Spelling: decend/decent -> descend/descent

bench/bench-lstm.hs

@@ -27,8 +27,8 @@ main = do
                       , bench "backwards-60"      $ nf (nfT3 . uncurry4 (testRun60'  layer60))  (rec60, input40, rec60, rec60)
                       , bench "backwards-512"     $ nf (nfT3 . uncurry4 (testRun512' layer512)) (rec512, input40, rec512, rec512)
                       ]
-    , bgroup "update" [ bench "matrix-60x60"      $ nf (uncurry3 (decendVector 1 1 1)) (upIn60, upIn60, upIn60)
-                      , bench "matrix-512x512"    $ nf (uncurry3 (decendVector 1 1 1)) (upIn512, upIn512, upIn512)
+    , bgroup "update" [ bench "matrix-60x60"      $ nf (uncurry3 (descendVector 1 1 1)) (upIn60, upIn60, upIn60)
+                      , bench "matrix-512x512"    $ nf (uncurry3 (descendVector 1 1 1)) (upIn512, upIn512, upIn512)
                       ]
     , bgroup "train"  [ bench "one-time-step"     $ whnf (nfT2 . trainRecurrent lp lstm 0) [(input40, Just input40)]
                       , bench "ten-time-steps"    $ whnf (nfT2 . trainRecurrent lp lstm 0) $ replicate 10 (input40, Just input40)

cbits/gradient_descent.c

@@ -1,6 +1,6 @@
-#include "gradient_decent.h"
+#include "gradient_descent.h"
 
-void decend_cpu(int len, double rate, double momentum, double regulariser,
+void descend_cpu(int len, double rate, double momentum, double regulariser,
     const double* weights,
     const double* gradient,
     const double* last,

cbits/gradient_descent.h

@@ -1,7 +1,7 @@
 #include <stdio.h>
 #include <stdint.h>
 
-void decend_cpu(int len, double rate, double momentum, double regulariser,
+void descend_cpu(int len, double rate, double momentum, double regulariser,
     const double* weights,
     const double* gradient,
     const double* last,

grenade.cabal

@@ -17,7 +17,7 @@ description:
     Grenade provides an API for composing layers of a neural network
     into a sequence parallel graph in a type safe manner; running
     networks with reverse automatic differentiation to calculate their
-    gradients; and applying gradient decent for learning.
+    gradients; and applying gradient descent for learning.
     .
     Documentation and examples are available on github
     <https://github.com/HuwCampbell/grenade>.
@@ -26,8 +26,8 @@ extra-source-files:
   README.md
   cbits/im2col.h
   cbits/im2col.c
-  cbits/gradient_decent.h
-  cbits/gradient_decent.c
+  cbits/gradient_descent.h
+  cbits/gradient_descent.c
   cbits/pad.h
   cbits/pad.c
 
@@ -108,10 +108,10 @@ library
                        Grenade.Utils.OneHot
 
   includes:            cbits/im2col.h
-                       cbits/gradient_decent.h
+                       cbits/gradient_descent.h
                        cbits/pad.h
   c-sources:           cbits/im2col.c
-                       cbits/gradient_decent.c
+                       cbits/gradient_descent.c
                        cbits/pad.c
 
   cc-options:          -std=c99 -O3 -msse4.2 -Wall -Werror -DCABAL=1

src/Grenade/Core/Network.hs

@@ -139,7 +139,7 @@ runGradient net tapes o =
       = (GNil, o)
 
 
--- | Apply one step of stochastic gradient decent across the network.
+-- | Apply one step of stochastic gradient descent across the network.
 applyUpdate :: LearningParameters
             -> Network layers shapes
             -> Gradients layers

src/Grenade/Layers/Convolution.hs

@@ -131,7 +131,7 @@ instance ( KnownNat channels
          ) => UpdateLayer (Convolution channels filters kernelRows kernelColumns strideRows strideColumns) where
   type Gradient (Convolution channels filters kernelRows kernelCols strideRows strideCols) = (Convolution' channels filters kernelRows kernelCols strideRows strideCols)
   runUpdate LearningParameters {..} (Convolution oldKernel oldMomentum) (Convolution' kernelGradient) =
-    let (newKernel, newMomentum) = decendMatrix learningRate learningMomentum learningRegulariser oldKernel kernelGradient oldMomentum
+    let (newKernel, newMomentum) = descendMatrix learningRate learningMomentum learningRegulariser oldKernel kernelGradient oldMomentum
     in Convolution newKernel newMomentum
 
   createRandom = randomConvolution

src/Grenade/Layers/Deconvolution.hs

@@ -130,7 +130,7 @@ instance ( KnownNat channels
          ) => UpdateLayer (Deconvolution channels filters kernelRows kernelColumns strideRows strideColumns) where
   type Gradient (Deconvolution channels filters kernelRows kernelCols strideRows strideCols) = (Deconvolution' channels filters kernelRows kernelCols strideRows strideCols)
   runUpdate LearningParameters {..} (Deconvolution oldKernel oldMomentum) (Deconvolution' kernelGradient) =
-    let (newKernel, newMomentum) = decendMatrix learningRate learningMomentum learningRegulariser oldKernel kernelGradient oldMomentum
+    let (newKernel, newMomentum) = descendMatrix learningRate learningMomentum learningRegulariser oldKernel kernelGradient oldMomentum
     in Deconvolution newKernel newMomentum
 
   createRandom = randomDeconvolution

src/Grenade/Layers/FullyConnected.hs

@@ -39,8 +39,8 @@ instance (KnownNat i, KnownNat o) => UpdateLayer (FullyConnected i o) where
   type Gradient (FullyConnected i o) = (FullyConnected' i o)
 
   runUpdate LearningParameters {..} (FullyConnected (FullyConnected' oldBias oldActivations) (FullyConnected' oldBiasMomentum oldMomentum)) (FullyConnected' biasGradient activationGradient) =
-    let (newBias, newBiasMomentum)    = decendVector learningRate learningMomentum learningRegulariser oldBias biasGradient oldBiasMomentum
-        (newActivations, newMomentum) = decendMatrix learningRate learningMomentum learningRegulariser oldActivations activationGradient oldMomentum
+    let (newBias, newBiasMomentum)    = descendVector learningRate learningMomentum learningRegulariser oldBias biasGradient oldBiasMomentum
+        (newActivations, newMomentum) = descendMatrix learningRate learningMomentum learningRegulariser oldActivations activationGradient oldMomentum
     in FullyConnected (FullyConnected' newBias newActivations) (FullyConnected' newBiasMomentum newMomentum)
 
   createRandom = randomFullyConnected

src/Grenade/Layers/Internal/Update.hs

@@ -1,7 +1,7 @@
 {-# LANGUAGE ForeignFunctionInterface #-}
 module Grenade.Layers.Internal.Update (
-    decendMatrix
-  , decendVector
+    descendMatrix
+  , descendVector
   ) where
 
 import           Data.Maybe ( fromJust )
@@ -17,8 +17,8 @@ import qualified Numeric.LinearAlgebra.Devel as U
 
 import           System.IO.Unsafe ( unsafePerformIO )
 
-decendMatrix :: (KnownNat rows, KnownNat columns) => Double -> Double -> Double -> L rows columns -> L rows columns -> L rows columns -> (L rows columns, L rows columns)
-decendMatrix rate momentum regulariser weights gradient lastUpdate =
+descendMatrix :: (KnownNat rows, KnownNat columns) => Double -> Double -> Double -> L rows columns -> L rows columns -> L rows columns -> (L rows columns, L rows columns)
+descendMatrix rate momentum regulariser weights gradient lastUpdate =
   let (rows, cols) = size weights
       len          = rows * cols
       -- Most gradients come in in ColumnMajor,
@@ -29,7 +29,7 @@ decendMatrix rate momentum regulariser weights gradient lastUpdate =
       weights'     = flatten . tr . extract $ weights
       gradient'    = flatten . tr . extract $ gradient
       lastUpdate'  = flatten . tr . extract $ lastUpdate
-      (vw, vm)     = decendUnsafe len rate momentum regulariser weights' gradient' lastUpdate'
+      (vw, vm)     = descendUnsafe len rate momentum regulariser weights' gradient' lastUpdate'
 
       -- Note that it's ColumnMajor, as we did a transpose before
       -- using the internal vectors.
@@ -37,17 +37,17 @@ decendMatrix rate momentum regulariser weights gradient lastUpdate =
       mm           = U.matrixFromVector U.ColumnMajor rows cols vm
   in  (fromJust . create $ mw, fromJust . create $ mm)
 
-decendVector :: (KnownNat r) => Double -> Double -> Double -> R r -> R r -> R r -> (R r, R r)
-decendVector rate momentum regulariser weights gradient lastUpdate =
+descendVector :: (KnownNat r) => Double -> Double -> Double -> R r -> R r -> R r -> (R r, R r)
+descendVector rate momentum regulariser weights gradient lastUpdate =
   let len          = size weights
       weights'     = extract weights
       gradient'    = extract gradient
       lastUpdate'  = extract lastUpdate
-      (vw, vm)     = decendUnsafe len rate momentum regulariser weights' gradient' lastUpdate'
+      (vw, vm)     = descendUnsafe len rate momentum regulariser weights' gradient' lastUpdate'
   in  (fromJust $ create vw, fromJust $ create vm)
 
-decendUnsafe :: Int -> Double -> Double -> Double -> Vector Double -> Vector Double -> Vector Double -> (Vector Double, Vector Double)
-decendUnsafe len rate momentum regulariser weights gradient lastUpdate =
+descendUnsafe :: Int -> Double -> Double -> Double -> Vector Double -> Vector Double -> Vector Double -> (Vector Double, Vector Double)
+descendUnsafe len rate momentum regulariser weights gradient lastUpdate =
   unsafePerformIO $ do
     outWPtr <- mallocForeignPtrArray len
     outMPtr <- mallocForeignPtrArray len
@@ -60,11 +60,11 @@ decendUnsafe len rate momentum regulariser weights gradient lastUpdate =
         withForeignPtr lPtr $ \lPtr' ->
           withForeignPtr outWPtr $ \outWPtr' ->
             withForeignPtr outMPtr $ \outMPtr' ->
-              decend_cpu len rate momentum regulariser wPtr' gPtr' lPtr' outWPtr' outMPtr'
+              descend_cpu len rate momentum regulariser wPtr' gPtr' lPtr' outWPtr' outMPtr'
 
     return (U.unsafeFromForeignPtr0 outWPtr len, U.unsafeFromForeignPtr0 outMPtr len)
 
 foreign import ccall unsafe
-    decend_cpu
+    descend_cpu
       :: Int -> Double -> Double -> Double -> Ptr Double -> Ptr Double -> Ptr Double -> Ptr Double -> Ptr Double -> IO ()
 

src/Grenade/Recurrent/Layers/LSTM.hs

@@ -87,11 +87,11 @@ instance (KnownNat i, KnownNat o) => UpdateLayer (LSTM i o) where
     -- Utility function for updating with the momentum, gradients, and weights.
     u :: forall x ix out. (KnownNat ix, KnownNat out) => (x -> (L out ix)) -> x -> x -> x -> ((L out ix), (L out ix))
     u e (e -> weights) (e -> momentum) (e -> gradient) =
-      decendMatrix learningRate learningMomentum learningRegulariser weights gradient momentum
+      descendMatrix learningRate learningMomentum learningRegulariser weights gradient momentum
 
     v :: forall x ix. (KnownNat ix) => (x -> (R ix)) -> x -> x -> x -> ((R ix), (R ix))
     v e (e -> weights) (e -> momentum) (e -> gradient) =
-      decendVector learningRate learningMomentum learningRegulariser weights gradient momentum
+      descendVector learningRate learningMomentum learningRegulariser weights gradient momentum
 
   -- There's a lot of updates here, so to try and minimise the number of data copies
   -- we'll create a mutable bucket for each.