feat(verbose): print more information using -v or --verbose flags

examples/naivebayes-doc-classifier.hs

@@ -8,32 +8,55 @@ module Main
     import Debug.Trace
     import Data.List.Split
     import Control.Arrow ((&&&))
+    import Control.Monad (when)
+    import System.Environment
 
     main = do
+      args <- getArgs
       dataset <- readFile "examples/doc-classifier-data/data-reuters"
       test <- readFile "examples/doc-classifier-data/data-reuters-test"
 
       classes <- map (filter (/= ' ')) . lines <$> readFile "examples/doc-classifier-data/data-classes"
       sws <- lines <$> readFile "examples/stopwords"
 
+      let verbose = or [elem "-v" args, elem "--verbose" args]
+      when (not verbose) $ putStrLn "use --verbose to print more information"
+
       let intClasses = [0..length classes - 1]
           documents = cleanDocuments $ removeWords sws $ createDocuments classes dataset
           testDocuments = cleanDocuments $ createDocuments classes test
           devTestDocuments = take 30 testDocuments
           nb = train documents intClasses
 
-          results = map (\(Document text c) -> (c, run text nb)) testDocuments
+          results = session testDocuments nb
-          -- results = map (\(Document text c) -> (c, run text nb)) devTestDocuments
+          -- results = session devTestDocuments nb
 
-      print (text $ head documents)
+      when verbose $ print (text $ head documents)
 
       let showResults (c, (r, confidence)) = putStrLn (classes !! c ++ " ~ " ++ classes !! r)
-      mapM_ showResults results
+      when verbose $ mapM_ showResults results
 
-      putStrLn $ "Recall: " ++ show (recall results)
+      when verbose $
-      putStrLn $ "Precision: " ++ show (precision results)
+        putStrLn $ "The training data is imbalanced which causes the classifier to be biased towards\n"
-      putStrLn $ "F Measure: " ++ show (fmeasure results)
+                ++ "some classes, `earn` is an example, the class alone has around 90% accuracy while\n"
-      putStrLn $ "Accuracy: " ++ show (accuracy results)
+                ++ "the rest of classes have a much lower accuracy and it's commonly seen that most inputs\n"
+                ++ "are incorrectly classified as `earn`.\n"
+
+      let
+        accuracies =
+          let as = zip intClasses $ map (\c -> filter ((==c) . fst) results) intClasses
+              av = filter (not . null . snd) as
+              calculated = map (fst &&& accuracy . snd) av
+          in sortBy (\(_, a) (_, b) -> b `compare` a) calculated
+
+      when verbose $
+        mapM_ (\(c, a) -> putStrLn $ "Accuracy(" ++ classes !! c ++ ") = " ++ show a) accuracies
+
+      putStrLn $ "\nAverages: "
+      putStrLn $ "Recall = " ++ show (recall results)
+      putStrLn $ "Precision = " ++ show (precision results)
+      putStrLn $ "F Measure = " ++ show (fmeasure results)
+      putStrLn $ "Accuracy = " ++ show (accuracy results)
 
     createDocuments classes content =
       let splitted = splitOn (replicate 10 '-' ++ "\n") content

src/Sibe/NaiveBayes.hs

@@ -3,6 +3,7 @@ module Sibe.NaiveBayes
    NB(..),
    train,
    run,
+   session,
    ordNub,
    accuracy,
    precision,
@@ -91,22 +92,30 @@ module Sibe.NaiveBayes
               grams = map (\(i, w) -> concat . intersperse "_" $ w:((take (n - 1) . drop (i+1)) ws)) pairs
           in unwords ("<b>":grams)
 
+    session :: [Document] -> NB -> [(Class, (Class, Double))]
+    session docs nb =
+      let results = map (\(Document text c) -> (c, run text nb)) docs
+      in results
+
     run :: String -> NB -> (Class, Double)
     run txt (NB documents classes vocabulary megadoc cd cw cgram) =
       let scores = map (score . fst) classes
           index = argmax scores
           m = maximum scores
-          confidence = m / sum scores
+      in (index, m)
-      in (index, 0)
       where
         score c =
           let prior = snd (classes !! c)
 
           -- below is the formula according to Multinominal Naive Bayes, but it seems
-          -- using a uniform prior probability seems to work better
+          -- using a uniform prior probability seems to work better when working with imbalanced
+          -- training datasets, instead, we help rare classes get higher scores using
+          -- alpha = (1 - prior * ALPHA)
           -- in prior * product (map (prob c) (words txt))
 
-          in product (map (prob c) (words txt))
+              alpha = 1 - (log 1 + prior)
+
+          in alpha * product (map (prob c) (words txt))
 
         prob c w =
           let tctM = find ((== w) . fst) (snd (cw !! c))