Multiple linear regression (gradient descent) Flashcards
Q: How is multiple linear regression written using vector notation?
Q: What is the purpose of vectorizing parameters w1 to wn?
A: To treat them as a single vector w for more efficient computation.
Q: What does the cost function J depend on in vectorized notation?
A: It depends on the vector of parameters w and the scalar b.
Q: What is the difference in the error term between univariate and multivariate regression?
A: In multivariate regression, both w and
x are vectors.
Q: What is an alternate method to gradient descent for finding w and b?
A: The normal equation, which is a direct method using linear algebra techniques.
Q: What is a disadvantage of the normal equation method?
A: It’s not generalizable to other learning algorithms and can be slow with many features.
Q: What significant aspect makes the gradient descent more preferred?
A: Its adaptability to different learning problems and efficient implementation in libraries like NumPy.
Q: What are some tricks to improve the performance of multiple linear regression?
A: Properly scaling features and choosing a suitable learning rate α.
Q: Why is multiple linear regression considered widely used?
A: It is practical for a variety of prediction problems and easy to implement.
Q: Why is it recommended to understand NumPy for implementing regression?
A: NumPy efficiently handles large calculations and supports vectorization, crucial for implementing ML algorithms.
Q: What is essential before executing gradient descent efficiently?
A: Understanding how to calculate predictions and cost in multiple regression using vector operations.
Q: How can the choice of learning rate α affect performance?
A: The correct learning rate can significantly speed up convergence and improve model performance.
