Maths · Module 5Module material
Numbers on a Computer interview questions
10 questions interviewers ask on this module’s topics, from first principles to the follow-ups. Each comes with a model answer.
- Why does
0.1 + 0.2 == 0.3return False? - What is machine epsilon?
- What is catastrophic cancellation? Give an example.
- How do you choose the step for a numerical gradient check?
- What are overflow and underflow, and how does NumPy report them?
- Why do Naive Bayes implementations add log probabilities?
- What is log-sum-exp, and why is it stable?
- How do you compute a softmax safely?
- Why should log loss be computed from the score rather than the probability?
- When would you train in float32 instead of float64, and what do you watch for?