Ensemble Methods and Gradient Boosting Quiz

Answers and explanations appear as you take the quiz.

1. What is the core operation in an ensemble method?
   • Combining predictions from multiple models
   • Using one very large model instead of many small ones
   • Training a model on as much data as possible
   • Removing all weak learners from a model
2. How does gradient boosting differ from bagging in the usual training setup?
   • Boosting trains trees in parallel; bagging trains them sequentially
   • Boosting adds learners sequentially; independently fitted bagging learners can be trained in parallel
   • Boosting only works on images; bagging only on text
   • They are identical in practice
3. In gradient boosting, what target does a new tree approximate?
   • The original raw labels
   • A random subset of the features only
   • The negative gradient of the current loss, often called a pseudo-residual
   • The average of all previous trees
4. What is a possible effect of using a smaller learning rate in boosting?
   • To increase the size of each tree
   • To delete trees that perform poorly
   • To convert trees into neural networks
   • It can make updates more gradual and may require more trees
5. What type of input is used in the cited gradient-boosted-tree examples and studies?
   • Structured or tabular data
   • High-resolution images
   • Raw audio waveforms
   • Long natural-language documents
6. Which is a common regularization technique used to keep boosting from overfitting?
   • Removing the loss function
   • Limiting tree depth and subsampling rows/features
   • Using a learning rate of exactly 1.0 always
   • Training infinitely many trees
7. What does an out-of-bag score estimate in a bagged tree ensemble?
   • The accuracy on examples used by every tree
   • The number of leaves in the largest tree
   • Generalization performance using training examples omitted from a tree’s bootstrap sample
   • The learning rate of a gradient-boosted model
8. What does early stopping monitor during gradient-boosting training?
   • Only the training-set row order
   • Whether the model has no trees left
   • The number of categorical columns in a spreadsheet
   • A validation metric and stops adding trees when it no longer improves