e6293b8 · grenade · git.iko.soy

LICENSE+1−1

@@ -1,4 +1,4 @@

22All rights reserved.

33

44Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

README.md+6−2

@@ -20,8 +20,12 @@ specified and initialised with random weights in a few lines of code with

2020```haskell

2121type MNIST

2222 = Network

23− '[ Convolution 1 10 5 5 1 1, Pooling 2 2 2 2, Relu, Convolution 10 16 5 5 1 1, Pooling 2 2 2 2, FlattenLayer, Relu, FullyConnected 256 80, Logit, FullyConnected 80 10, Logit]

24− '[ 'D2 28 28, 'D3 24 24 10, 'D3 12 12 10, 'D3 12 12 10, 'D3 8 8 16, 'D3 4 4 16, 'D1 256, 'D1 256, 'D1 80, 'D1 80, 'D1 10, 'D1 10]

23+ '[ Convolution 1 10 5 5 1 1, Pooling 2 2 2 2, Relu

24+ , Convolution 10 16 5 5 1 1, Pooling 2 2 2 2, FlattenLayer, Relu

25+ , FullyConnected 256 80, Logit, FullyConnected 80 10, Logit]

26+ '[ 'D2 28 28, 'D3 24 24 10, 'D3 12 12 10, 'D3 12 12 10

27+ , 'D3 8 8 16, 'D3 4 4 16, 'D1 256, 'D1 256

28+ , 'D1 80, 'D1 80, 'D1 10, 'D1 10]

2529

2630randomMnist :: MonadRandom m => m MNIST

2731randomMnist = randomNetwork

examples/LICENSEnew file+10

@@ -0,0 +1,10 @@

3+

4+Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

5+

6+1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.

7+

8+2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.

9+

10+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

examples/grenade-examples.cabalnew file+108

@@ -0,0 +1,108 @@

1+name: grenade-examples

2+version: 0.0.1

3+license: BSD2

4+license-file: LICENSE

5+author: Huw Campbell <huw.campbell@gmail.com>

6+maintainer: Huw Campbell <huw.campbell@gmail.com>

8+synopsis: grenade-examples

9+category: System

10+cabal-version: >= 1.8

11+build-type: Simple

12+description: grenade-examples

13+

14+source-repository head

15+ type: git

16+ location: https://github.com/HuwCampbell/grenade.git

17+

18+library

19+

20+executable feedforward

21+ ghc-options: -Wall -threaded -O2

22+ main-is: main/feedforward.hs

23+ build-depends: base

24+ , grenade

25+ , attoparsec

26+ , bytestring

27+ , cereal

28+ , either

29+ , optparse-applicative == 0.13.*

30+ , text == 1.2.*

31+ , mtl >= 2.2.1 && < 2.3

32+ , hmatrix

33+ , transformers

34+ , singletons

35+ , semigroups

36+ , MonadRandom

37+

38+executable mnist

39+ ghc-options: -Wall -threaded -O2

40+ main-is: main/mnist.hs

41+ build-depends: base

42+ , grenade

43+ , attoparsec

44+ , either

45+ , optparse-applicative == 0.13.*

46+ , text == 1.2.*

47+ , mtl >= 2.2.1 && < 2.3

48+ , hmatrix >= 0.18 && < 0.19

49+ , transformers

50+ , semigroups

51+ , singletons

52+ , MonadRandom

53+ , vector

54+

55+executable gan-mnist

56+ ghc-options: -Wall -threaded -O2

57+ main-is: main/gan-mnist.hs

58+ build-depends: base

59+ , grenade

60+ , attoparsec

61+ , bytestring

62+ , cereal

63+ , either

64+ , optparse-applicative == 0.13.*

65+ , text == 1.2.*

66+ , mtl >= 2.2.1 && < 2.3

67+ , hmatrix >= 0.18 && < 0.19

68+ , transformers

69+ , semigroups

70+ , singletons

71+ , MonadRandom

72+ , vector

73+

74+executable recurrent

75+ ghc-options: -Wall -threaded -O2

76+ main-is: main/recurrent.hs

77+ build-depends: base

78+ , grenade

79+ , attoparsec

80+ , either

81+ , optparse-applicative == 0.13.*

82+ , text == 1.2.*

83+ , mtl >= 2.2.1 && < 2.3

84+ , hmatrix >= 0.18 && < 0.19

85+ , transformers

86+ , semigroups

87+ , singletons

88+ , MonadRandom

89+

90+executable shakespeare

91+ ghc-options: -Wall -threaded -O2

92+ main-is: main/shakespeare.hs

93+ build-depends: base

94+ , grenade

95+ , attoparsec

96+ , bytestring

97+ , cereal

98+ , either

99+ , optparse-applicative == 0.13.*

100+ , text == 1.2.*

101+ , mtl >= 2.2.1 && < 2.3

102+ , hmatrix >= 0.18 && < 0.19

103+ , transformers

104+ , semigroups

105+ , singletons

106+ , vector

107+ , MonadRandom

108+ , containers

main/feedforward.hs → examples/main/feedforward.hsrenamed

The content is the same.

main/gan-mnist.hs → examples/main/gan-mnist.hsrenamed+4−3

@@ -58,7 +58,7 @@ import Grenade

5858import Grenade.Utils.OneHot

5959

6060type Discriminator = Network '[ Convolution 1 10 5 5 1 1, Pooling 2 2 2 2, Relu, Convolution 10 16 5 5 1 1, Pooling 2 2 2 2, Reshape, Relu, FullyConnected 256 80, Logit, FullyConnected 80 1, Logit]

61− '[ 'D2 28 28, 'D3 24 24 10, 'D3 12 12 10, 'D3 12 12 10, 'D3 8 8 16, 'D3 4 4 16, 'D1 256, 'D1 256, 'D1 80, 'D1 80, 'D1 1, 'D1 1]

61+ '[ 'D2 28 28, 'D3 24 24 10, 'D3 12 12 10, 'D3 12 12 10, 'D3 8 8 16, 'D3 4 4 16, 'D1 256, 'D1 256, 'D1 80, 'D1 80, 'D1 1, 'D1 1]

6262

6363type Generator = Network '[ FullyConnected 100 10240, Relu, Reshape, Convolution 10 10 5 5 1 1, Relu, Convolution 10 1 1 1 1 1, Logit, Reshape]

6464 '[ 'D1 100, 'D1 10240, 'D1 10240, 'D3 32 32 10, 'D3 28 28 10, 'D3 28 28 10, 'D3 28 28 1, 'D3 28 28 1, 'D2 28 28 ]

@@ -77,9 +77,10 @@ trainExample rate discriminator generator realExample noiseSource

7777 (discriminatorTapeFake, guessFake) = runNetwork discriminator fakeExample

7878

7979 (discriminator'real, _) = runGradient discriminator discriminatorTapeReal ( guessReal - 1 )

80− (discriminator'fake, push) = runGradient discriminator discriminatorTapeFake guessFake

80+ (discriminator'fake, _) = runGradient discriminator discriminatorTapeFake guessFake

81+ (_, push) = runGradient discriminator discriminatorTapeFake ( guessFake - 1)

8182

82− (generator', _) = runGradient generator generatorTape (-push)

83+ (generator', _) = runGradient generator generatorTape push

8384

8485 newDiscriminator = foldl' (applyUpdate rate { learningRegulariser = learningRegulariser rate * 10}) discriminator [ discriminator'real, discriminator'fake ]

8586 newGenerator = applyUpdate rate generator generator'

main/mnist.hs → examples/main/mnist.hsrenamed+18−2

@@ -34,8 +34,24 @@ import Grenade.Utils.OneHot

3434

3535-- With the mnist data from Kaggle normalised to doubles between 0 and 1, learning rate of 0.01 and 15 iterations,

3636-- this network should get down to about a 1.3% error rate.

37−type MNIST = Network '[ Convolution 1 10 5 5 1 1, Pooling 2 2 2 2, Relu, Convolution 10 16 5 5 1 1, Pooling 2 2 2 2, Reshape, Relu, FullyConnected 256 80, Logit, FullyConnected 80 10, Logit]

38− '[ 'D2 28 28, 'D3 24 24 10, 'D3 12 12 10, 'D3 12 12 10, 'D3 8 8 16, 'D3 4 4 16, 'D1 256, 'D1 256, 'D1 80, 'D1 80, 'D1 10, 'D1 10]

37+--

38+-- /NOTE:/ This model is actually too complex for MNIST, and one should use the type given in the readme instead.

39+-- This one is just here to demonstrate Inception layers in use.

40+--

41+type MNIST =

42+ Network

43+ '[ Reshape

44+ , Inception 28 28 1 5 5 5, Pooling 2 2 2 2, Relu

45+ , Inception 14 14 15 5 5 5, Pooling 2 2 2 2, Relu

46+ , Reshape

47+ , FullyConnected 735 80, Logit

48+ , FullyConnected 80 10, Logit]

49+ '[ 'D2 28 28, 'D3 28 28 1

50+ , 'D3 28 28 15, 'D3 14 14 15, 'D3 14 14 15

51+ , 'D3 14 14 15, 'D3 7 7 15, 'D3 7 7 15

52+ , 'D1 735

53+ , 'D1 80, 'D1 80

54+ , 'D1 10, 'D1 10]

3955

4056randomMnist :: MonadRandom m => m MNIST

4157randomMnist = randomNetwork

main/recurrent.hs → examples/main/recurrent.hsrenamed

The content is the same.

main/shakespeare.hs → examples/main/shakespeare.hsrenamed

The content is the same.

grenade.cabal+36−121

@@ -4,7 +4,7 @@ license: BSD2

44license-file: LICENSE

55author: Huw Campbell <huw.campbell@gmail.com>

66maintainer: Huw Campbell <huw.campbell@gmail.com>

88synopsis: grenade

99category: System

1010cabal-version: >= 1.8

@@ -12,12 +12,12 @@ build-type: Simple

1212description: grenade.

1313

1414extra-source-files:

15− cbits/im2col.h

16− cbits/im2col.c

17− cbits/gradient_decent.h

18− cbits/gradient_decent.c

19− cbits/pad.h

20− cbits/pad.c

15+ cbits/im2col.h

16+ cbits/im2col.c

17+ cbits/gradient_decent.h

18+ cbits/gradient_decent.c

19+ cbits/pad.h

20+ cbits/pad.c

2121

2222source-repository head

2323 type: git

@@ -27,25 +27,26 @@ library

2727 build-depends:

2828 base >= 4.8 && < 5

2929 , bytestring == 0.10.*

30− , containers

31− , deepseq

32− , either == 4.4.*

33− , cereal

30+ , containers >= 0.5 && < 0.6

31+ , cereal >= 0.5 && < 0.6

32+ , deepseq >= 1.4 && < 1.5

3433 , exceptions == 0.8.*

3534 , hmatrix == 0.18.*

36− , MonadRandom

37− , mtl >= 2.2.1 && < 2.3

38− , primitive

35+ , MonadRandom >= 0.4 && < 0.6

36+ , mtl >= 2.2.1 && < 2.3

37+ , primitive >= 0.6 && < 0.7

3938 , text == 1.2.*

40− , transformers

41− , singletons >= 2.1 && < 2.3

42− , vector == 0.11.*

39+ , singletons >= 2.1 && < 2.3

40+ , vector >= 0.11 && < 0.13

4341

4442 ghc-options:

4543 -Wall

4644 hs-source-dirs:

4745 src

4846

47+ if impl(ghc < 8.0)

48+ ghc-options: -fno-warn-incomplete-patterns

49+

4950

5051 exposed-modules:

5152 Grenade

@@ -55,19 +56,24 @@ library

5556 Grenade.Core.Network

5657 Grenade.Core.Runner

5758 Grenade.Core.Shape

58− Grenade.Layers.Crop

59+

60+ Grenade.Layers

5961 Grenade.Layers.Concat

6062 Grenade.Layers.Convolution

63+ Grenade.Layers.Crop

6164 Grenade.Layers.Dropout

65+ Grenade.Layers.Elu

6266 Grenade.Layers.FullyConnected

63− Grenade.Layers.Reshape

67+ Grenade.Layers.Inception

6468 Grenade.Layers.Logit

6569 Grenade.Layers.Merge

66− Grenade.Layers.Relu

67− Grenade.Layers.Elu

68− Grenade.Layers.Tanh

6970 Grenade.Layers.Pad

7071 Grenade.Layers.Pooling

72+ Grenade.Layers.Relu

73+ Grenade.Layers.Reshape

74+ Grenade.Layers.Softmax

75+ Grenade.Layers.Tanh

76+ Grenade.Layers.Trivial

7177

7278 Grenade.Layers.Internal.Convolution

7379 Grenade.Layers.Internal.Pad

@@ -81,111 +87,20 @@ library

8187 Grenade.Recurrent.Core.Network

8288 Grenade.Recurrent.Core.Runner

8389

90+ Grenade.Recurrent.Layers

8491 Grenade.Recurrent.Layers.BasicRecurrent

8592 Grenade.Recurrent.Layers.LSTM

8693

8794 Grenade.Utils.OneHot

8895

89− includes: cbits/im2col.h

90− cbits/gradient_decent.h

91− cbits/pad.h

92− c-sources: cbits/im2col.c

93− cbits/gradient_decent.c

94− cbits/pad.c

95−

96− cc-options: -std=c99 -O3 -msse4.2 -Wall -Werror -DCABAL=1

97−

98−executable feedforward

99− ghc-options: -Wall -threaded -O2

100− main-is: main/feedforward.hs

101− build-depends: base

102− , grenade

103− , attoparsec

104− , bytestring

105− , cereal

106− , either

107− , optparse-applicative == 0.13.*

108− , text == 1.2.*

109− , mtl >= 2.2.1 && < 2.3

110− , hmatrix

111− , transformers

112− , singletons

113− , semigroups

114− , MonadRandom

115−

116−executable mnist

117− ghc-options: -Wall -threaded -O2

118− main-is: main/mnist.hs

119− build-depends: base

120− , grenade

121− , attoparsec

122− , either

123− , optparse-applicative == 0.13.*

124− , text == 1.2.*

125− , mtl >= 2.2.1 && < 2.3

126− , hmatrix >= 0.18 && < 0.19

127− , transformers

128− , semigroups

129− , singletons

130− , MonadRandom

131− , vector

132−

133−executable gan-mnist

134− ghc-options: -Wall -threaded -O2

135− main-is: main/gan-mnist.hs

136− build-depends: base

137− , grenade

138− , attoparsec

139− , bytestring

140− , cereal

141− , either

142− , optparse-applicative == 0.13.*

143− , text == 1.2.*

144− , mtl >= 2.2.1 && < 2.3

145− , hmatrix >= 0.18 && < 0.19

146− , transformers

147− , semigroups

148− , singletons

149− , MonadRandom

150− , vector

151−

152−executable recurrent

153− ghc-options: -Wall -threaded -O2

154− main-is: main/recurrent.hs

155− build-depends: base

156− , grenade

157− , attoparsec

158− , either

159− , optparse-applicative == 0.13.*

160− , text == 1.2.*

161− , mtl >= 2.2.1 && < 2.3

162− , hmatrix >= 0.18 && < 0.19

163− , transformers

164− , semigroups

165− , singletons

166− , MonadRandom

167−

168−

169−executable shakespeare

170− ghc-options: -Wall -threaded -O2

171− main-is: main/shakespeare.hs

172− build-depends: base

173− , grenade

174− , attoparsec

175− , bytestring

176− , cereal

177− , either

178− , optparse-applicative == 0.13.*

179− , text == 1.2.*

180− , mtl >= 2.2.1 && < 2.3

181− , hmatrix >= 0.18 && < 0.19

182− , transformers

183− , semigroups

184− , singletons

185− , vector

186− , MonadRandom

187− , containers

96+ includes: cbits/im2col.h

97+ cbits/gradient_decent.h

98+ cbits/pad.h

99+ c-sources: cbits/im2col.c

100+ cbits/gradient_decent.c

101+ cbits/pad.c

188102

103+ cc-options: -std=c99 -O3 -msse4.2 -Wall -Werror -DCABAL=1

189104

190105test-suite test

191106 type: exitcode-stdio-1.0

src/Grenade.hs+49−20

@@ -1,23 +1,52 @@

11module Grenade (

2− module X

2+ -- | This is an empty module which simply re-exports public definitions

3+ -- for machine learning with Grenade.

4+

5+ -- * Exported modules

6+ --

7+ -- | The core types and runners for Grenade.

8+ module Grenade.Core

9+

10+ -- | The neural network layer zoo

11+ , module Grenade.Layers

12+

13+

14+ -- * Overview of the library

15+ -- $library

16+

17+ -- * Example usage

18+ -- $example

19+

320 ) where

421

5−import Grenade.Core.LearningParameters as X

6−import Grenade.Core.Layer as X

7−import Grenade.Core.Network as X

8−import Grenade.Core.Runner as X

9−import Grenade.Core.Shape as X

10−import Grenade.Layers.Concat as X

11−import Grenade.Layers.Crop as X

12−import Grenade.Layers.Dropout as X

13−import Grenade.Layers.Pad as X

14−import Grenade.Layers.Pooling as X

15−import Grenade.Layers.Reshape as X

16−import Grenade.Layers.FullyConnected as X

17−import Grenade.Layers.Logit as X

18−import Grenade.Layers.Merge as X

19−import Grenade.Layers.Convolution as X

20−import Grenade.Layers.Relu as X

21−import Grenade.Layers.Elu as X

22−import Grenade.Layers.Tanh as X

23−import Grenade.Layers.Softmax as X

22+import Grenade.Core

23+import Grenade.Layers

24+

25+{- $library

26+Grenade is a purely functional deep learning library.

27+

28+It provides an expressive type level API for the construction

29+of complex neural network architectures. Backing this API is and

30+implementation written using BLAS and LAPACK, mostly provided by

31+the hmatrix library.

32+-}

33+

34+{- $example

35+A few examples are provided at https://github.com/HuwCampbell/grenade

36+under the examples folder.

37+

38+The starting place is to write your neural network type and a

39+function to create a random layer of that type. The following

40+is a simple example which runs a logistic regression.

41+

42+> type MyNet = Network '[ FullyConnected 10 1, Logit ] '[ 'D1 10, 'D1 1, 'D1 1 ]

43+>

44+> randomMyNet :: MonadRandom MyNet

45+> randomMyNet = randomNetwork

46+

47+The function `randomMyNet` witnesses the `CreatableNetwork`

48+constraint of the neural network, that is it ensures the network

49+can be built, and hence, that the architecture is sound.

50+-}

51+

52+

src/Grenade/Core.hs+10−5

@@ -1,8 +1,13 @@

11module Grenade.Core (

2− module X

2+ module Grenade.Core.Layer

3+ , module Grenade.Core.LearningParameters

4+ , module Grenade.Core.Network

5+ , module Grenade.Core.Runner

6+ , module Grenade.Core.Shape

37 ) where

48

5−import Grenade.Core.Layer as X

6−import Grenade.Core.LearningParameters as X

7−import Grenade.Core.Shape as X

8−import Grenade.Core.Network as X

9+import Grenade.Core.Layer

10+import Grenade.Core.LearningParameters

11+import Grenade.Core.Network

12+import Grenade.Core.Runner

13+import Grenade.Core.Shape

src/Grenade/Core/LearningParameters.hs+4

@@ -1,4 +1,8 @@

11module Grenade.Core.LearningParameters (

2+ -- | This module contains learning algorithm specific

3+ -- code. Currently, this module should be consifered

4+ -- unstable, due to issue #26.

5+

26 LearningParameters (..)

37 ) where

48

src/Grenade/Core/Network.hs+2−7

@@ -1,4 +1,3 @@

1−{-# LANGUAGE CPP #-}

21{-# LANGUAGE DataKinds #-}

32{-# LANGUAGE BangPatterns #-}

43{-# LANGUAGE GADTs #-}

@@ -19,10 +18,6 @@ This module defines the core data types and functions

1918for non-recurrent neural networks.

2019-}

2120

22−#if __GLASGOW_HASKELL__ < 800

23−{-# OPTIONS_GHC -fno-warn-incomplete-patterns #-}

24−#endif

25−

2621module Grenade.Core.Network (

2722 Network (..)

2823 , Gradients (..)

@@ -47,9 +42,9 @@ import Grenade.Core.Shape

4742

4843-- | Type of a network.

4944--

50−-- The [*] type specifies the types of the layers.

45+-- The @[*]@ type specifies the types of the layers.

5146--

52−-- The [Shape] type specifies the shapes of data passed between the layers.

47+-- The @[Shape]@ type specifies the shapes of data passed between the layers.

5348--

5449-- Can be considered to be a heterogeneous list of layers which are able to

5550-- transform the data shapes of the network.

src/Grenade/Core/Shape.hs−9

@@ -1,4 +1,3 @@

1−{-# LANGUAGE CPP #-}

21{-# LANGUAGE DataKinds #-}

32{-# LANGUAGE GADTs #-}

43{-# LANGUAGE KindSignatures #-}

@@ -8,11 +7,6 @@

87{-# LANGUAGE FlexibleContexts #-}

98{-# LANGUAGE ScopedTypeVariables #-}

109{-# LANGUAGE RankNTypes #-}

11−

12−-- Ghc 7.10 fails to recognise n2 is complete.

13−#if __GLASGOW_HASKELL__ < 800

14−{-# OPTIONS_GHC -fno-warn-incomplete-patterns #-}

15−#endif

1610{-|

1711Module : Grenade.Core.Shape

1812Description : Core definition of the Shapes of data we understand

@@ -65,17 +59,14 @@ data Shape

6559-- All shapes are held in contiguous memory.

6660-- 3D is held in a matrix (usually row oriented) which has height depth * rows.

6761data S (n :: Shape) where

68− -- | One dimensional data

6962 S1D :: ( KnownNat len )

7063 => R len

7164 -> S ('D1 len)

7265

73− -- | Two dimensional data

7466 S2D :: ( KnownNat rows, KnownNat columns )

7567 => L rows columns

7668 -> S ('D2 rows columns)

7769

78− -- | Three dimensional data

7970 S3D :: ( KnownNat rows

8071 , KnownNat columns

8172 , KnownNat depth

src/Grenade/Layers.hsnew file+33

@@ -0,0 +1,33 @@

1+module Grenade.Layers (

2+ module Grenade.Layers.Concat

3+ , module Grenade.Layers.Convolution

4+ , module Grenade.Layers.Crop

5+ , module Grenade.Layers.Elu

6+ , module Grenade.Layers.FullyConnected

7+ , module Grenade.Layers.Inception

8+ , module Grenade.Layers.Logit

9+ , module Grenade.Layers.Merge

10+ , module Grenade.Layers.Pad

11+ , module Grenade.Layers.Pooling

12+ , module Grenade.Layers.Reshape

13+ , module Grenade.Layers.Relu

14+ , module Grenade.Layers.Softmax

15+ , module Grenade.Layers.Tanh

16+ , module Grenade.Layers.Trivial

17+ ) where

18+

19+import Grenade.Layers.Concat

20+import Grenade.Layers.Convolution

21+import Grenade.Layers.Crop

22+import Grenade.Layers.Elu

23+import Grenade.Layers.Pad

24+import Grenade.Layers.FullyConnected

25+import Grenade.Layers.Inception

26+import Grenade.Layers.Logit

27+import Grenade.Layers.Merge

28+import Grenade.Layers.Pooling

29+import Grenade.Layers.Reshape

30+import Grenade.Layers.Relu

31+import Grenade.Layers.Softmax

32+import Grenade.Layers.Tanh

33+import Grenade.Layers.Trivial

src/Grenade/Layers/Concat.hs+4−2

@@ -9,11 +9,13 @@

99{-# LANGUAGE ScopedTypeVariables #-}

1010{-# LANGUAGE StandaloneDeriving #-}

1111{-|

12−Module : Grenade.Core.Network

13−Description : Core definition a simple neural etwork

12+Module : Grenade.Layers.Concat

13+Description : Concatenation layer

1414Copyright : (c) Huw Campbell, 2016-2017

1515License : BSD2

1616Stability : experimental

17+

18+This module provides the concatenation layer, whic used to run two separate layers in parallel and combine their outputs.

1719-}

1820module Grenade.Layers.Concat (

1921 Concat (..)

src/Grenade/Layers/Inception.hs+26−12

@@ -9,10 +9,13 @@

99{-# LANGUAGE ScopedTypeVariables #-}

1010{-|

1111Module : Grenade.Core.Network

12−Description : Core definition a simple neural etwork

12+Description : Inception style parallel convolutional network composition.

1313Copyright : (c) Huw Campbell, 2016-2017

1414License : BSD2

1515Stability : experimental

16+

17+Export an Inception style type, which can be used to build up

18+complex multiconvolution size networks.

1619-}

1720module Grenade.Layers.Inception (

1821 Inception

@@ -25,24 +28,35 @@ import Grenade.Layers.Convolution

2528import Grenade.Layers.Pad

2629import Grenade.Layers.Concat

2730

28−

31+-- | Type of an inception layer.

32+--

33+-- It looks like a bit of a handful, but is actually pretty easy to use.

34+--

35+-- The first three type parameters are the size of the (3D) data the

36+-- inception layer will take. It will emit 3D data with the number of

37+-- channels being the sum of @chx@, @chy@, @chz@, which are the number

38+-- of convolution filters in the 3x3, 5x5, and 7x7 convolutions Layers

39+-- respectively.

40+--

41+-- The network get padded effectively before each convolution filters

42+-- such that the output dimension is the same x and y as the input.

2943type Inception rows cols channels chx chy chz

30− = Network '[ Concat ('D3 (rows - 2) (cols - 2) (chx + chy)) (InceptionS rows cols channels chx chy) ('D3 (rows - 2) (cols - 2) chz) (Inception7x7 rows cols channels chz) ]

31− '[ 'D3 rows cols channels, 'D3 (rows -2) (cols -2) (chx + chy + chz) ]

44+ = Network '[ Concat ('D3 rows cols (chx + chy)) (InceptionS rows cols channels chx chy) ('D3 rows cols chz) (Inception7x7 rows cols channels chz) ]

45+ '[ 'D3 rows cols channels, 'D3 rows cols (chx + chy + chz) ]

3246

3347type InceptionS rows cols channels chx chy

34− = Network '[ Concat ('D3 (rows - 2) (cols - 2) chx) (Inception3x3 rows cols channels chx) ('D3 (rows - 2) (cols - 2) chy) (Inception5x5 rows cols channels chy) ]

35− '[ 'D3 rows cols channels, 'D3 (rows -2) (cols -2) (chx + chy) ]

48+ = Network '[ Concat ('D3 rows cols chx) (Inception3x3 rows cols channels chx) ('D3 rows cols chy) (Inception5x5 rows cols channels chy) ]

49+ '[ 'D3 rows cols channels, 'D3 rows cols (chx + chy) ]

3650

3751type Inception3x3 rows cols channels chx

38− = Network '[ Convolution channels chx 3 3 1 1 ]

39− '[ 'D3 rows cols channels, 'D3 (rows -2) (cols -2) chx ]

52+ = Network '[ Pad 1 1 1 1, Convolution channels chx 3 3 1 1 ]

53+ '[ 'D3 rows cols channels, 'D3 (rows + 2) (cols + 2) channels, 'D3 rows cols chx ]

4054

4155type Inception5x5 rows cols channels chx

42− = Network '[ Pad 1 1 1 1, Convolution channels chx 5 5 1 1 ]

43− '[ 'D3 rows cols channels, 'D3 (rows + 2) (cols + 2) channels, 'D3 (rows - 2) (cols - 2) chx ]

56+ = Network '[ Pad 2 2 2 2, Convolution channels chx 5 5 1 1 ]

57+ '[ 'D3 rows cols channels, 'D3 (rows + 4) (cols + 4) channels, 'D3 rows cols chx ]

4458

4559type Inception7x7 rows cols channels chx

46− = Network '[ Pad 2 2 2 2, Convolution channels chx 7 7 1 1 ]

47− '[ 'D3 rows cols channels, 'D3 (rows + 4) (cols + 4) channels, 'D3 (rows - 2) (cols - 2) chx ]

60+ = Network '[ Pad 3 3 3 3, Convolution channels chx 7 7 1 1 ]

61+ '[ 'D3 rows cols channels, 'D3 (rows + 6) (cols + 6) channels, 'D3 rows cols chx ]

4862

src/Grenade/Layers/Trivial.hs+3−2

@@ -9,7 +9,7 @@ module Grenade.Layers.Trivial (

99

1010import Data.Serialize

1111

12−import Grenade.Core.Network

12+import Grenade.Core

1313

1414-- | A trivial layer.

1515data Trivial = Trivial

@@ -25,5 +25,6 @@ instance UpdateLayer Trivial where

2525 createRandom = return Trivial

2626

2727instance (a ~ b) => Layer Trivial a b where

28− runForwards _ = id

28+ type Tape Trivial a b = ()

29+ runForwards _ a = ((), a)

2930 runBackwards _ _ y = ((), y)

src/Grenade/Recurrent.hs+4−6

@@ -1,9 +1,7 @@

11module Grenade.Recurrent (

2− module X

2+ module Grenade.Recurrent.Core

3+ , module Grenade.Recurrent.Layers

34 ) where

45

5−import Grenade.Recurrent.Core.Layer as X

6−import Grenade.Recurrent.Core.Network as X

7−import Grenade.Recurrent.Core.Runner as X

8−import Grenade.Recurrent.Layers.BasicRecurrent as X

9−import Grenade.Recurrent.Layers.LSTM as X

6+import Grenade.Recurrent.Core

7+import Grenade.Recurrent.Layers

src/Grenade/Recurrent/Core.hs+6−3

@@ -1,6 +1,9 @@

11module Grenade.Recurrent.Core (

2− module X

2+ module Grenade.Recurrent.Core.Layer

3+ , module Grenade.Recurrent.Core.Network

4+ , module Grenade.Recurrent.Core.Runner

35 ) where

46

5−import Grenade.Recurrent.Core.Layer as X

6−import Grenade.Recurrent.Core.Network as X

7+import Grenade.Recurrent.Core.Layer

8+import Grenade.Recurrent.Core.Network

9+import Grenade.Recurrent.Core.Runner

src/Grenade/Recurrent/Core/Network.hs+1−5

@@ -1,4 +1,3 @@

1−{-# LANGUAGE CPP #-}

21{-# LANGUAGE DataKinds #-}

32{-# LANGUAGE GADTs #-}

43{-# LANGUAGE TypeOperators #-}

@@ -10,10 +9,7 @@

109{-# LANGUAGE RankNTypes #-}

1110{-# LANGUAGE BangPatterns #-}

1211{-# LANGUAGE ScopedTypeVariables #-}

13−

14−#if __GLASGOW_HASKELL__ < 800

15−{-# OPTIONS_GHC -fno-warn-incomplete-patterns #-}

16−#endif

12+{-# LANGUAGE UndecidableInstances #-}

1713

1814module Grenade.Recurrent.Core.Network (

1915 Recurrent

src/Grenade/Recurrent/Core/Runner.hs−5

@@ -3,16 +3,11 @@

33{-# LANGUAGE DataKinds #-}

44{-# LANGUAGE ScopedTypeVariables #-}

55{-# LANGUAGE TypeOperators #-}

6−{-# LANGUAGE CPP #-}

76{-# LANGUAGE TypeFamilies #-}

87{-# LANGUAGE FlexibleContexts #-}

98{-# LANGUAGE RankNTypes #-}

109{-# LANGUAGE RecordWildCards #-}

1110

12−#if __GLASGOW_HASKELL__ < 800

13−{-# OPTIONS_GHC -fno-warn-incomplete-patterns #-}

14−#endif

15−

1611module Grenade.Recurrent.Core.Runner (

1712 trainRecurrent

1813 , runRecurrent

src/Grenade/Recurrent/Layers.hsnew file+7

@@ -0,0 +1,7 @@

1+module Grenade.Recurrent.Layers (

2+ module Grenade.Recurrent.Layers.BasicRecurrent

3+ , module Grenade.Recurrent.Layers.LSTM

4+ ) where

5+

6+import Grenade.Recurrent.Layers.BasicRecurrent

7+import Grenade.Recurrent.Layers.LSTM

src/Grenade/Recurrent/Layers/BasicRecurrent.hs−5

@@ -1,4 +1,3 @@

1−{-# LANGUAGE CPP #-}

21{-# LANGUAGE DataKinds #-}

32{-# LANGUAGE GADTs #-}

43{-# LANGUAGE RecordWildCards #-}

@@ -8,10 +7,6 @@

87{-# LANGUAGE FlexibleContexts #-}

98{-# LANGUAGE UndecidableInstances #-}

109

11−-- GHC 7.10 doesn't see recurrent run functions as total.

12−#if __GLASGOW_HASKELL__ < 800

13−{-# OPTIONS_GHC -fno-warn-incomplete-patterns #-}

14−#endif

1510module Grenade.Recurrent.Layers.BasicRecurrent (

1611 BasicRecurrent (..)

1712 , randomBasicRecurrent

src/Grenade/Recurrent/Layers/LSTM.hs−6

@@ -1,5 +1,4 @@

11{-# LANGUAGE BangPatterns #-}

2−{-# LANGUAGE CPP #-}

32{-# LANGUAGE DataKinds #-}

43{-# LANGUAGE GADTs #-}

54{-# LANGUAGE RankNTypes #-}

@@ -11,11 +10,6 @@

1110{-# LANGUAGE ViewPatterns #-}

1211{-# LANGUAGE ScopedTypeVariables #-}

1312

14−-- GHC 7.10 doesn't see recurrent run functions as total.

15−#if __GLASGOW_HASKELL__ < 800

16−{-# OPTIONS_GHC -fno-warn-incomplete-patterns #-}

17−#endif

18−

1913module Grenade.Recurrent.Layers.LSTM (

2014 LSTM (..)

2115 , LSTMWeights (..)

test/Test/Grenade/Layers/PadCrop.hs+9

@@ -30,6 +30,15 @@ prop_pad_crop =

3030 (_ , grad) = runBackwards net tapes d

3131 in d ~~~ res .&&. grad ~~~ d

3232

33+prop_pad_crop_2d :: Property

34+prop_pad_crop_2d =

35+ let net :: Network '[Pad 2 3 4 6, Crop 2 3 4 6] '[ 'D2 7 9, 'D2 16 15, 'D2 7 9 ]

36+ net = Pad :~> Crop :~> NNil

37+ in gamble genOfShape $ \(d :: S ('D2 7 9)) ->

38+ let (tapes, res) = runForwards net d

39+ (_ , grad) = runBackwards net tapes d

40+ in d ~~~ res .&&. grad ~~~ d

41+

3342(~~~) :: S x -> S x -> Bool

3443(S1D x) ~~~ (S1D y) = norm_Inf (x - y) < 0.00001

3544(S2D x) ~~~ (S2D y) = norm_Inf (x - y) < 0.00001