nlp-scores 0.5.2 → 0.5.4

raw patch · 2 files changed

+22/−3 lines, 2 files

Files

• NLP/Scores.hs +21/−2
• nlp-scores.cabal +1/−1

@@ -26,6 +26,9 @@     , ari     , mi     , vi+    -- * Comparing probability distributions+    , kullbackLeibler+    , jensenShannon     -- * Auxiliary types and functions     , Count     , Counts@@ -103,6 +106,24 @@ vi cs@(Counts _ cx cy) = entropy (elems cx) + entropy (elems cy) - 2 * mi cs   where elems = Map.elems +-- | Kullback-Leibler divergence: KL(X,Y) = SUM_i P(X=i) log_2(P(X=i)/P(Y=i)). +-- The distributions can be unnormalized.+        +kullbackLeibler :: (Eq a, Floating a, F.Foldable f, T.Traversable t) => t a -> f a -> a+kullbackLeibler xs ys = sum . zipWithTF f xs $ ys+  where f !x !y = let px = x / sx in px `mult` logBase 2 (px/(y/sy))+        sx = sum xs+        sy = sum ys+        mult 0 _ = 0+        mult w p = w * p+        {-# INLINE mult #-}  ++-- | Jensen-Shannon divergence: JS(X,Y) = 1/2 KL(X,(X+Y)/2) + 1/2 KL(Y,(X+Y)/2).+-- The distributions can be unnormalized.+jensenShannon :: (Eq a, Floating a, T.Traversable t, T.Traversable u) => t a -> u a -> a+jensenShannon xs ys = 0.5 * kullbackLeibler xs zs + 0.5 * kullbackLeibler ys zs+  where zs = zipWithTF (+) xs ys+           -- | Adjusted Rand Index: <http://en.wikipedia.org/wiki/Rand_index> ari :: (Ord a, Ord b) => Counts a b -> Double ari (Counts cxy cx cy) =  (sum1 - sum2*sum3/choicen2) @@ -115,8 +136,6 @@ -- | The sum of a sequence of numbers sum :: (F.Foldable t, Num a) => t a -> a sum = F.foldl' (+) 0-{-# SPECIALIZE sum :: [Double] -> Double #-}-{-# SPECIALIZE sum :: [Int] -> Int #-} {-# INLINE sum #-}  -- | The mean of a sequence of numbers.

@@ -7,7 +7,7 @@ -- The package version. See the Haskell package versioning policy -- (http://www.haskell.org/haskellwiki/Package_versioning_policy) for -- standards guiding when and how versions should be incremented.-Version:             0.5.2+Version:             0.5.4  -- A short (one-line) description of the package. Synopsis:            Scoring functions commonly used for evaluation in NLP and IR