Add hardening

src/GEval/MatchingSpecification.hs

@@ -31,9 +31,13 @@ singletons [d|data MatchingSpecification = ExactMatch -- ^ exact match, i.e. ide
                                                                           -- is matched and then proceed with some matching spec.
                                          | SmartMatch MatchingSpecification -- ^ do fuzzy matching only on values
                                                                             -- containing letters
+                                         | Harden MatchingSpecification -- ^ harden a soft match
                                          deriving (Eq)
              |]
 
+hardeningThreshold :: Double
+hardeningThreshold = 0.8
+
 getMatchingFunctionForString :: MatchingSpecification -> String -> String -> Double
 getMatchingFunctionForString ExactMatch got expected
   | got == expected = 1.0
@@ -51,6 +55,11 @@ getMatchingFunctionForString (SmartMatch smatchSpec) got expected = getMatchingF
                       then smatchSpec
                       else ExactMatch
 
+getMatchingFunctionForString (Harden smatchSpec) got expected = if softMatch >= hardeningThreshold
+                                                                then 1.0
+                                                                else 0.0
+  where softMatch = getMatchingFunctionForString smatchSpec got expected
+
 -- | Whether suitable for fuzzy matching when in the "smart" match mode.
 -- At the moment we check whether it contains at least one letter
 -- (we require the exact match for, for instance, numbers written with digits.

src/GEval/Metric.hs

@@ -84,6 +84,7 @@ instance Show Metric where
   show (MultiLabelFMeasure beta FuzzyMatch) = "Fuzzy/" ++ (show $ MultiLabelFMeasure beta ExactMatch)
   show (MultiLabelFMeasure beta (CutLabel matchSpec)) = "CutLabel/" ++ (show $ MultiLabelFMeasure beta matchSpec)
   show (MultiLabelFMeasure beta (SmartMatch matchSpec)) = "Smart/" ++ (show $ MultiLabelFMeasure beta matchSpec)
+  show (MultiLabelFMeasure beta (Harden matchSpec)) = "Harden/" ++ (show $ MultiLabelFMeasure beta matchSpec)
   show MultiLabelLogLoss = "MultiLabel-Logloss"
   show MultiLabelLikelihood = "MultiLabel-Likelihood"
   show (Mean metric) = "Mean/" ++ (show metric)
@@ -108,6 +109,9 @@ instance Read Metric where
   readsPrec p ('S':'m':'a':'r':'t':'/':theRest) = case readsPrec p theRest of
     [(metric, theRest)] -> [(applyMatchingSpecification SmartMatch metric, theRest)]
     _ -> []
+  readsPrec p ('H':'a':'r':'d':'e':'n':'/':theRest) = case readsPrec p theRest of
+    [(metric, theRest)] -> [(applyMatchingSpecification Harden metric, theRest)]
+    _ -> []
   readsPrec _ ('R':'M':'S':'E':theRest) = [(RMSE, theRest)]
   readsPrec _ ('M':'S':'E':theRest) = [(MSE, theRest)]
   readsPrec _ ('P':'e':'a':'r':'s':'o':'n':theRest) = [(Pearson, theRest)]

test/Spec.hs

@@ -351,6 +351,8 @@ main = hspec $ do
       runGEvalTest "multilabel-f1-ie-fuzzy" `shouldReturnAlmost` 0.681777777777
     it "information extraction with smart fuzzy matching" $ do
       runGEvalTest "multilabel-f1-ie-fuzzy-smart" `shouldReturnAlmost` 0.598444
+    it "information extraction with smart fuzzy matching hardened" $ do
+      runGEvalTest "multilabel-f1-ie-fuzzy-harden" `shouldReturnAlmost` 0.555555555
   describe "Mean/MultiLabel-F" $ do
     it "simple" $ do
       runGEvalTest "mean-multilabel-f1-simple" `shouldReturnAlmost` 0.5

test/multilabel-f1-ie-fuzzy-harden/multilabel-f1-ie-fuzzy-harden-solution/test-A/out.tsv

@@ -0,0 +1,3 @@
+important-person=JOHN_BROWN important-person=JOHN_SMITH company-name=Axaxaxaas_Mlo profit=12031
+company-name=Foo_Bar profit=1220
+company-name=Whatever important-person=PIERRE_MENARD

test/multilabel-f1-ie-fuzzy-harden/multilabel-f1-ie-fuzzy-harden/config.txt

@@ -0,0 +1 @@
+--metric Harden/CutLabel/Smart/Fuzzy/MultiLabel-F1:ls<_(inc|ltd)\.?(\s|$)><\2>

test/multilabel-f1-ie-fuzzy-harden/multilabel-f1-ie-fuzzy-harden/test-A/expected.tsv

@@ -0,0 +1,3 @@
+company-name=Axaxaxas_Mlö profit=12031 important-person=John_Smith important-person=James_Brown
+company-name=Orbis_Tertius profit=1020 important-person=Anna_Smith
+company-name=Whatever_Inc profit=5600 important-person=Pierre_Menard