chapter4-sml

What is the purpose of supervised machine learning?

• To generate random predictions

• To train models using labeled data

• To perform unsupervised learning

• To analyze unstructured data

What is the error function used to analyze the performance of an already learned model?

• Misclassification rate

• Squared error

• Cross-validation error

• Loss function

What is the purpose of hold-out validation data in machine learning?

• To test the model on training data

• To visualize the data distribution

• To train the model

• To estimate the new data error

What is the main advantage of k-fold cross-validation over hold-out validation?

• It is faster to implement

• It requires a larger validation dataset

• It provides an unbiased estimate of the new data error

• It uses less computational resources

Why is it important to have a separate test dataset in machine learning?

• To evaluate the model's performance on unseen data

• To increase the training data size

• To estimate the test error

• To validate the model during training

What is the ultimate goal in supervised machine learning?

• To overfit the training data

• To increase the model complexity

• To achieve a small new data error

• To minimize the training error

Answer explanation

The ultimate goal in supervised machine learning is to achieve a small new data error, indicating the model's ability to generalize well to unseen data.

What is the purpose of selecting models and hyperparameters based on k-fold cross-validation?

• To minimize the training error

• To increase the model complexity

• To overfit the validation data

• To obtain an unbiased estimate of the new data error

What insight can be gained by further analyzing the new data error mathematically?

• The training data distribution

• The generalization gap

• The feature importance

• The model complexity

What does the bias term in the bias-variance decomposition of the expected squared error represent?

• Systematic error in the model that remains regardless of the training data

• Amount of noise in the experiment each time it is performed

• Random variable representing the model learned from training data

• Irreducible error due to the assumption of independent noise

In the bias-variance decomposition, what does the variance term describe?

• How much the model varies each time it is trained on a different dataset

• Irreducible error due to the assumption of independent noise

• Systematic error in the model that remains regardless of the training data

• Amount of noise in the experiment each time it is performed