I am trying vectorize text data for a Naive Bayes classifier that will be
trained in Hadoop and then the corresponding model will be deployed in a
Java app. My basic approach is to tokenize a string of text data using
Lucene and then encode each token using a StaticWordValueEncoder here are
the relevant code snippets
private static FeatureVectorEncoder memoEncoder = new
StaticWordValueEncoder("memo");
Vector v = new RandomAccessSparseVector(FEATURES);
StringReader reader = new StringReader(text);
StandardTokenizer source = new StandardTokenizer(Version.LUCENE_46, reader);
ShingleFilter sf = new ShingleFilter(source);
sf.setOutputUnigrams(true);
CharTermAttribute charTermAttribute =
sf.addAttribute(CharTermAttribute.class);
sf.reset();
while (sf.incrementToken()) {
memoEncoder.addToVector(charTermAttribute.toString(), 1,v);
}
In the Mahout in Action book I got the impression that the term "memo" will
seed the random number generator and I wanted to confirm that means I will
have consistency if I deploy this vectorizer in both my Hadoop environment
as well as my Java app. In particular, I am fixing the vector size to be
of length FEATURES and I am using "memo" as the name of my encoder. Will
those two things guarantee consistency of my text vectorization?
