lopusz / milelab

milelab

MachIne LEarning Lab

Currently, this library contains only a wrapper to LDA (Latent Dirichlet Allocation) computed by the Mallet package.

Examples

Dead simple analysis with two topics:

(require '[clojure.pprint :as pp])
(require '[ milelab.instances :as mi])
(require '[ milelab.topics :as mt])

(def ptm
  (let [
         num-topics 2
         doc-names [ 1 2 3 4 ]
         doc-data
         [ [ "Jane" "lives" "in" "Warsaw" ]
           [ "Warsaw" "is" "a" "beautiful" "city" ]
           [ "Sister" "of" "Jane" "lives" "in" "LA" ]
           [ "LA" "is" "far" "away" "from" "Warsaw"] ]
         instance-list (mi/create-instance-list doc-names doc-data)
        ]
      (mt/estimate-topics instance-list num-topics)))
(pp/pprint (mt/get-topic ptm 0))
(pp/pprint (mt/get-topic ptm 1))

Note that get-topic returns java class LinkedHashMap mapping word to its probability in a given topic. This Java class guarantees that iteration order through words with decreasing probability. You can convert this LinkedHashMap to Clojure immutable hash map, however, this looses the iteration order guarantee.

(pp/pprint (into (hash-map) (mt/get-topic ptm 0)))

License

Distributed under the Eclipse Public License, the same as Clojure.