Model Evaluation Checkpoint

Trace the representation, diagnose a boundary failure, and reconstruct a changed case.

Get every question correct to pass this checkpoint. You can review feedback, revise your answers, and submit again.

1. TP=12, FP=3, FN=8. What is precision? Select one.

Select one answer.

0.8 0.6 0.5 0.75

2. For the same counts TP=12, FP=3, FN=8, what is recall? Select one.

Select one answer.

0.75 0.6 0.4 0.8

3. Only 1 of 100 cases is positive. A classifier always predicts negative. What are accuracy and positive recall? Select one.

Select one answer.

0% and 0% 99% and 0% 99% and 99% 1% and 100%

4. Precision and recall both equal 0.5. What is F1?

5. You repeatedly choose a threshold using the test set. Why is its final score unreliable as an independent estimate? Select one.

Select one answer.

Thresholds never affect classification Precision becomes mathematically undefined Test sets must have balanced labels The test labels have influenced model selection

6. Scores [0.9,0.8,0.7] have labels [1,0,1]. Predict positive for score >= threshold. Lowering threshold from 0.8 to 0.7 does what? Select one.

Select one answer.

Precision decreases and recall decreases Precision stays 1/2 and recall stays 1/2 Precision becomes 1 and recall becomes 2/3 Precision rises from 1/2 to 2/3 and recall rises from 1/2 to 1

7. What should confusion_matrix([1,0,1],[0,0,1]) return? Select one.

Select one answer.

[[1,0],[1,1]] {'TP':1,'TN':1,'FP':1,'FN':0} {'TP':1,'TN':1,'FP':0,'FN':1} {'TP':2,'TN':1,'FP':0,'FN':0}

8. To estimate performance on new users, what is wrong with randomly splitting individual records when each user has many records? Select one.

Select one answer.

Train and evaluation can contain the same user, changing the generalization question Randomness prevents metrics from being computed Every split must have exactly equal classes A test set must be larger than the training set

Auto-advance on pass

Join Discord