1.We just need to try out a few values before settling on one.
2.Low values of K can be noisy and subject to the effects of outliers.
3.Large values of K smooth over thins, but if k is very large, a category with only a few samples will always be out voted by other categories.