-- Hoogle documentation, generated by Haddock
-- See Hoogle, http://www.haskell.org/hoogle/


-- | Reverse Automatic Differentiation.
--   
--   Reverse-Mode Automatic Differentiation via overloading. Existential
--   type "branding" is used to prevent sensitivity confusion.
@package rad
@version 0.1.6


-- | Reverse Mode Automatic Differentiation via overloading to perform
--   nonstandard interpretation that replaces original numeric type with a
--   bundle that contains a value of the original type and the tape that
--   will be used to recover the value of the sensitivity.
--   
--   This package uses StableNames internally to recover sharing
--   information from the tape to avoid combinatorial explosion, and thus
--   runs asymptotically faster than it could without such sharing
--   information, but the use of side-effects contained herein is benign.
--   
--   The API has been built to be close to the design of
--   <tt>Numeric.FAD</tt> from the <tt>fad</tt> package by Barak
--   Pearlmutter and Jeffrey Mark Siskind and contains portions of that
--   code, with minor liberties taken.
module Numeric.RAD
data RAD s a

-- | The <a>lift</a> function injects a primal number into the RAD data
--   type with a 0 derivative. If reverse-mode AD numbers formed a monad,
--   then <a>lift</a> would be <a>return</a>.
lift :: a -> RAD s a

-- | The <a>diffUU</a> function calculates the first derivative of a
--   scalar-to-scalar function.
diffUU :: Num a => (forall s. RAD s a -> RAD s a) -> a -> a

-- | The <a>diffUF</a> function calculates the first derivative of
--   scalar-to-nonscalar function.
diffUF :: (Functor f, Num a) => (forall s. RAD s a -> f (RAD s a)) -> a -> f a

-- | The <a>diff2UU</a> function calculates the value and derivative, as a
--   pair, of a scalar-to-scalar function.
diff2UU :: Num a => (forall s. RAD s a -> RAD s a) -> a -> (a, a)

-- | Note that the signature differs from that used in Numeric.FAD, because
--   while you can always <a>unzip</a> an arbitrary functor, not all
--   functors can be zipped.
diff2UF :: (Functor f, Num a) => (forall s. RAD s a -> f (RAD s a)) -> a -> f (a, a)

-- | The <a>diff</a> function is a synonym for <a>diffUU</a>.
diff :: Num a => (forall s. RAD s a -> RAD s a) -> a -> a

-- | The <a>diff2</a> function is a synonym for <a>diff2UU</a>.
diff2 :: Num a => (forall s. RAD s a -> RAD s a) -> a -> (a, a)

-- | The <a>jacobian</a> function calcualtes the Jacobian of a
--   nonscalar-to-nonscalar function, using m invocations of reverse AD,
--   where m is the output dimensionality. When the output dimensionality
--   is significantly greater than the input dimensionality you should use
--   <tt>Numeric.FAD.jacobian</tt> instead.
jacobian :: (Traversable f, Functor g, Num a) => (forall s. f (RAD s a) -> g (RAD s a)) -> f a -> g (f a)

-- | The <a>jacobian2</a> function calcualtes both the result and the
--   Jacobian of a nonscalar-to-nonscalar function, using m invocations of
--   reverse AD, where m is the output dimensionality. 'fmap snd' on the
--   result will recover the result of <a>jacobian</a>
jacobian2 :: (Traversable f, Functor g, Num a) => (forall s. f (RAD s a) -> g (RAD s a)) -> f a -> g (a, f a)
grad :: (Traversable f, Num a) => (forall s. f (RAD s a) -> RAD s a) -> f a -> f a
grad2 :: (Traversable f, Num a) => (forall s. f (RAD s a) -> RAD s a) -> f a -> (a, f a)

-- | The <a>zeroNewton</a> function finds a zero of a scalar function using
--   Newton's method; its output is a stream of increasingly accurate
--   results. (Modulo the usual caveats.)
--   
--   TEST CASE: <tt>take 10 $ zeroNewton (\x-&gt;x^2-4) 1 -- converge to
--   2.0</tt>
--   
--   TEST CASE :module Data.Complex Numeric.RAD <tt>take 10 $ zeroNewton
--   ((+1).(^2)) (1 :+ 1) -- converge to (0 :+ 1)</tt>
zeroNewton :: Fractional a => (forall s. RAD s a -> RAD s a) -> a -> [a]

-- | The <a>inverseNewton</a> function inverts a scalar function using
--   Newton's method; its output is a stream of increasingly accurate
--   results. (Modulo the usual caveats.)
--   
--   TEST CASE: <tt>take 10 $ inverseNewton sqrt 1 (sqrt 10) -- converge to
--   10</tt>
inverseNewton :: Fractional a => (forall s. RAD s a -> RAD s a) -> a -> a -> [a]

-- | The <a>fixedPointNewton</a> function find a fixedpoint of a scalar
--   function using Newton's method; its output is a stream of increasingly
--   accurate results. (Modulo the usual caveats.)
fixedPointNewton :: Fractional a => (forall s. RAD s a -> RAD s a) -> a -> [a]

-- | The <a>extremumNewton</a> function finds an extremum of a scalar
--   function using Newton's method; produces a stream of increasingly
--   accurate results. (Modulo the usual caveats.)
extremumNewton :: Fractional a => (forall s t. RAD t (RAD s a) -> RAD t (RAD s a)) -> a -> [a]

-- | The <a>argminNaiveGradient</a> function performs a multivariate
--   optimization, based on the naive-gradient-descent in the file
--   <tt>stalingrad/examples/flow-tests/pre-saddle-1a.vlad</tt> from the
--   VLAD compiler Stalingrad sources. Its output is a stream of
--   increasingly accurate results. (Modulo the usual caveats.) This is
--   <i>O(n)</i> faster than <tt>Numeric.FAD.argminNaiveGradient</tt>
argminNaiveGradient :: (Fractional a, Ord a) => (forall s. [RAD s a] -> RAD s a) -> [a] -> [[a]]
instance Monad S
instance Floating a => Floating (RAD s a)
instance Fractional a => Fractional (RAD s a)
instance RealFrac a => RealFrac (RAD s a)
instance RealFloat a => RealFloat (RAD s a)
instance Real a => Real (RAD s a)
instance Num a => Num (RAD s a)
instance (Num a, Enum a) => Enum (RAD s a)
instance Bounded a => Bounded (RAD s a)
instance Ord a => Ord (RAD s a)
instance Eq a => Eq (RAD s a)
instance MuRef (RAD s a)
instance Show a => Show (RAD s a)