Large-scale clustering with Mahout

Mahout (http://mahout.apache.org/) is a machine learning library intended for use in distributed computing environments. Version 0.9 of the library targets Hadoop and is the version we'll be using here.

Note

At the time of writing, Mahout 0.10 has just been released and also targets Spark. Spark is an alternative distributed computing framework that we'll be introducing in the next chapter.

We saw in the previous chapter that one of Hadoop's abstractions is the sequence file: a binary representation of Java keys and values. Many of Mahout's algorithms expect to operate on sequence files, and we'll need to create one as input to Mahout's k-means algorithm. Mahout's k-means algorithm also expects to receive its input as a vector, represented by one of Mahout's vector types.

Although Mahout contains classes and utility programs that will extract vectors from text, we'll use this as an opportunity to demonstrate how...