Add bootstrap resampling

Handler/Evaluate.hs

@@ -17,12 +17,14 @@ import qualified Data.Map.Strict as M
 
 import GEval.Core
 import GEval.EvaluationScheme
-import GEval.Common (MetricValue)
 import GEval.OptionsParser
 import GEval.ParseParams (parseParamsFromFilePath, OutputFileParsed(..))
+import GEval.Common (GEvalException, MetricResult(..), MetricValue)
+import GEval.Formatting (formatTheResult)
 
 import Options.Applicative
 import Data.Conduit.SmartSource
+import Data.Conduit.Bootstrap (defaultConfidenceLevel, getConfidenceBounds)
 
 import System.FilePath (takeFileName, dropExtensions, (-<.>))
 
@@ -222,12 +224,14 @@ checkOrInsertEvaluation repoDir chan version out = do
         Right (Left _) -> do
           err chan "Cannot parse options, check the challenge repo"
         Right (Right (_, Just [(_, [result])])) -> do
-          msg chan $ concat [ "Evaluated! Score ", (formatNonScientifically result) ]
+          msg chan $ concat [ "Evaluated! Score ", (T.pack $ formatTheResult Nothing result) ]
           time <- liftIO getCurrentTime
-          _ <- runDB $ insert $ Evaluation {
+          _ <- runDB $ insert $ let (pointResult, errorBound) = extractResult result
+                               in Evaluation {
                                    evaluationTest=outTest out,
                                    evaluationChecksum=outChecksum out,
-            evaluationScore=Just result,
+                                   evaluationScore=Just pointResult,
+                                   evaluationErrorBound=errorBound,
                                    evaluationErrorMessage=Nothing,
                                    evaluationStamp=time,
                                    evaluationVersion=Just version }
@@ -239,12 +243,17 @@ checkOrInsertEvaluation repoDir chan version out = do
         Left exception -> do
           err chan $ "Evaluation failed: " ++ (T.pack $ show exception)
 
+extractResult :: MetricResult -> (MetricValue, Maybe MetricValue)
+extractResult (SimpleRun r) = (r, Nothing)
+extractResult (BootstrapResampling vals) = ((upperBound + lowerBound) / 2.0, Just ((upperBound - lowerBound) / 2.0))
+  where (lowerBound, upperBound) = getConfidenceBounds defaultConfidenceLevel vals
+
 rawEval :: FilePath
           -> EvaluationScheme
           -> FilePath
           -> Text
           -> FilePath
-          -> IO (Either GEvalException (Either (ParserResult GEvalOptions) (GEvalOptions, Maybe [(SourceSpec, [MetricValue])])))
+          -> IO (Either GEvalException (Either (ParserResult GEvalOptions) (GEvalOptions, Maybe [(SourceSpec, [MetricResult])])))
 rawEval challengeDir metric repoDir name outF = Import.try (runGEvalGetOptions [
                                                           "--alt-metric", (show metric),
                                                           "--expected-directory", challengeDir,

Handler/Query.hs

@@ -301,6 +301,7 @@ lineByLineTable (Entity testId test) theStamp = mempty
           evaluationTest = testId,
           evaluationChecksum = testChecksum test,
           evaluationScore = Just score,
+          evaluationErrorBound = Nothing,
           evaluationErrorMessage = Nothing,
           evaluationStamp = theStamp,
           evaluationVersion = Nothing }

Handler/Shared.hs

@@ -41,6 +41,7 @@ import Text.Regex.TDFA
 
 import GEval.Core
 import GEval.EvaluationScheme
+import GEval.Formatting (formatTheResultWithErrorBounds)
 
 import qualified Data.Vector as DV
 
@@ -449,7 +450,7 @@ formatTruncatedScore :: Maybe Int -> Maybe Evaluation -> Text
 formatTruncatedScore Nothing e = formatFullScore e
 formatTruncatedScore _ Nothing  = formatFullScore Nothing
 formatTruncatedScore (Just precision) (Just evaluation) = case evaluationScore evaluation of
-  Just score -> T.pack $ printf "%0.*f" precision score
+  Just score -> T.pack $ formatTheResultWithErrorBounds (Just precision) score (evaluationErrorBound evaluation)
   Nothing -> formatFullScore Nothing
 
 formatScore :: Maybe Int -> Double -> Text

config/models

@@ -122,6 +122,7 @@ Evaluation
     test TestId
     checksum SHA1
     score Double Maybe
+    errorBound Double Maybe
     errorMessage Text Maybe
     stamp UTCTime default=now()
     -- Should be just SHA1 (without Maybe) - Maybe is just a legacy

gonito.cabal

@@ -128,7 +128,7 @@ library
                  , filemanip
                  , cryptohash
                  , markdown
-                 , geval >= 1.27 && < 1.29
+                 , geval >= 1.31.1 && < 1.32
                  , filepath
                  , yesod-table
                  , regex-tdfa