(2) Curse of dimensionality: Nearestneighbor classication tends to be more successful when the test point has training points in close proximity. In this problem we investigate whether this is feasible for high dimensional datasets. Suppose our training points S = {231,1'2, ..., urn} all have the same class label, and are chosen independently and uniformly from the unit ball B dened by: B = {1176le : HIEHQ S 1} (a) Suppose our test point is located at the origin, and that the nearestneighbor classier is accurate if there is a training sample in S within 0.1 of the origin (measured in Euclidean distance). In terms of n and d, what is the probability the nearestneighbor estimator is accurate? (b) What does your answer to the previous part say as d becomes large? I.e.7 roughly how much data is necessary for the probability of accurate classication to be larger than some constant