Module 12 Knowledge Check
5 questions. Pass mark 4 out of 5. Answer every question before checking the answer key below, then retry after reading the feedback.
1. Why write the problem framing before touching the data?
- A. To satisfy a rubric
- B. Because the error costs decide the threshold, and the threshold decides what the model does
- C. To estimate training time
- D. To choose an architecture
2. 600 rows covering 200 learners, three rows each. How should you split?
- A. Randomly by row
- B. By learner, so no learner appears on both sides
- C. By date
- D. 80/20 with shuffling
3. Your network beats the majority baseline but ties logistic regression. What should you ship?
- A. The network
- B. Logistic regression
- C. Both
- D. Neither
4. Best validation occurred at epoch 60 and you trained to 250 without checkpointing. What happened?
- A. Nothing — the final model is fine
- B. You spent 190 epochs making it worse and discarded the best model
- C. The model converged
- D. Validation loss is unreliable
5. A missed at-risk learner costs far more than an unnecessary contact. What follows for the threshold?
- A. Keep it at 0.5
- B. Lower it to favour recall, and record why
- C. Raise it to favour precision
- D. The threshold is unrelated
---
Answer Key and Explanations
Check these only after attempting every question.
1. B — Because the error costs decide the threshold, and the threshold decides what the model does
A model trained without it optimises a number nobody asked for, and 0.5 becomes a threshold chosen by default rather than by evidence.
2. B — By learner, so no learner appears on both sides
A row split puts the same learner in train and test, so the model recognises the learner rather than the pattern.
3. B — Logistic regression
It is faster, interpretable and easier to maintain. The network has not earned its complexity.
4. B — You spent 190 epochs making it worse and discarded the best model
That is what epochs_trained_past_best is for, and why checkpointing on validation matters.
5. B — Lower it to favour recall, and record why
The cost asymmetry decides the cut-off. Recording the reason lets a reviewer challenge the trade-off rather than guess at it.
Practical Check
Apply this module to your own work: complete the module activity for *Capstone: Deep Learning Portfolio and Model Card*, then write one sentence naming what your result shows and one naming what it does not.
Strong Answer Pattern
A strong answer names the task, the evidence used, the check performed, and the remaining limitation. It avoids "proved", "guaranteed", or "always" unless the evidence genuinely supports it.