Introduce existential features

This commit is contained in:
Filip Gralinski 2019-01-26 17:18:41 +01:00
parent dbf5c961af
commit ea5de5c719
2 changed files with 25 additions and 18 deletions

View File

@ -9,6 +9,7 @@ module GEval.FeatureExtractor
PeggedFactor(..),
Feature(..),
SimpleFactor(..),
ExistentialFactor(..),
AtomicFactor(..),
FeatureNamespace(..))
where
@ -41,15 +42,21 @@ data PeggedFactor = PeggedFactor FeatureNamespace SimpleFactor
instance Show PeggedFactor where
show (PeggedFactor namespace factor) = (show namespace) ++ ":" ++ (show factor)
data SimpleFactor = SimpleAtomicFactor AtomicFactor | BigramFactor AtomicFactor AtomicFactor | NumericalFactor (Maybe Double) Int
data SimpleFactor = SimpleExistentialFactor ExistentialFactor | NumericalFactor (Maybe Double) Int
deriving (Eq, Ord)
instance Show SimpleFactor where
show (SimpleAtomicFactor factor) = show factor
show (BigramFactor factorA factorB) = (show factorA) ++ "++" ++ (show factorB)
show (SimpleExistentialFactor factor) = show factor
show (NumericalFactor (Just v) _) = ("=" ++ (show v))
show (NumericalFactor (Nothing) l) = ("=#" ++ (show l))
data ExistentialFactor = SimpleAtomicFactor AtomicFactor | BigramFactor AtomicFactor AtomicFactor
deriving (Eq, Ord)
instance Show ExistentialFactor where
show (SimpleAtomicFactor factor) = show factor
show (BigramFactor factorA factorB) = (show factorA) ++ "++" ++ (show factorB)
data AtomicFactor = TextFactor Text | ShapeFactor WordShape
deriving (Eq, Ord)
@ -77,15 +84,15 @@ extractAtomicFactors mTokenizer bbdo t = [Data.List.map TextFactor tokens] ++
where tokens = nub $ (tokenizeForFactors mTokenizer) t
extractSimpleFactors :: (Maybe Tokenizer) -> BlackBoxDebuggingOptions -> Text -> [SimpleFactor]
extractSimpleFactors mTokenizer bbdo t = Data.List.concat $ (Prelude.map (Prelude.map SimpleAtomicFactor) atomss) ++
(if bbdoBigrams bbdo
then Prelude.map bigramFactors atomss
else [])
++
extractSimpleFactors mTokenizer bbdo t = Data.List.concat $ (Prelude.map (Prelude.map SimpleExistentialFactor) existentials) ++
(if bbdoConsiderNumericalFeatures bbdo
then [numericalFactor t]
else [])
where atomss = extractAtomicFactors mTokenizer bbdo t
existentials = (Prelude.map (Prelude.map SimpleAtomicFactor) atomss) ++
(if bbdoBigrams bbdo
then Prelude.map bigramFactors atomss
else [])
bigramFactors atoms = Prelude.map (\(a, b) -> BigramFactor a b) $ bigrams atoms
numericalFactor t = [NumericalFactor (readMaybe $ unpack t) (Data.Text.length t)]
extractFactors :: (Maybe Tokenizer) -> BlackBoxDebuggingOptions -> Text -> Text -> [PeggedFactor]

View File

@ -522,23 +522,23 @@ main = hspec $ do
bbdoConsiderNumericalFeatures = True }
(sort $ extractFactorsFromTabbed Nothing bbdo "in" "I like this\t34.3\ttests") `shouldBe` [
PeggedFactor (FeatureTabbedNamespace "in" 1)
(SimpleAtomicFactor (TextFactor "I")),
(SimpleExistentialFactor (SimpleAtomicFactor (TextFactor "I"))),
PeggedFactor (FeatureTabbedNamespace "in" 1)
(SimpleAtomicFactor (TextFactor "like")),
(SimpleExistentialFactor (SimpleAtomicFactor (TextFactor "like"))),
PeggedFactor (FeatureTabbedNamespace "in" 1)
(SimpleAtomicFactor (TextFactor "this")),
(SimpleExistentialFactor (SimpleAtomicFactor (TextFactor "this"))),
PeggedFactor (FeatureTabbedNamespace "in" 1)
(BigramFactor (TextFactor "I") (TextFactor "like")),
(SimpleExistentialFactor (BigramFactor (TextFactor "I") (TextFactor "like"))),
PeggedFactor (FeatureTabbedNamespace "in" 1)
(BigramFactor (TextFactor "like") (TextFactor "this")),
(SimpleExistentialFactor (BigramFactor (TextFactor "like") (TextFactor "this"))),
PeggedFactor (FeatureTabbedNamespace "in" 1)
(NumericalFactor Nothing 11),
PeggedFactor (FeatureTabbedNamespace "in" 2)
(SimpleAtomicFactor (TextFactor "34.3")),
(SimpleExistentialFactor (SimpleAtomicFactor (TextFactor "34.3"))),
PeggedFactor (FeatureTabbedNamespace "in" 2)
(NumericalFactor (Just 34.3) 4),
PeggedFactor (FeatureTabbedNamespace "in" 3)
(SimpleAtomicFactor (TextFactor "tests")),
(SimpleExistentialFactor (SimpleAtomicFactor (TextFactor "tests"))),
PeggedFactor (FeatureTabbedNamespace "in" 3)
(NumericalFactor Nothing 5) ]