clj-naive-bayes

Warning: This project is under heavy development. Things will break!

Usage

First of all you will need a new classifier:

(require '[clj_naive_bayes.core :as nb])

(def my-classifier (nb/new-classifier {:name :ngram-nb :ngram-size 2 :ngram-type :multinomial}))

Available options are

• :name : Currently :ngram-nb, :multinomial-nb and :binary-nb are supported. (Default :multinomial-nb)

• :ngram-size : Sets ngram size. (Default 2)

• :ngram-type : Whether the ngram should be :binary or :multinomial

• :boost-start : Boolean. (Default false). This flag has only effect with ngrams.

• :keep-sorted : Boolean. (Default false). With this flag on all tokens in ngram keys are stores in alphabetical order.

Train

Suppose you have a training dataset. This should be a CSV file, consisting of lines with <document,class> or <document,class,count> elements. In the second case, the count column should contain the number of occurences of each sample. This is purely for space-saving purposes, so e.g. instead of using five lines of the same <document,class> pair, a single <document,class,5> line can be used instead.

(require '[clj_naive_bayes.train :as train])

(train/parallel-train-from-file my-classifier "resources/train.csv" :limit 400000)

Classify

Now we can try classifying a new document:

(nb/classify my-classifier "iphone 6s")
=> "40"

Export Probabilities to a Hashmap

This could be useful for e.g. persisting the classifier:

(def out (nb/export a))
=> #'user/out
(keys out)
=> (:terms :cats)

Evaluate Performance

(use 'clj_naive_bayes.core)
(use 'clj_naive_bayes.eval)

(def logs (parallel-classifications my-classifier "resources/test.json"))

Persist classifiers

Currently only file disk persistance is supported. Suppose you have a trained classifier named my-classifier you can write it to a file:

(use 'clj_naive_bayes.utils)

(persist-classifier my-classifier "resources/data.clj")

And later on load it:

(use 'clj_naive_bayes.utils)

  (load-classifier my-classifier "resources/data.clj")

Testing

lein test will run all tests.

lein test [TEST] will run only tests in the TESTS namespaces.

Tooling

Kibit

lein kibit will analyze code

Marginalia

lein marg will produce documentation under /docs