. The one through the origin, so it needs no bias term
    View answer

    ✓ C. The one with the widest margin to each class

    A wide margin generalises better; only the closest points, the support vectors, decide it.

    Lesson: Support Vector Machines

  2. Question 2 of 5

    What does the kernel trick let an SVM do?

    1. A. Draw non-linear boundaries without building the space
    2. B. Train faster on data that's already linearly separable
    3. C. Handle missing values in the features automatically
    4. D. Remove the margin so more points can fit inside it
    View answer

    ✓ A. Draw non-linear boundaries without building the space

    Kernels such as the RBF kernel give curved boundaries at the cost of a linear model.

    Lesson: Support Vector Machines

  3. Question 3 of 5

    When does k-nearest neighbours tend to fail?

    1. A. With many or unscaled features; slow to predict
    2. B. Never: it works equally well on any kind of data
    3. C. When the classes are evenly balanced in the data
    4. D. With small datasets of only a few hundred rows
    View answer

    ✓ A. With many or unscaled features; slow to predict

    In high dimensions all points look roughly equally far apart. Scale features and reduce dimensions.

    Lesson: k-Nearest Neighbours

  4. Question 4 of 5

    Naive Bayes' independence assumption is almost always wrong. Why does it still classify well?

    1. A. The independence assumption isn't actually wrong for text data
    2. B. It ignores the data and uses priors only
    3. C. Classifying only needs the right class to score highest
    4. D. It uses deep learning under the hood
    View answer

    ✓ C. Classifying only needs the right class to score highest

    Its probabilities are often over-confident, but the ranking of classes is frequently correct.

    Lesson: Naive Bayes

  5. Question 5 of 5

    What is the difference between LDA and QDA?

    1. A. There's no real difference between them
    2. B. LDA is built from decision trees
    3. C. QDA is unsupervised, while LDA needs labels for training
    4. D. LDA: one shared covariance; QDA: one per class
    View answer

    ✓ D. LDA: one shared covariance; QDA: one per class

    QDA is more flexible but needs more data to estimate each covariance.

    Lesson: Discriminant Analysis: LDA and QDA