Stanford Machine Learning Practice Test

Answer choices

1. A. Separating training data from evaluation data
2. B. Using clustering before every prediction task
3. C. Skipping generalization checks
4. D. Treating every feature as a target

Wrong-answer review

• B. Using clustering before every prediction task: This option handles a neighboring evaluation or preprocessing idea rather than the requested concept.
• C. Skipping generalization checks: This option points to a separate optimization, modeling, or data-preparation step.
• D. Treating every feature as a target: This option follows another ML workflow path rather than the requirement described here.

Answer choices

1. A. Image denoising only
2. B. Recommender systems
3. C. Polynomial feature expansion
4. D. Gradient checking

Answer choices

1. A. Whether the labels should be removed from training
2. B. Whether clustering would fix class imbalance automatically
3. C. Whether accuracy alone is the right evaluation metric
4. D. Whether the optimizer should be replaced by k-means

Answer choices

1. A. Nearest-neighbor voting
2. B. Hard clustering
3. C. Backpropagation
4. D. Manual feature labeling

Answer choices

1. A. It focuses only on one supervised algorithm for the entire program
2. B. It also covers unsupervised learning, recommender systems, tree-based models, and neural networks
3. C. It excludes all practical implementation work with Python tools
4. D. It removes any discussion of model evaluation and overfitting

Answer choices

1. A. Increase the number of clusters k until the distortion drops to exactly zero
2. B. Run the k-means algorithm multiple times with different random centroid initializations, and select the run that yields the lowest final distortion (cost)
3. C. Initialize all centroids at the exact coordinate origin (all zeros) to ensure deterministic convergence
4. D. Apply L2 regularization to the centroid coordinates to smooth the cost surface

Answer choices

1. A. K-means clustering
2. B. PCA
3. C. Nearest-centroid segmentation
4. D. Decision trees

Answer choices

1. A. Skip evaluation and retrain immediately
2. B. Tune only the learning rate
3. C. Remove unusual examples before reviewing them
4. D. Error analysis

Answer choices

1. A. Train a single binary classifier that predicts a continuous value from 1 to 4, then round the output to the nearest integer
2. B. Train six pairwise binary classifiers and use a voting system to determine which category wins the most duels
3. C. Train four separate binary classifiers, one for each category, and assign the ticket to the category whose classifier outputs the highest probability
4. D. Perform k-means clustering with k=4 on the support tickets and assign the categories based on the final cluster centroids

Answer choices

1. A. Regularization failure
2. B. Convex optimization
3. C. Cold start
4. D. One-hot encoding