Running the experiment

Remember, earlier we defined functions to break a sequence of tokens into features of various sorts: unigrams, bigrams, trigrams, and POS-tagged unigrams. We can take these and automatically test both the classifiers against all of these types of features. Let's see how.

First, we'll define some top-level variables that associate label keywords with the functions that we want to test at that point in the process (that is, classifiers or feature-generators):

(def classifiers
  {:naive-bayes a/k-fold-naive-bayes
:maxent a/k-fold-logistic})
(def feature-factories
  {:unigram t/unigrams
:bigram t/bigrams
:trigram t/trigrams
:pos (let [pos-model 
              (t/read-me-tagger "data/en-pos-maxent.bin")]
          (fn [ts] (t/with-pos pos-model ts)))})

We can now iterate over both of these hash maps and cross-validate these classifiers on these features. We'll average the error information (the precision and recall) for all of them and return the averages. Once we've executed...