When evaluating nearest neighbor models, we can compute the accuracy on the training data, holding out one training point at a time so that a point is not considered its own neighbor. This is often referred to as leave-one-out cross-validation (LOOCV).

(a) Construct a dataset where 1-NN has an LOOCV accuracy of 0% but 3-NN has an LOOCV accuracy of 100%.

(b) Describe a dataset with n points (for some n 10 of your choice) where (n 1)-NN achieves LOOCV accuracy of 100%.