Class FeatureDocFreqPipe

  • All Implemented Interfaces:

    public class FeatureDocFreqPipe
    extends Pipe
    Pruning low-count features can be a good way to save memory and computation. However, in order to use Vectors2Vectors, you need to write out the unpruned instance list, read it back into memory, collect statistics, create new instances, and then write everything back out.

    This class supports a simpler method that makes two passes over the data: one to collect statistics and create an augmented "stop list", and a second to actually create instances.

    See Also:
    Serialized Form
    • Constructor Detail

      • FeatureDocFreqPipe

        public FeatureDocFreqPipe()
      • FeatureDocFreqPipe

        public FeatureDocFreqPipe​(Alphabet dataAlphabet,
                                  Alphabet targetAlphabet)
    • Method Detail

      • pipe

        public Instance pipe​(Instance instance)
        Description copied from class: Pipe
        Really this should be 'protected', but isn't for historical reasons.
        pipe in class Pipe
      • addPrunedWordsToStoplist

        public void addPrunedWordsToStoplist​(SimpleTokenizer tokenizer,
                                             double docFrequencyCutoff)
        Add all pruned words to the internal stoplist of a SimpleTokenizer.
        docFrequencyCutoff - Remove words that occur in greater than this proportion of documents. 0.05 corresponds to IDF >= 3.