ad-4.3.4: Automatic Differentiation

Copyright(c) Edward Kmett 2010-2015
LicenseBSD3
Maintainerekmett@gmail.com
Stabilityexperimental
PortabilityGHC only
Safe HaskellNone
LanguageHaskell2010

Numeric.AD.Mode.Kahn

Contents

Description

This module provides reverse-mode Automatic Differentiation using post-hoc linear time topological sorting.

For reverse mode AD we use StableName to recover sharing information from the tape to avoid combinatorial explosion, and thus run asymptotically faster than it could without such sharing information, but the use of side-effects contained herein is benign.

Synopsis

Documentation

data AD s a Source #

Instances

Bounded a => Bounded (AD s a) Source # 

Methods

minBound :: AD s a #

maxBound :: AD s a #

Enum a => Enum (AD s a) Source # 

Methods

succ :: AD s a -> AD s a #

pred :: AD s a -> AD s a #

toEnum :: Int -> AD s a #

fromEnum :: AD s a -> Int #

enumFrom :: AD s a -> [AD s a] #

enumFromThen :: AD s a -> AD s a -> [AD s a] #

enumFromTo :: AD s a -> AD s a -> [AD s a] #

enumFromThenTo :: AD s a -> AD s a -> AD s a -> [AD s a] #

Eq a => Eq (AD s a) Source # 

Methods

(==) :: AD s a -> AD s a -> Bool #

(/=) :: AD s a -> AD s a -> Bool #

Floating a => Floating (AD s a) Source # 

Methods

pi :: AD s a #

exp :: AD s a -> AD s a #

log :: AD s a -> AD s a #

sqrt :: AD s a -> AD s a #

(**) :: AD s a -> AD s a -> AD s a #

logBase :: AD s a -> AD s a -> AD s a #

sin :: AD s a -> AD s a #

cos :: AD s a -> AD s a #

tan :: AD s a -> AD s a #

asin :: AD s a -> AD s a #

acos :: AD s a -> AD s a #

atan :: AD s a -> AD s a #

sinh :: AD s a -> AD s a #

cosh :: AD s a -> AD s a #

tanh :: AD s a -> AD s a #

asinh :: AD s a -> AD s a #

acosh :: AD s a -> AD s a #

atanh :: AD s a -> AD s a #

log1p :: AD s a -> AD s a #

expm1 :: AD s a -> AD s a #

log1pexp :: AD s a -> AD s a #

log1mexp :: AD s a -> AD s a #

Fractional a => Fractional (AD s a) Source # 

Methods

(/) :: AD s a -> AD s a -> AD s a #

recip :: AD s a -> AD s a #

fromRational :: Rational -> AD s a #

Num a => Num (AD s a) Source # 

Methods

(+) :: AD s a -> AD s a -> AD s a #

(-) :: AD s a -> AD s a -> AD s a #

(*) :: AD s a -> AD s a -> AD s a #

negate :: AD s a -> AD s a #

abs :: AD s a -> AD s a #

signum :: AD s a -> AD s a #

fromInteger :: Integer -> AD s a #

Ord a => Ord (AD s a) Source # 

Methods

compare :: AD s a -> AD s a -> Ordering #

(<) :: AD s a -> AD s a -> Bool #

(<=) :: AD s a -> AD s a -> Bool #

(>) :: AD s a -> AD s a -> Bool #

(>=) :: AD s a -> AD s a -> Bool #

max :: AD s a -> AD s a -> AD s a #

min :: AD s a -> AD s a -> AD s a #

Read a => Read (AD s a) Source # 

Methods

readsPrec :: Int -> ReadS (AD s a) #

readList :: ReadS [AD s a] #

readPrec :: ReadPrec (AD s a) #

readListPrec :: ReadPrec [AD s a] #

Real a => Real (AD s a) Source # 

Methods

toRational :: AD s a -> Rational #

RealFloat a => RealFloat (AD s a) Source # 

Methods

floatRadix :: AD s a -> Integer #

floatDigits :: AD s a -> Int #

floatRange :: AD s a -> (Int, Int) #

decodeFloat :: AD s a -> (Integer, Int) #

encodeFloat :: Integer -> Int -> AD s a #

exponent :: AD s a -> Int #

significand :: AD s a -> AD s a #

scaleFloat :: Int -> AD s a -> AD s a #

isNaN :: AD s a -> Bool #

isInfinite :: AD s a -> Bool #

isDenormalized :: AD s a -> Bool #

isNegativeZero :: AD s a -> Bool #

isIEEE :: AD s a -> Bool #

atan2 :: AD s a -> AD s a -> AD s a #

RealFrac a => RealFrac (AD s a) Source # 

Methods

properFraction :: Integral b => AD s a -> (b, AD s a) #

truncate :: Integral b => AD s a -> b #

round :: Integral b => AD s a -> b #

ceiling :: Integral b => AD s a -> b #

floor :: Integral b => AD s a -> b #

Show a => Show (AD s a) Source # 

Methods

showsPrec :: Int -> AD s a -> ShowS #

show :: AD s a -> String #

showList :: [AD s a] -> ShowS #

Erf a => Erf (AD s a) Source # 

Methods

erf :: AD s a -> AD s a #

erfc :: AD s a -> AD s a #

erfcx :: AD s a -> AD s a #

normcdf :: AD s a -> AD s a #

InvErf a => InvErf (AD s a) Source # 

Methods

inverf :: AD s a -> AD s a #

inverfc :: AD s a -> AD s a #

invnormcdf :: AD s a -> AD s a #

Mode a => Mode (AD s a) Source # 

Associated Types

type Scalar (AD s a) :: * Source #

Methods

isKnownConstant :: AD s a -> Bool Source #

isKnownZero :: AD s a -> Bool Source #

auto :: Scalar (AD s a) -> AD s a Source #

(*^) :: Scalar (AD s a) -> AD s a -> AD s a Source #

(^*) :: AD s a -> Scalar (AD s a) -> AD s a Source #

(^/) :: AD s a -> Scalar (AD s a) -> AD s a Source #

zero :: AD s a Source #

type Scalar (AD s a) Source # 
type Scalar (AD s a) = Scalar a

data Kahn a Source #

Kahn is a Mode using reverse-mode automatic differentiation that provides fast diffFU, diff2FU, grad, grad2 and a fast jacobian when you have a significantly smaller number of outputs than inputs.

Instances

(Num a, Bounded a) => Bounded (Kahn a) # 

Methods

minBound :: Kahn a #

maxBound :: Kahn a #

(Num a, Enum a) => Enum (Kahn a) # 

Methods

succ :: Kahn a -> Kahn a #

pred :: Kahn a -> Kahn a #

toEnum :: Int -> Kahn a #

fromEnum :: Kahn a -> Int #

enumFrom :: Kahn a -> [Kahn a] #

enumFromThen :: Kahn a -> Kahn a -> [Kahn a] #

enumFromTo :: Kahn a -> Kahn a -> [Kahn a] #

enumFromThenTo :: Kahn a -> Kahn a -> Kahn a -> [Kahn a] #

(Num a, Eq a) => Eq (Kahn a) # 

Methods

(==) :: Kahn a -> Kahn a -> Bool #

(/=) :: Kahn a -> Kahn a -> Bool #

Floating a => Floating (Kahn a) # 

Methods

pi :: Kahn a #

exp :: Kahn a -> Kahn a #

log :: Kahn a -> Kahn a #

sqrt :: Kahn a -> Kahn a #

(**) :: Kahn a -> Kahn a -> Kahn a #

logBase :: Kahn a -> Kahn a -> Kahn a #

sin :: Kahn a -> Kahn a #

cos :: Kahn a -> Kahn a #

tan :: Kahn a -> Kahn a #

asin :: Kahn a -> Kahn a #

acos :: Kahn a -> Kahn a #

atan :: Kahn a -> Kahn a #

sinh :: Kahn a -> Kahn a #

cosh :: Kahn a -> Kahn a #

tanh :: Kahn a -> Kahn a #

asinh :: Kahn a -> Kahn a #

acosh :: Kahn a -> Kahn a #

atanh :: Kahn a -> Kahn a #

log1p :: Kahn a -> Kahn a #

expm1 :: Kahn a -> Kahn a #

log1pexp :: Kahn a -> Kahn a #

log1mexp :: Kahn a -> Kahn a #

Fractional a => Fractional (Kahn a) # 

Methods

(/) :: Kahn a -> Kahn a -> Kahn a #

recip :: Kahn a -> Kahn a #

fromRational :: Rational -> Kahn a #

Num a => Num (Kahn a) # 

Methods

(+) :: Kahn a -> Kahn a -> Kahn a #

(-) :: Kahn a -> Kahn a -> Kahn a #

(*) :: Kahn a -> Kahn a -> Kahn a #

negate :: Kahn a -> Kahn a #

abs :: Kahn a -> Kahn a #

signum :: Kahn a -> Kahn a #

fromInteger :: Integer -> Kahn a #

(Num a, Ord a) => Ord (Kahn a) # 

Methods

compare :: Kahn a -> Kahn a -> Ordering #

(<) :: Kahn a -> Kahn a -> Bool #

(<=) :: Kahn a -> Kahn a -> Bool #

(>) :: Kahn a -> Kahn a -> Bool #

(>=) :: Kahn a -> Kahn a -> Bool #

max :: Kahn a -> Kahn a -> Kahn a #

min :: Kahn a -> Kahn a -> Kahn a #

Real a => Real (Kahn a) # 

Methods

toRational :: Kahn a -> Rational #

RealFloat a => RealFloat (Kahn a) # 
RealFrac a => RealFrac (Kahn a) # 

Methods

properFraction :: Integral b => Kahn a -> (b, Kahn a) #

truncate :: Integral b => Kahn a -> b #

round :: Integral b => Kahn a -> b #

ceiling :: Integral b => Kahn a -> b #

floor :: Integral b => Kahn a -> b #

Show a => Show (Kahn a) Source # 

Methods

showsPrec :: Int -> Kahn a -> ShowS #

show :: Kahn a -> String #

showList :: [Kahn a] -> ShowS #

MuRef (Kahn a) Source # 

Associated Types

type DeRef (Kahn a) :: * -> * #

Methods

mapDeRef :: Applicative f => (forall b. (MuRef b, ((* -> *) ~ DeRef (Kahn a)) (DeRef b)) => b -> f u) -> Kahn a -> f (DeRef (Kahn a) u) #

Erf a => Erf (Kahn a) # 

Methods

erf :: Kahn a -> Kahn a #

erfc :: Kahn a -> Kahn a #

erfcx :: Kahn a -> Kahn a #

normcdf :: Kahn a -> Kahn a #

InvErf a => InvErf (Kahn a) # 

Methods

inverf :: Kahn a -> Kahn a #

inverfc :: Kahn a -> Kahn a #

invnormcdf :: Kahn a -> Kahn a #

Num a => Mode (Kahn a) Source # 

Associated Types

type Scalar (Kahn a) :: * Source #

Methods

isKnownConstant :: Kahn a -> Bool Source #

isKnownZero :: Kahn a -> Bool Source #

auto :: Scalar (Kahn a) -> Kahn a Source #

(*^) :: Scalar (Kahn a) -> Kahn a -> Kahn a Source #

(^*) :: Kahn a -> Scalar (Kahn a) -> Kahn a Source #

(^/) :: Kahn a -> Scalar (Kahn a) -> Kahn a Source #

zero :: Kahn a Source #

Num a => Jacobian (Kahn a) Source # 

Associated Types

type D (Kahn a) :: * Source #

Methods

unary :: (Scalar (Kahn a) -> Scalar (Kahn a)) -> D (Kahn a) -> Kahn a -> Kahn a Source #

lift1 :: (Scalar (Kahn a) -> Scalar (Kahn a)) -> (D (Kahn a) -> D (Kahn a)) -> Kahn a -> Kahn a Source #

lift1_ :: (Scalar (Kahn a) -> Scalar (Kahn a)) -> (D (Kahn a) -> D (Kahn a) -> D (Kahn a)) -> Kahn a -> Kahn a Source #

binary :: (Scalar (Kahn a) -> Scalar (Kahn a) -> Scalar (Kahn a)) -> D (Kahn a) -> D (Kahn a) -> Kahn a -> Kahn a -> Kahn a Source #

lift2 :: (Scalar (Kahn a) -> Scalar (Kahn a) -> Scalar (Kahn a)) -> (D (Kahn a) -> D (Kahn a) -> (D (Kahn a), D (Kahn a))) -> Kahn a -> Kahn a -> Kahn a Source #

lift2_ :: (Scalar (Kahn a) -> Scalar (Kahn a) -> Scalar (Kahn a)) -> (D (Kahn a) -> D (Kahn a) -> D (Kahn a) -> (D (Kahn a), D (Kahn a))) -> Kahn a -> Kahn a -> Kahn a Source #

Num a => Grad (Kahn a) [a] (a, [a]) a Source # 

Methods

pack :: Kahn a -> [Kahn a] -> Kahn a Source #

unpack :: ([a] -> [a]) -> [a] Source #

unpack' :: ([a] -> (a, [a])) -> (a, [a]) Source #

Grad i o o' a => Grad (Kahn a -> i) (a -> o) (a -> o') a Source # 

Methods

pack :: (Kahn a -> i) -> [Kahn a] -> Kahn a Source #

unpack :: ([a] -> [a]) -> a -> o Source #

unpack' :: ([a] -> (a, [a])) -> a -> o' Source #

type DeRef (Kahn a) Source # 
type DeRef (Kahn a) = Tape a
type Scalar (Kahn a) Source # 
type Scalar (Kahn a) = a
type D (Kahn a) Source # 
type D (Kahn a) = Id a

auto :: Mode t => Scalar t -> t Source #

Embed a constant

Gradient

grad :: (Traversable f, Num a) => (forall s. f (AD s (Kahn a)) -> AD s (Kahn a)) -> f a -> f a Source #

The grad function calculates the gradient of a non-scalar-to-scalar function with kahn-mode AD in a single pass.

>>> grad (\[x,y,z] -> x*y+z) [1,2,3]
[2,1,1]

grad' :: (Traversable f, Num a) => (forall s. f (AD s (Kahn a)) -> AD s (Kahn a)) -> f a -> (a, f a) Source #

The grad' function calculates the result and gradient of a non-scalar-to-scalar function with kahn-mode AD in a single pass.

>>> grad' (\[x,y,z] -> 4*x*exp y+cos z) [1,2,3]
(28.566231899122155,[29.5562243957226,29.5562243957226,-0.1411200080598672])

gradWith :: (Traversable f, Num a) => (a -> a -> b) -> (forall s. f (AD s (Kahn a)) -> AD s (Kahn a)) -> f a -> f b Source #

grad g f function calculates the gradient of a non-scalar-to-scalar function f with kahn-mode AD in a single pass. The gradient is combined element-wise with the argument using the function g.

grad = gradWith (_ dx -> dx)
id = gradWith const

gradWith' :: (Traversable f, Num a) => (a -> a -> b) -> (forall s. f (AD s (Kahn a)) -> AD s (Kahn a)) -> f a -> (a, f b) Source #

grad' g f calculates the result and gradient of a non-scalar-to-scalar function f with kahn-mode AD in a single pass the gradient is combined element-wise with the argument using the function g.

grad' == gradWith' (_ dx -> dx)

Jacobian

jacobian :: (Traversable f, Functor g, Num a) => (forall s. f (AD s (Kahn a)) -> g (AD s (Kahn a))) -> f a -> g (f a) Source #

The jacobian function calculates the jacobian of a non-scalar-to-non-scalar function with kahn AD lazily in m passes for m outputs.

>>> jacobian (\[x,y] -> [y,x,x*y]) [2,1]
[[0,1],[1,0],[1,2]]
>>> jacobian (\[x,y] -> [exp y,cos x,x+y]) [1,2]
[[0.0,7.38905609893065],[-0.8414709848078965,0.0],[1.0,1.0]]

jacobian' :: (Traversable f, Functor g, Num a) => (forall s. f (AD s (Kahn a)) -> g (AD s (Kahn a))) -> f a -> g (a, f a) Source #

The jacobian' function calculates both the result and the Jacobian of a nonscalar-to-nonscalar function, using m invocations of kahn AD, where m is the output dimensionality. Applying fmap snd to the result will recover the result of jacobian | An alias for gradF'

ghci> jacobian' ([x,y] -> [y,x,x*y]) [2,1] [(1,[0,1]),(2,[1,0]),(2,[1,2])]

jacobianWith :: (Traversable f, Functor g, Num a) => (a -> a -> b) -> (forall s. f (AD s (Kahn a)) -> g (AD s (Kahn a))) -> f a -> g (f b) Source #

'jacobianWith g f' calculates the Jacobian of a non-scalar-to-non-scalar function f with kahn AD lazily in m passes for m outputs.

Instead of returning the Jacobian matrix, the elements of the matrix are combined with the input using the g.

jacobian = jacobianWith (_ dx -> dx)
jacobianWith const = (f x -> const x <$> f x)

jacobianWith' :: (Traversable f, Functor g, Num a) => (a -> a -> b) -> (forall s. f (AD s (Kahn a)) -> g (AD s (Kahn a))) -> f a -> g (a, f b) Source #

jacobianWith g f' calculates both the result and the Jacobian of a nonscalar-to-nonscalar function f, using m invocations of kahn AD, where m is the output dimensionality. Applying fmap snd to the result will recover the result of jacobianWith

Instead of returning the Jacobian matrix, the elements of the matrix are combined with the input using the g.

jacobian' == jacobianWith' (_ dx -> dx)

Hessian

hessian :: (Traversable f, Num a) => (forall s. f (AD s (On (Kahn (Kahn a)))) -> AD s (On (Kahn (Kahn a)))) -> f a -> f (f a) Source #

Compute the hessian via the jacobian of the gradient. gradient is computed in Kahn mode and then the jacobian is computed in Kahn mode.

However, since the grad f :: f a -> f a is square this is not as fast as using the forward-mode jacobian of a reverse mode gradient provided by hessian.

>>> hessian (\[x,y] -> x*y) [1,2]
[[0,1],[1,0]]

hessianF :: (Traversable f, Functor g, Num a) => (forall s. f (AD s (On (Kahn (Kahn a)))) -> g (AD s (On (Kahn (Kahn a))))) -> f a -> g (f (f a)) Source #

Compute the order 3 Hessian tensor on a non-scalar-to-non-scalar function via the Kahn-mode Jacobian of the Kahn-mode Jacobian of the function.

Less efficient than hessianF.

>>> hessianF (\[x,y] -> [x*y,x+y,exp x*cos y]) [1,2]
[[[0.0,1.0],[1.0,0.0]],[[0.0,0.0],[0.0,0.0]],[[-1.1312043837568135,-2.4717266720048188],[-2.4717266720048188,1.1312043837568135]]]

Derivatives

diff :: Num a => (forall s. AD s (Kahn a) -> AD s (Kahn a)) -> a -> a Source #

Compute the derivative of a function.

>>> diff sin 0
1.0
>>> cos 0
1.0

diff' :: Num a => (forall s. AD s (Kahn a) -> AD s (Kahn a)) -> a -> (a, a) Source #

The diff' function calculates the value and derivative, as a pair, of a scalar-to-scalar function.

>>> diff' sin 0
(0.0,1.0)

diffF :: (Functor f, Num a) => (forall s. AD s (Kahn a) -> f (AD s (Kahn a))) -> a -> f a Source #

Compute the derivatives of a function that returns a vector with regards to its single input.

>>> diffF (\a -> [sin a, cos a]) 0
[1.0,0.0]

diffF' :: (Functor f, Num a) => (forall s. AD s (Kahn a) -> f (AD s (Kahn a))) -> a -> f (a, a) Source #

Compute the derivatives of a function that returns a vector with regards to its single input as well as the primal answer.

>>> diffF' (\a -> [sin a, cos a]) 0
[(0.0,1.0),(1.0,0.0)]