Python Machine Learning Cookbook - Second Edition

There's more...

The main advantages of this method are that it does not require learning or the construction of a model; it can adapt its decision boundaries in an arbitrary way, producing a representation of the most flexible model; and it also guarantees the possibility of increasing the training set. However, this algorithm also has many drawbacks, including being susceptible to data noise, being sensitive to the presence of irrelevant features, and requiring a similarity measure to evaluate proximity.