summaryrefslogtreecommitdiff
path: root/src/GF/Probabilistic/Probabilistic.hs
blob: 935175ed91dc0e8145233b4870b5eb7d8f860af4 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
----------------------------------------------------------------------
-- |
-- Module      : Probabilistic
-- Maintainer  : AR
-- Stability   : (stable)
-- Portability : (portable)
--
-- > CVS $Date: 2005/11/01 09:20:09 $ 
-- > CVS $Author: aarne $
-- > CVS $Revision: 1.5 $
--
-- Probabilistic abstract syntax. AR 30\/10\/2005
--
-- (c) Aarne Ranta 2005 under GNU GPL
--
-- Contents: parsing and random generation with probabilistic grammars.
-- To begin with, we use simple types and don't
-- guarantee the correctness of bindings\/dependences.
-----------------------------------------------------------------------------

module GF.Probabilistic.Probabilistic (
  generateRandomTreesProb -- :: Options -> StdGen -> GFCGrammar -> Probs -> Cat -> [Exp]
 ,checkGrammarProbs       -- :: GFCGrammar -> Probs -> Err ()
 ,computeProbTree         -- :: Probs -> Tree -> Double
 ,rankByScore             -- :: Ord n => [(a,n)] -> [(a,n)]
 ,Probs                   -- = BinTree Ident Double
 ,getProbsFromFile        -- :: Opts -> IO Probs
 ,emptyProbs              -- :: Probs
 ,prProbs                 -- :: Probs -> String
  ) where

import GF.Canon.GFC
import GF.Grammar.LookAbs
import GF.Grammar.PrGrammar
import GF.Grammar.Macros
import GF.Grammar.Values
import GF.Grammar.Grammar -- (Cat,EInt,K)

import GF.Infra.Ident
import GF.Data.Zipper
import GF.Data.Operations
import GF.Infra.Option

import Data.Char
import Data.List
import Control.Monad
import System.Random

-- | this parameter tells how many constructors at most are generated in a tree
timeout :: Int
timeout = 99

-- | generate an infinite list of trees, with their probabilities
generateRandomTreesProb :: Options -> StdGen -> GFCGrammar -> Probs -> Cat -> [Exp]
generateRandomTreesProb opts gen gr probs cat = 
  map str2tr $ randomTrees gen gr' cat' where
    gr'  = gr2sgr gr probs
    cat' = prt $ snd cat

-- | check that probabilities attached to a grammar make sense
checkGrammarProbs :: GFCGrammar -> Probs -> Err Probs
checkGrammarProbs gr probs = 
  err Bad (return . gr2probs) $ checkSGrammar $ gr2sgr gr probs where
    gr2probs sgr = buildTree [(zIdent f,p) | (_,rs) <- tree2list sgr, ((p,f),_) <- rs]

-- | compute the probability of a given tree
computeProbTree :: Probs -> Tree -> Double
computeProbTree probs (Tr (N (_,at,_,_,_),ts)) = case at of
  AtC (_,f) -> case lookupTree prt f probs of
    Ok p -> p * product (map prob ts)
    _ -> product (map prob ts)
  _ -> 1.0 ----
 where
   prob = computeProbTree probs

-- | rank from highest to lowest score, e.g. probability
rankByScore :: Ord n => [(a,n)] -> [(a,n)]
rankByScore = sortBy (\ (_,p) (_,q) -> compare q p)

getProbsFromFile :: Options -> FilePath -> IO Probs
getProbsFromFile opts file = do 
  s <- maybe (readFile file) readFile $ getOptVal opts probFile
  return $ buildTree $ concatMap pProb $ lines s
-- where
pProb s = case words s of
     "--#":"prob":f:p:_ | isDouble p -> [(zIdent f, read p)]
     f:ps@(g:rest) -> case span (/= "--#") ps of
       (_,_:"prob":p:_) | isDouble p -> [(zIdent f', readD p)] where 
         f' = if elem f ["fun","lin","data"] then ident g else ident f
       _ -> []
     _ -> []
  where
   isDouble = all (flip elem ('.':['0'..'9']))
   ident = takeWhile (flip notElem ".:")
   readD :: String -> Double
   readD = read

type Probs = BinTree Ident Double

emptyProbs :: Probs
emptyProbs = emptyBinTree

prProbs :: Probs -> String
prProbs = unlines . map pr . tree2list where
  pr (f,p) = prt f ++ "\t" ++ show p

------------------------------------------
-- translate grammar to simpler form and generated trees back

gr2sgr :: GFCGrammar -> Probs -> SGrammar
gr2sgr gr probs = buildTree [(c,fillProb rs) | rs@((_,(_,c)):_) <- rules] where
  rules =
    groupBy (\x y -> scat x == scat y) $
      sortBy (\x y -> compare (scat x) (scat y)) 
        [(trId f, ty') | (f,ty) <- funRulesOf gr, ty' <- trTy ty]
  trId (_,f) = let f' = prt f in case lookupTree prt f probs of
    Ok p -> (p,f')
    _ -> (2.0, f')
  trTy ty = case catSkeleton ty of
    Ok (mcs,mc) -> [(map trCat mcs, trCat mc)]
    _ -> []
  trCat (m,c) = prt c ---
  scat (_,(_,c)) = c

str2tr :: STree -> Exp
str2tr t = case t of
  SApp ((_,f),ts) -> mkApp (trId f) (map str2tr ts) 
  SMeta _     -> mkMeta 0
  SString s   -> K s
  SInt i      -> EInt i
  SFloat i    -> EFloat i
 where
   trId = cn . zIdent

type SGrammar = BinTree SCat [SRule]
type SIdent = String
type SRule = (SFun,SType)
type SType = ([SCat],SCat)
type SCat = SIdent
type SFun = (Double,SIdent)

allRules gr = concat [rs  | (c,rs) <- tree2list gr]

data STree = 
    SApp (SFun,[STree]) 
--  | SAppN (SIdent,[STree])  -- no probability given 
  | SMeta SCat
  | SString String
  | SInt Integer
  | SFloat Double
   deriving (Show,Eq)

probTree :: STree -> Double
probTree t = case t of
  SApp ((p,_),ts) -> p * product (map probTree ts)
  _ -> 1

rankTrees :: [STree] -> [(STree,Double)]
rankTrees ts = sortBy (\ (_,p) (_,q) -> compare q p) [(t,probTree t) | t <- ts]

randomTrees :: StdGen -> SGrammar -> SCat -> [STree]
randomTrees gen = genTrees (randomRs (0.0, 1.0) gen)

genTrees :: [Double] -> SGrammar -> SCat -> [STree]
genTrees ds0 gr cat = 
  let (ds,ds2) = splitAt (timeout+1) ds0  -- for time out, else ds
      (t,k) = genTree ds gr cat      
  in (if k>timeout then id else (t:))     -- don't accept with metas
           (genTrees ds2 gr cat)          -- else (drop k ds)

genTree :: [Double] -> SGrammar -> SCat -> (STree,Int)
genTree rs gr = gett rs where
  gett [] cat = (SMeta cat,1) -- time-out case
  gett ds "String" = (SString "foo",1)
  gett ds "Int" = (SInt 1978,1)
  gett ds "Float" = (SFloat 3.1415926, 1)
  gett ds cat = case look cat of
    [] -> (SMeta cat,1) -- if no productions, return ? 
    fs -> let 
        d:ds2     = ds
        (pf,args) = getf d fs
        (ts,k)    = getts ds2 args
      in (SApp (pf,ts), k+1)
  getf d fs = hitRegion d [(p,(pf,args)) | (pf@(p,_),(args,_)) <- fs]
  getts ds cats = case cats of
    c:cs -> let 
        (t, k)  = gett ds c
        (ts,ks) = getts (drop k ds) cs 
      in (t:ts, k + ks)
    _ -> ([],0)
  look cat = errVal [] $ lookupTree id cat gr

hitRegion :: Double -> [(Double,a)] -> a
hitRegion d vs = case vs of
  (p1,v1):vs2 -> 
    if d < p1 then v1 else hitRegion d [(p+p1,v) | (p,v) <- vs2]

--- this should recover from rounding errors

checkSGrammar :: SGrammar -> Err SGrammar
checkSGrammar = mapMTree chCat where
  chCat (c,rs) = case sum [p | ((p,f),_) <- rs] of
    s | abs (s - 1.0) > 0.01 -> 
      Bad $ "illegal probability sum " ++ show s ++ " in " ++ c
    _ -> return (c,rs)

-- for cases where explicit probability is not given (encoded as
-- p > 1) divide the remaining mass by the number of such cases

fillProb :: [SRule] -> [SRule]
fillProb rs = [((defa p,f),ty) | ((p,f),ty) <- rs] where
  defa p = if p > 1.0 then def else p
  def = (1 - sum given) / genericLength nope 
  (nope,given) = partition (> 1.0) [p | ((p,_),_) <- rs]


------------------------------------------
-- to test outside GF

prSTree t = case t of
  SApp ((p,f),ts) -> f ++ prParenth (show p) ++ concat (map pr1 ts)
  SMeta c -> '?':c
  SString s -> prQuotedString s
  SInt i -> show i 
  SFloat i -> show i 
 where
  pr1 t@(SApp (_,ts)) = ' ' : (if null ts then id else prParenth) (prSTree t)
  pr1 t = prSTree t


mkSGrammar :: [SRule] -> SGrammar 
mkSGrammar rules = 
  buildTree [(c, fillProb rs) | rs@((_,(_,c)):_) <- rules'] where
    rules' =     
      groupBy (\x y -> scat x == scat y) $
        sortBy (\x y -> compare (scat x) (scat y)) 
          rules 
    scat (_,(_,c)) = c

pSRule :: String -> SRule
pSRule s = case words s of
  p : f : c : cs -> 
    if isDigit (head p) 
      then ((read p, f),(init cs', last cs')) 
      else ((2.0, p),(init (c:cs'), last (c:cs'))) --- hack for automatic probability
     where cs' = [cs !! i | i <- [0,2..length cs - 1]]
  _ -> error $ "not a rule" +++ s

exSgr = mkSGrammar $ map pSRule [
  "0.8 a : A"
 ,"0.2 b : A"
 ,"0.2 n : A -> S -> S"
 ,"0.8 e : S"
 ]

ex1 :: IO ()
ex1 = do
  g <- newStdGen
  mapM_ (putStrLn . prSTree) $ randomTrees g exSgr "S"