filipg/geval
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Clip loess and switch to gaussian in Calibration

Browse Source
This commit is contained in:
Filip Gralinski 2019-03-19 20:41:06 +01:00
parent de40851b5a
commit e5220d71d8
4 changed files with 18 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/Data/Statistics/Calibration.hs
View File

@ -1,7 +1,7 @@
module Data.Statistics.Calibration
(calibration, softCalibration) where
import Data.Statistics.Loess(loess)
import Data.Statistics.Loess (clippedLoess)
import Numeric.Integration.TanhSinh
import Data.List (minimum, maximum)
import qualified Data.Vector.Unboxed as DVU
@ -34,7 +34,7 @@ softCalibration [] _ = error "too few booleans in calibration"
softCalibration _ [] = error "too few probabilities in calibration"
softCalibration results probs
| band probs < minBand = handleNarrowBand results probs
| otherwise = 1.0 - (min 1.0 (2.0 * (integrate (lowest, highest) (\x -> abs ((loess (DVU.fromList probs) (DVU.fromList results) x) - x))) / (highest - lowest)))
| otherwise = 1.0 - (min 1.0 (2.0 * (integrate (lowest, highest) (\x -> abs ((clippedLoess (DVU.fromList probs) (DVU.fromList results) x) - x))) / (highest - lowest)))
where lowest = (minimum probs) + epsilon -- integrating loess gets crazy at edges
highest = (maximum probs) - epsilon
epsilon = 0.0001

15
src/Data/Statistics/Loess.hs
View File

@ -1,23 +1,32 @@
module Data.Statistics.Loess
(loess) where
(loess, clippedLoess) where
import qualified Statistics.Matrix.Types as SMT
import Statistics.Regression (ols)
import Data.Vector.Unboxed((!), zipWith, length, (++), map)
import Statistics.Matrix(transpose)
import Statistics.Distribution.Normal (standard)
import Statistics.Distribution (density)
lambda :: Double
lambda = 2.0
lambda = 8.0
triCube :: Double -> Double
triCube d = (1.0 - (abs d) ** 3) ** 3
gaussian :: Double -> Double
gaussian = density standard
clippedLoess :: SMT.Vector -> SMT.Vector -> Double -> Double
clippedLoess inputs outputs x = min 1.0 $ max 0.0 $ loess inputs outputs x
loess :: SMT.Vector -> SMT.Vector -> Double -> Double
loess inputs outputs x = a * x + b
where a = params ! 1
b = params ! 0
params = ols inputMatrix scaledOutputs
weights = Data.Vector.Unboxed.map (\v -> lambda * triCube (lambda * (x - v))) inputs
weights = Data.Vector.Unboxed.map (\v -> lambda * gaussian (lambda * (x - v))) inputs
scaledOutputs = Data.Vector.Unboxed.zipWith (*) outputs weights
scaledInputs = Data.Vector.Unboxed.zipWith (*) inputs weights
inputMatrix = transpose (SMT.Matrix 2 (Data.Vector.Unboxed.length inputs) 1000 (weights Data.Vector.Unboxed.++ scaledInputs))

4
src/GEval/Core.hs
View File

@ -103,7 +103,7 @@ import qualified Data.Vector.Unboxed as DVU
import Statistics.Correlation
import Data.Statistics.Calibration (softCalibration)
import Data.Statistics.Loess(loess)
import Data.Statistics.Loess (clippedLoess)
import Data.Proxy
@ -755,7 +755,7 @@ gevalCore' (ProbabilisticSoftFMeasure beta) _ = gevalCoreWithoutInput parseAnnot
probabilisticSoftAgg = CC.foldl probabilisticSoftFolder ([], [], fromInteger 0, 0)
probabilisticSoftFolder (r1, p1, g1, e1) (r2, p2, g2, e2) = (r1 ++ r2, p1 ++ p2, g1 + g2, e1 + e2)
loessGraph :: ([Double], [Double], Double, Int) -> Maybe GraphSeries
loessGraph (results, probs, _, _) = Just $ GraphSeries $ Prelude.map (\x -> (x, loess probs' results' x)) $ Prelude.filter (\p -> p > lowest && p < highest) $ Prelude.map (\d -> 0.01 * (fromIntegral d)) [1..99]
loessGraph (results, probs, _, _) = Just $ GraphSeries $ Prelude.map (\x -> (x, clippedLoess probs' results' x)) $ Prelude.filter (\p -> p > lowest && p < highest) $ Prelude.map (\d -> 0.01 * (fromIntegral d)) [1..99]
where results' = DVU.fromList results
probs' = DVU.fromList probs
lowest = Data.List.minimum probs

2
src/GEval/OptionsParser.hs
View File

@ -9,8 +9,6 @@ module GEval.OptionsParser
precisionArgParser
) where
import Debug.Trace
import Paths_geval (version)
import Data.Version (showVersion)