01
True or false
1 point
A held-out test set can estimate a model’s performance on future data when the test examples are sufficiently similar to the data the model will encounter in use.
Free Practice Quiz Question List
8 Generalization and Overfitting Online Quiz Questions
Use this free practice quiz with 20 questions to review 8 Generalization and Overfitting, test your knowledge, and prepare for your next test or exam.
20 questions
02
Choose one
1 point
A model has poor performance on both its training data and its validation data. Which interpretation best fits this score pattern?
- A
Overfitting
- B
Underfitting
- C
A useful fit with a modest gap
03
Fill in the blank
1 point
A model fits its training examples closely, including random noise that does not recur, and performs substantially worse on new examples. This is called .
04
Choose all
1 point
Which actions are suggested ways to address a model that is overfitting? Select all that apply.
- A
Reduce the model’s complexity
- B
Remove unhelpful features
- C
Make the model more expressive
- D
Use regularization
05
Written response
1 point
Which regularization method discourages excessively large model weights?
06
Fill in the blank
1 point
If a preprocessing step is learned using the full dataset before it is split into training and evaluation data, information can improperly cross the split. This problem is called .
07
Choose one
1 point
A model performs well on a held-out test set, but the distribution of real-world examples later changes. What is the most appropriate conclusion?
- A
The original test score guarantees the model will perform just as well.
- B
The model must be overfitting because its test score was high.
- C
Performance may change because the real-world data distribution has changed.
- D
The training data should be used as the final evaluation set.
08
True or false
1 point
Adding more training examples is guaranteed to improve performance, even if the new examples do not represent the cases the model must handle.
- A
True
- B
False
09
Choose all
1 point
Which practices help preserve a meaningful final evaluation while comparing models and settings? Select all that apply.
- A
Compare candidate models using cross-validation.
- B
Choose the model based on repeated adjustments to its test-set results.
- C
Keep a separate test set for the final evaluation.
- D
Use the test set to fit each candidate model.
10
Choose one
1 point
What can learning curves and validation curves help reveal?
- A
They guarantee that the validation data match future real-world data.
- B
They replace the need for a separate final test set.
- C
They can help reveal underfitting or overfitting as training-set size or model settings change.
- D
They improve performance by automatically removing noisy examples.
11
Choose one
1 point
Which sequence best describes a practical workflow for improving model performance?
- A
Establish a simple baseline, inspect training and validation performance, make a targeted change, and compare using the same validation method.
- B
Start with the most complex model and select it using the test set.
- C
Change several parts of the model at once and compare scores from different validation methods.
- D
Skip baseline measurements and evaluate only on training data.
12
Open ended
1 point
A model performs poorly on both its training data and validation data. Explain the likely diagnosis and describe suitable next steps for improving it.
13
Written response
1 point
What is the term for a model’s ability to make useful predictions on examples it did not train on?
14
Choose one
1 point
A model’s validation score looks strong, but the metric may not reflect what matters for the task. What should you do?
- A
Keep the metric because it is commonly used, regardless of the task.
- B
Check whether the metric reflects the goal of the task.
- C
Use training performance alone to decide whether the metric is suitable.
- D
Choose whichever metric gives the model the highest score.
15
Choose one
1 point
A model scores very well on its training set but substantially worse on a representative validation set. Which interpretation is most likely?
- A
Underfitting
- B
Overfitting
- C
Useful fit with a modest gap
- D
Data leakage
16
Choose one
1 point
A straight-line model performs poorly on both training and validation data, but the underlying relationship is strongly curved. Which change most directly addresses the likely problem?
- A
Add more training examples without checking what they represent
- B
Increase regularization to penalize complexity more strongly
- C
Try a more expressive model that can capture curvature
- D
Stop training as soon as validation performance stops improving
17
Written response
1 point
What resampling method is recommended for comparing candidate models and settings?
18
Choose one
1 point
When can performance on a held-out test set serve as a useful estimate of a model's performance on future data?
- A
The test examples are relevant to the intended use and sufficiently similar to future data
- B
The model achieves excellent training performance, regardless of the test examples
- C
The test set is repeatedly used to choose model settings
- D
The test set is larger than the training set
19
True or false
1 point
True or false: After selecting a model using validation results, it is good practice to keep a separate test set for final evaluation.
- A
True
- B
False
20
Written response
1 point
What technique penalizes model complexity, with L2 being an example that discourages excessively large weights?