[http://bit.ly/k-NN] The nearest-neighbour algorithm is sensitive to the choice of distance function. Euclidean distance (L2) is a common choice, but it may lead to sub-optimal performance. We discuss Minkowski (p-norm) distance functions, which generalise the Euclidean distance, and can approximate some logical functions (AND, OR). We also mention similarity/distance measures appropriate for histogram data and for text.
