added simple script for estimating the coverage on the PennTreebank

treebanks/PennTreebank/coverage.hs

@@ -0,0 +1,45 @@
+import PGF
+import Data.Maybe
+
+main = do
+  ls <- fmap (filterExprs . lines) $ readFile "log4.txt"
+  let (c,m1,m2) = foldl counts (0,0,0) (map (\l -> fromMaybe (error l) (readExpr (f l))) ls)
+  print (length ls)
+  print ((c / (c+m1+m2))*100)
+  print (((c+m2) / (c+m1+m2))*100)
+  meta_dist [length [l | l <- ls, length [c | c <- l, c == '?'] == n] | n <- [0..27]]
+  meta_dist [length [l | l <- ls, length [x | x <- (zip l (tail l)), x == ('(','?')] == n] | n <- [0..27]]
+  cs <- fmap (map (length . words) . lines) $ readFile "wsj.eng"
+  print (average [fromIntegral c / fromIntegral (max n 1) | (c,l) <- zip cs ls, let n = length [c | c <- l, c == '?']])
+  print (average [fromIntegral c / fromIntegral (max n 1) | (c,l) <- zip cs ls, let n = length [x | x <- (zip l (tail l)), x == ('(','?')]])
+
+average xs = sum xs / fromIntegral (length xs)
+
+filterExprs []          = []
+filterExprs (l:ls)
+  | null l              = filterExprs ls
+  | elem (head l) "+#*" = drop 2 l : filterExprs ls
+  | otherwise           = filterExprs ls
+
+f []       = []
+f ('[':cs) = let (xs,']':ys) = break (==']') cs
+             in f ('?' : ys)
+f ('?':cs) = 'Q' : f cs
+f (c:cs)   = c   : f cs
+
+counts (c,m1,m2) e = c `seq` m1 `seq` m2 `seq`
+  case unApp e of
+    Just (f,es) | f == mkCId "Q" -> if null es
+                                      then foldl counts (c,m1,m2+1) es
+                                      else foldl counts (c,m1+1,m2) es
+                | otherwise      -> foldl counts (c+1,m1,m2) es
+    Nothing   -> case unStr e of
+                   Just _        -> (c+1,m1,m2)
+                   Nothing       -> error ("counts ("++show e++")")
+
+meta_dist cs = do
+  print cs
+  let cnt = fromIntegral (sum cs)
+      avg = fromIntegral (sum [n*c | (n,c) <- zip [0..] cs]) / cnt
+      dev = sqrt (sum [((fromIntegral n-avg) ^ 2)*fromIntegral c | (n,c) <- zip [0..] cs] / cnt)
+  print (avg,dev)