nonlinear-optimization-ad 0.2.0 → 0.2.1

raw patch · 4 files changed

+44/−15 lines, 4 filesdep +csvdep +nonlinear-optimization-addep ~addep ~basedep ~vectornew-component:exe:LinearRegression

Dependencies added: csv, nonlinear-optimization-ad

Dependency ranges changed: ad, base, vector

Files

• .travis.yml +3/−3
• nonlinear-optimization-ad.cabal +20/−1
• samples/LinearRegression.hs +19/−8
• src/Numeric/Optimization/Algorithms/HagerZhang05/AD.hs +2/−3

@@ -17,7 +17,7 @@ # - CABALVER=1.18 GHCVER=7.8.1  # see note about Alex/Happy for GHC >= 7.8 # - CABALVER=1.18 GHCVER=7.8.2  - CABALVER=1.18 GHCVER=7.8.3- - CABALVER=1.22 GHCVER=7.10.1+ - CABALVER=1.22 GHCVER=7.10.2 # - CABALVER=head GHCVER=head   # see section about GHC HEAD snapshots  # Note: the distinction between `before_install` and `install` is not important.@@ -31,12 +31,12 @@  - cabal --version  - echo "$(ghc --version) [$(ghc --print-project-git-commit-id 2> /dev/null || echo '?')]"  - travis_retry cabal update- - cabal install --only-dependencies --enable-tests --enable-benchmarks+ - cabal install --only-dependencies --enable-tests --enable-benchmarks -fBuildSamplePrograms  # Here starts the actual work to be performed for the package under test; any command which exits with a non-zero exit code causes the build to fail. script:  - if [ -f configure.ac ]; then autoreconf -i; fi- - cabal configure --enable-tests --enable-benchmarks -v2  # -v2 provides useful information for debugging+ - cabal configure --enable-tests --enable-benchmarks -fBuildSamplePrograms -v2  # -v2 provides useful information for debugging  - cabal build   # this builds all libraries and executables (including tests/benchmarks)  - cabal test  - cabal check

@@ -2,7 +2,7 @@ -- further documentation, see http://haskell.org/cabal/users-guide/  name:                nonlinear-optimization-ad-version:             0.2.0+version:             0.2.1 synopsis:            Wrapper of nonlinear-optimization package for using with AD package description:         Wrapper of nonlinear-optimization package for using with AD package homepage:            https://github.com/msakai/nonlinear-optimization-ad@@ -25,6 +25,11 @@   type:     git   location: git://github.com/msakai/nonlinear-optimization-ad.git +flag BuildSamplePrograms+  Description: build sample programs+  Default: False+  Manual: True+ library   exposed-modules:     Numeric.Optimization.Algorithms.HagerZhang05.AD   build-depends:@@ -42,3 +47,17 @@     TypeFamilies     CPP +-- Sample Programs++Executable LinearRegression+  If !flag(BuildSamplePrograms)+    Buildable: False+  Main-is: LinearRegression.hs+  HS-Source-Dirs: samples+  Build-Depends:+    base,+    csv,+    nonlinear-optimization-ad+  Default-Language: Haskell2010+  Other-extensions:+    TypeFamilies

@@ -1,26 +1,37 @@+{-# LANGUAGE GADTs #-}+{-# OPTIONS_GHC -Wall #-} module Main where -import Control.Monad-import qualified Data.Vector as V-import Numeric.AD import qualified Numeric.Optimization.Algorithms.HagerZhang05.AD as CG import Text.Printf import qualified Text.CSV as CSV  main :: IO () main = do-  Right csv <- CSV.parseCSVFromFile "galton.csv"+  Right csv <- CSV.parseCSVFromFile "samples/galton.csv"   let samples :: [(Double, Double)]       samples = [(read parent, read child) | [child,parent] <- tail csv]             -- hypothesis       h [theta0,theta1] x = theta0 + theta1 * x       -- cost function-      cost theta = mse [(auto x, auto y) | (x,y) <- samples] (h theta)+      -- cost :: (Fractional m, CG.Mode m, CG.Scalar m ~ Double) => [m] -> m]+      cost theta = mse [(CG.auto x, CG.auto y) | (x,y) <- samples] (h theta)       params   = CG.defaultParameters{ CG.printFinal = True, CG.printParams = True, CG.verbose = CG.Verbose }       grad_tol = 0.0000001-  (theta, result, stat) <- CG.optimize params grad_tol [0,0] cost-  print theta+  (theta@[theta0,theta1], _result, _stat) <- CG.optimize params grad_tol [0,0] cost+  printf "y = %f x + %f\n" theta1 theta0+  printf "MSE = %f\n" (mse samples (h theta))+  printf "R^2 = %f\n" (r2 samples (h theta))  -- mean squared error mse :: Fractional y => [(x,y)] -> (x -> y) -> y-mse samples h = sum [(h x - y)^(2::Int) | (x,y) <- samples] / fromIntegral (length samples)+mse samples h = mean [(h x - y)^(2::Int) | (x,y) <- samples]++r2 :: Fractional y => [(x,y)] -> (x -> y) -> y+r2 samples h = 1 - mse samples h / sum [(y - ym)^(2::Int) | y <- ys]+  where+    ys = map snd samples+    ym = mean ys++mean :: Fractional x => [x] -> x+mean xs = sum [x | x <- xs] / fromIntegral (length xs)

@@ -3,9 +3,6 @@ module Numeric.Optimization.Algorithms.HagerZhang05.AD   ( -- * Main function     optimize-    -- ** Kinds of function types-  , Simple-  , Mutable     -- * Result and statistics   , Result(..)   , Statistics(..)@@ -18,6 +15,8 @@   , EstimateError(..)     -- * Technical parameters   , TechParameters(..)+    -- * Re-export+  , Mode (..)   ) where  import Prelude hiding (mapM)