-----------------------------------------------------------------------------
-- |
-- Module      :  DSP.Matrix.Simplex
-- Copyright   :  (c) Matthew Donadio 2003
-- License     :  GPL
--
-- Maintainer  :  m.p.donadio@ieee.org
-- Stability   :  experimental
-- Portability :  portable
--
-- Two-step simplex algorithm
--
-- I only guarantee that this module wastes inodes
--
-----------------------------------------------------------------------------

-- Originally based off the code in Sedgewick, but modified to match the
-- conventions from Papadimitriou and Steiglitz.

-- TODO: Is our column/row selection the same as Bland's anti-cycle
-- algorithm?

-- TODO: Add check for redundant rows in two-phase algorithm

-- TODO: Lots of testing

module Matrix.Simplex (Simplex(..), simplex, twophase) where

import Data.Array

eps :: Double
eps = 1.0e-10

-------------------------------------------------------------------------------

-- Pivot around a!(p,q)

pivot :: Int -> Int -> Array (Int,Int) Double -> Array (Int,Int) Double
pivot p q a0 = step4 $ step3 $ step2 $ step1 $ a0
    where step1 a = a // [ ((j,k), a!(j,k) - a!(p,k) * a!(j,q) / a!(p,q)) | k <- [0..m], j <- [ph..n], j /= p && k /= q ]
          step2 a = a // [ ((j,q),0) | j <- [ph..n], j /= p ]
          step3 a = a // [ ((p,k), a!(p,k) / a!(p,q)) | k <- [0..m], k /= q ]
          step4 a = a // [ ((p,q),1) ]
          ((ph,_),(n,m)) = bounds a0

-- chooseq picks the lowest numbered favorable column.  If there are no
-- favorable columns, then q==m is returned, and we have reached an
-- optimum.


chooseq :: (Ord b, Num b, Ix a, Ix b, Num a) =>
           Array (a, b) Double -> b
chooseq a0 = chooseq' 1 a0
    where chooseq' q a | q > m          = q
                       | a!(0,q) < -eps = q
                       | otherwise      = chooseq' (q+1) a
          ((_,_),(_,m)) = bounds a0

-- choosep picks a row with a positive element in column q.  If no such
-- element exists, then the p==n is returned, and the problem is
-- unfeasible.

choosep :: (Ord a, Num a, Ix a, Ix b) =>
           b -> Array (a, b) Double -> a
choosep q a0 = choosep' 1 a0
    where choosep' p a | p > n         = p
                       | a!(p,q) > eps = p
                       | otherwise     = choosep' (p+1) a
          ((_,_),(n,_)) = bounds a0

-- refinep picks the row using the ratio test.

refinep :: (Ord a, Num a, Num b, Ix a, Ix b) =>
           a -> b -> Array (a, b) Double -> a
refinep p0 q a0 = refinep' (p0+1) p0 a0
    where refinep' i p a | i > n = p
                         | a!(i,q) > eps && a!(i,0) / a!(i,q) < a!(p,0) / a!(p,q) = refinep' (i+1) i a
                         | otherwise = refinep' (i+1) p a
          ((_,_),(n,_)) = bounds a0

-- * Types

-- | Type for results of the simplex algorithm

data Simplex a = Unbounded | Infeasible | Optimal a deriving (Read,Show)

-- * Functions

-- | The simplex algorithm for standard form:
--
-- min   c'x
--
-- where Ax = b, x >= 0
--
-- a!(0,0) = -z
--
-- a!(0,j) = c'
--
-- a!(i,0) = b
--
-- a!(i,j) = A_ij

simplex :: Array (Int,Int) Double -- ^ stating tableau
        -> Simplex (Array (Int,Int) Double) -- ^ solution

simplex a | q > m      = Optimal a
          | p > n      = Unbounded
          | otherwise  = simplex $ pivot p' q $ a
    where q = chooseq a
          p = choosep q a
          p' = refinep p q a
          ((_,_),(n,m)) = bounds a

-------------------------------------------------------------------------------

addart :: (Num e, Enum a, Ix a, Num a) =>
          Array (a, a) e -> Array (a, a) e
addart a = array ((-1,0),(n,m+n)) $ z ++ xsi ++ b ++ art ++ x
    where z = ((-1,0), a!(0,0)) : [ ((-1,j),0) | j <- [1..n] ] ++ [ ((-1,j+n),a!(0,j)) | j <- [1..m] ]
          xsi = ((0,0), -colsum a 0) : [ ((0,j),0) | j <- [1..n] ] ++ [ ((0,j+n), -colsum a j) | j <- [1..m] ]
          b = [ ((i,0), a!(i,0)) | i <- [1..n] ]
          art = [ ((i,j), if i == j then 1 else 0) | i <- [1..n], j <- [1..n] ]
          x = [ ((i,j+n), a!(i,j)) | i <- [1..n], j <- [1..m] ]
          ((_,_),(n,m)) = bounds a

colsum :: (Num e, Num a, Enum a, Ix a, Ix b) =>
          Array (a, b) e -> b -> e
colsum a j = sum [ a!(i,j) | i <- [1..n] ]
    where ((_,_),(n,_)) = bounds a

delart :: (Enum a, Ix a, Num a) =>
          Array (a, a) e -> Array (a, a) e -> Array (a, a) e
delart a a' = array ((0,0),(n,m)) $ z ++ b ++ x
    where z = ((0,0), a'!(-1,0)) : [ ((0,j), a!(0,j)) | j <- [1..m] ]
          b = [ ((i,0), a'!(i,0)) | i <- [1..n] ]
          x = [ ((i,j), a'!(i,j+n)) | i <- [1..n], j <- [1..m] ]
          ((_,_),(n,m)) = bounds a

-- | The two-phase simplex algorithm

twophase :: Array (Int,Int) Double -- ^ stating tableau
         -> Simplex (Array (Int,Int) Double) -- ^ solution

twophase a | cost a' > eps = Infeasible
           | otherwise     = simplex $ delart a (gettab a')
    where a' = simplex $ addart $ a


-- How to handle cases where 'simplex' does not return Optimal?
gettab :: Simplex a -> a
gettab (Optimal a) = a

cost :: (Num e, Ix a, Ix b, Num a, Num b) =>
        Simplex (Array (a, b) e) -> e
cost (Optimal a) = negate $ a!(0,0)

-------------------------------------------------------------------------------

{-

Test vectors

This is from Sedgewick

> x1 = listArray ((0,0),(5,8)) [  0, -1, -1, -1, 0, 0, 0, 0, 0,
>                                 5, -1,  1,  0, 1, 0, 0, 0, 0,
>                                45,  1,  4,  0, 0, 1, 0, 0, 0,
>                                27,  2,  1,  0, 0, 0, 1, 0, 0,
>                                24,  3, -4,  0, 0, 0, 0, 1, 0,
>                                 4,  0,  0,  1, 0, 0, 0, 0, 1 ] :: Array (Int,Int) Double

P&S, Example 2.6

> x2 = listArray ((0,0),(3,5)) [ 0, 1, 1, 1, 1, 1,
>                                1, 3, 2, 1, 0, 0,
>                                3, 5, 1, 1, 1, 0,
>                                4, 2, 5, 1, 0, 1 ] :: Array (Int,Int) Double

P&S, Example 2.6 (after BFS selection)

> x2' = listArray ((0,0),(3,5)) [ -6, -3, -3,  0,  0,  0,
>                                 1,  3,  2,  1,  0,  0,
>                                 2,  2, -1,  0,  1,  0,
>                                 3, -1,  3,  0,  0,  1 ] :: Array (Int,Int) Double

P&S, Example 2.2 / Section 2.9

> x3 = listArray ((0,0),(4,7)) [ -34, -1, -14, -6, 0, 0, 0, 0,
>                                  4,  1,   1,  1, 1, 0, 0, 0,
>                                  2,  1,   0,  0, 0, 1, 0, 0,
>                                  3,  0,   0,  1, 0, 0, 1, 0,
>                                  6,  0,   3,  1, 0, 0, 0, 1 ] :: Array (Int,Int) Double

P&S, Example 2.7

> x4 = listArray ((0,0),(3,7)) [ 3, -3/4,  20, -1/2, 6, 0, 0, 0,
>                                0,  1/4,  -8,   -1, 9, 1, 0, 0,
>                                0,  1/2, -12, -1/2, 3, 0, 1, 0,
>                                1,    0,   0,    1, 0, 0, 0, 1 ] :: Array (Int,Int) Double

These come in handy for testing

> row j a = listArray (0,m) [ a!(j,k) | k <- [0..m] ]
>    where ((_,_),(n,m)) = bounds a

> column k a = listArray (0,n) [ a!(j,k) | j <- [0..n] ]
>    where ((_,_),(n,m)) = bounds a

> solution (Optimal a) = listArray (1,m) $ [ find a j | j <- [1..m] ]
>    where ((_,_),(n,m)) = bounds a

> find a j = findone' a 1 j
>     where findone' a i j | i > n          = 0
>                          | a!(i,j) == 1.0 = b!i
>                          | otherwise      = findone' a (i+1) j
>           b = listArray (1,n) [ a!(i,0) | i <- [1..n] ]
>           ((_,_),(n,m)) = bounds a

-}