QuizFind out where it breaks/10 min
Checkpoint: honest testing
Eight questions on training examples, labels, splitting data, overfitting, and which score you report.
8 questions. You need 70% to pass, and you can retake it as many times as you want.
1
What is one labeled training example for the image lab?
2
Forty nearly identical photos show one mug. What would improve coverage?
3
Two people disagree on what ripe means. What should happen before training?
4
Which evaluation better estimates performance on new images?
5
When should you reserve the final test set?
6
Training accuracy is high but development accuracy is low. What does that suggest?
7
With true classes in rows, where is a glass predicted as mug recorded?
8
A shoe receives a very high mug score. What does this demonstrate?
0 of 8 answered