Topic 16

Question 1

Crossvalidation

Answer 1

Partition the data available for training into a training dataset and a validation dataset, allowing us to diagnose overfitting without seeing the real test data

Question 2

Why not use test data to choose K?

Answer 2

We won’t correctly evaluate the model’s performance when we have truly unlabeled data

Question 3

Bias and Variance Depending on the Chosen K Value for Validation

Answer 3

Low K: high bias, low variance, High K: low bias, high variance