Check the code of tree and rpart packages to see how they select attributes at each node. We also would like to know how they handle numeric attributes. Based on you observeration do you think that decision tree is a suitable method for the USPS set?
Write a short report (<= 2 pages in English).