grenade/grenade.cabal

name:                  grenade
version:               0.1.0
license:               BSD2
license-file:          LICENSE
author:                Huw Campbell <huw.campbell@gmail.com>
maintainer:            Huw Campbell <huw.campbell@gmail.com>
copyright:             (c) 2016-2017 Huw Campbell.
synopsis:              Practical Deep Learning in Haskell
category:              AI, Machine Learning
cabal-version:         >= 1.8
build-type:            Simple
description:
    Grenade is a composable, dependently typed, practical, and fast
    recurrent neural network library for precise specifications and
    complex deep neural networks in Haskell.
    .
    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 descent for learning.
    .
    Documentation and examples are available on github
    <https://github.com/HuwCampbell/grenade>.

extra-source-files:
  README.md
  cbits/im2col.h
  cbits/im2col.c
  cbits/gradient_descent.h
  cbits/gradient_descent.c
  cbits/pad.h
  cbits/pad.c

source-repository head
  type: git
  location: https://github.com/HuwCampbell/grenade.git

library
  build-depends:
                       base                            >= 4.8         && < 5
                     , bytestring                      == 0.10.*
                     , containers                      >= 0.5         && < 0.6
                     , cereal                          >= 0.5         && < 0.6
                     , deepseq                         >= 1.4         && < 1.5
                     , hmatrix                         == 0.18.*
                     , MonadRandom                     >= 0.4         && < 0.6
                     , primitive                       >= 0.6         && < 0.7
                     , singletons                      >= 2.1         && < 2.4
                     , vector                          >= 0.11        && < 0.13

  ghc-options:
                       -Wall
  hs-source-dirs:
                       src

  if impl(ghc < 8.0)
    ghc-options:       -fno-warn-incomplete-patterns


  exposed-modules:
                       Grenade
                       Grenade.Core
                       Grenade.Core.Layer
                       Grenade.Core.LearningParameters
                       Grenade.Core.Network
                       Grenade.Core.Runner
                       Grenade.Core.Shape

                       Grenade.Layers
                       Grenade.Layers.Concat
                       Grenade.Layers.Convolution
                       Grenade.Layers.Crop
                       Grenade.Layers.Deconvolution
                       Grenade.Layers.Dropout
                       Grenade.Layers.Elu
                       Grenade.Layers.FullyConnected
                       Grenade.Layers.Inception
                       Grenade.Layers.Logit
                       Grenade.Layers.Merge
                       Grenade.Layers.Pad
                       Grenade.Layers.Pooling
                       Grenade.Layers.Relu
                       Grenade.Layers.Reshape
                       Grenade.Layers.Sinusoid
                       Grenade.Layers.Softmax
                       Grenade.Layers.Tanh
                       Grenade.Layers.Trivial

                       Grenade.Layers.Internal.Convolution
                       Grenade.Layers.Internal.Pad
                       Grenade.Layers.Internal.Pooling
                       Grenade.Layers.Internal.Update

                       Grenade.Recurrent

                       Grenade.Recurrent.Core
                       Grenade.Recurrent.Core.Layer
                       Grenade.Recurrent.Core.Network
                       Grenade.Recurrent.Core.Runner

                       Grenade.Recurrent.Layers
                       Grenade.Recurrent.Layers.BasicRecurrent
                       Grenade.Recurrent.Layers.ConcatRecurrent
                       Grenade.Recurrent.Layers.LSTM

                       Grenade.Utils.OneHot

  includes:            cbits/im2col.h
                       cbits/gradient_descent.h
                       cbits/pad.h
  c-sources:           cbits/im2col.c
                       cbits/gradient_descent.c
                       cbits/pad.c

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

test-suite test
  type:                exitcode-stdio-1.0

  main-is:             test.hs

  ghc-options:         -Wall -threaded -O2

  hs-source-dirs:
                       test


  other-modules:       Test.Hedgehog.Compat
                       Test.Hedgehog.Hmatrix
                       Test.Hedgehog.TypeLits

                       Test.Grenade.Network
                       Test.Grenade.Layers.Convolution
                       Test.Grenade.Layers.FullyConnected
                       Test.Grenade.Layers.Nonlinear
                       Test.Grenade.Layers.PadCrop
                       Test.Grenade.Layers.Pooling
                       Test.Grenade.Layers.Internal.Convolution
                       Test.Grenade.Layers.Internal.Pooling
                       Test.Grenade.Layers.Internal.Reference

                       Test.Grenade.Recurrent.Layers.LSTM
                       Test.Grenade.Recurrent.Layers.LSTM.Reference

  if impl(ghc < 8.0)
    ghc-options:       -fno-warn-incomplete-patterns

  build-depends:
                       base                            >= 4.8         && < 5
                     , grenade
                     , hedgehog                        >= 0.5         && < 0.6
                     , hmatrix
                     , mtl
                     , singletons
                     , text                            == 1.2.*
                     , typelits-witnesses
                     , transformers
                     , constraints
                     , MonadRandom
                     , random
                     , ad
                     , reflection
                     , vector


benchmark bench
  type:                exitcode-stdio-1.0

  main-is:             bench.hs

  ghc-options:         -Wall -threaded -O2

  hs-source-dirs:
                       bench

  build-depends:
                       base                            >= 3          && < 5
                     , bytestring                      == 0.10.*
                     , criterion                       >= 1.1        && < 1.3
                     , grenade
                     , hmatrix

benchmark bench-lstm
  type:                exitcode-stdio-1.0

  main-is:             bench-lstm.hs

  ghc-options:         -Wall -threaded -O2

  hs-source-dirs:
                       bench

  build-depends:
                       base                            >= 3          && < 5
                     , bytestring                      == 0.10.*
                     , criterion                       == 1.1.*
                     , grenade
                     , hmatrix
Initial commit 2016-06-23 15:12:57 +03:00			`name: grenade`
Bump version for release 2017-04-12 10:15:24 +03:00			`version: 0.1.0`
grenade.cabal: Specify BSD2 to match the LICENSE file 2016-06-28 15:08:45 +03:00			`license: BSD2`
Sketch out travis 2017-02-07 10:30:10 +03:00			`license-file: LICENSE`
Replace licence in cabal file 2016-06-28 02:31:44 +03:00			`author: Huw Campbell <huw.campbell@gmail.com>`
			`maintainer: Huw Campbell <huw.campbell@gmail.com>`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`copyright: (c) 2016-2017 Huw Campbell.`
Docs 2017-04-07 15:50:28 +03:00			`synopsis: Practical Deep Learning in Haskell`
Update categories 2017-04-12 09:06:56 +03:00			`category: AI, Machine Learning`
Initial commit 2016-06-23 15:12:57 +03:00			`cabal-version: >= 1.8`
			`build-type: Simple`
Docs 2017-04-07 15:50:28 +03:00			`description:`
			`Grenade is a composable, dependently typed, practical, and fast`
Update categories 2017-04-12 09:06:56 +03:00			`recurrent neural network library for precise specifications and`
			`complex deep neural networks in Haskell.`
Docs 2017-04-07 15:50:28 +03:00			`.`
			`Grenade provides an API for composing layers of a neural network`
Update categories 2017-04-12 09:06:56 +03:00			`into a sequence parallel graph in a type safe manner; running`
			`networks with reverse automatic differentiation to calculate their`
Spelling: decend/decent -> descend/descent 2017-09-25 12:56:42 +03:00			`gradients; and applying gradient descent for learning.`
Docs 2017-04-07 15:50:28 +03:00			`.`
			`Documentation and examples are available on github`
			`<https://github.com/HuwCampbell/grenade>.`
Initial commit 2016-06-23 15:12:57 +03:00
Add stupidly fast im2col in c 2016-12-12 15:35:00 +03:00			`extra-source-files:`
Update categories 2017-04-12 09:06:56 +03:00			`README.md`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`cbits/im2col.h`
			`cbits/im2col.c`
Spelling: decend/decent -> descend/descent 2017-09-25 12:56:42 +03:00			`cbits/gradient_descent.h`
			`cbits/gradient_descent.c`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`cbits/pad.h`
			`cbits/pad.c`
Add stupidly fast im2col in c 2016-12-12 15:35:00 +03:00
Sketch out travis 2017-02-07 10:30:10 +03:00			`source-repository head`
			`type: git`
			`location: https://github.com/HuwCampbell/grenade.git`

Initial commit 2016-06-23 15:12:57 +03:00			`library`
			`build-depends:`
			`base >= 4.8 && < 5`
			`, bytestring == 0.10.*`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`, containers >= 0.5 && < 0.6`
			`, cereal >= 0.5 && < 0.6`
			`, deepseq >= 1.4 && < 1.5`
Make Pad 3D faster. Using dependent types in the deeper functions and requiring a Proxy to reach them meant we required dictionary passing to get the Nats. This made the pad and crop layers almost 1000 times slower than they should have been. 2017-02-03 13:04:27 +03:00			`, hmatrix == 0.18.*`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`, MonadRandom >= 0.4 && < 0.6`
			`, primitive >= 0.6 && < 0.7`
Allow a recurrent net to be a recurrent layer 2017-07-14 07:05:41 +03:00			`, singletons >= 2.1 && < 2.4`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`, vector >= 0.11 && < 0.13`
Initial commit 2016-06-23 15:12:57 +03:00
			`ghc-options:`
			`-Wall`
			`hs-source-dirs:`
			`src`

Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`if impl(ghc < 8.0)`
			`ghc-options: -fno-warn-incomplete-patterns`

Initial commit 2016-06-23 15:12:57 +03:00
			`exposed-modules:`
			`Grenade`
Add extra composition 2017-02-01 14:11:55 +03:00			`Grenade.Core`
			`Grenade.Core.Layer`
			`Grenade.Core.LearningParameters`
Initial commit 2016-06-23 15:12:57 +03:00			`Grenade.Core.Network`
			`Grenade.Core.Runner`
			`Grenade.Core.Shape`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00
			`Grenade.Layers`
Add c 3d pad and crop functions 2017-02-02 12:27:02 +03:00			`Grenade.Layers.Concat`
Initial commit 2016-06-23 15:12:57 +03:00			`Grenade.Layers.Convolution`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`Grenade.Layers.Crop`
Add Deconvolution layer and update the GAN example to use it This cuts the runtime by about 70% which is nice, and it's a better algorithm for it anyway. I've also refactored the Convolution layer such that there's only one actual implementation instead of two, and with that provided a few more instances for 2D and 3D shapes in and out. Update to the README and mnist show higher levels of composition. 2017-04-06 06:34:43 +03:00			`Grenade.Layers.Deconvolution`
Initial commit 2016-06-23 15:12:57 +03:00			`Grenade.Layers.Dropout`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`Grenade.Layers.Elu`
Initial commit 2016-06-23 15:12:57 +03:00			`Grenade.Layers.FullyConnected`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`Grenade.Layers.Inception`
Initial commit 2016-06-23 15:12:57 +03:00			`Grenade.Layers.Logit`
Add c 3d pad and crop functions 2017-02-02 12:27:02 +03:00			`Grenade.Layers.Merge`
Add pad and crop layers, add pad to mnist example 2016-07-01 05:16:50 +03:00			`Grenade.Layers.Pad`
Initial commit 2016-06-23 15:12:57 +03:00			`Grenade.Layers.Pooling`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`Grenade.Layers.Relu`
			`Grenade.Layers.Reshape`
added the sinusoid activation function layer 2018-01-26 12:52:41 +03:00			`Grenade.Layers.Sinusoid`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`Grenade.Layers.Softmax`
			`Grenade.Layers.Tanh`
			`Grenade.Layers.Trivial`
Initial commit 2016-06-23 15:12:57 +03:00
Tests 2016-12-08 01:03:29 +03:00			`Grenade.Layers.Internal.Convolution`
Add c 3d pad and crop functions 2017-02-02 12:27:02 +03:00			`Grenade.Layers.Internal.Pad`
Tests 2016-12-08 01:03:29 +03:00			`Grenade.Layers.Internal.Pooling`
Remove primes on shape instantiations Add singletons for Shape and remove hacks on recurrent nets Add Recurrent Nets 2016-12-20 08:31:09 +03:00			`Grenade.Layers.Internal.Update`

			`Grenade.Recurrent`

Make the recurrent api like core 2017-02-06 13:11:51 +03:00			`Grenade.Recurrent.Core`
			`Grenade.Recurrent.Core.Layer`
Remove primes on shape instantiations Add singletons for Shape and remove hacks on recurrent nets Add Recurrent Nets 2016-12-20 08:31:09 +03:00			`Grenade.Recurrent.Core.Network`
			`Grenade.Recurrent.Core.Runner`

Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`Grenade.Recurrent.Layers`
Remove primes on shape instantiations Add singletons for Shape and remove hacks on recurrent nets Add Recurrent Nets 2016-12-20 08:31:09 +03:00			`Grenade.Recurrent.Layers.BasicRecurrent`
Allow a recurrent net to be a recurrent layer 2017-07-14 07:05:41 +03:00			`Grenade.Recurrent.Layers.ConcatRecurrent`
Remove primes on shape instantiations Add singletons for Shape and remove hacks on recurrent nets Add Recurrent Nets 2016-12-20 08:31:09 +03:00			`Grenade.Recurrent.Layers.LSTM`

			`Grenade.Utils.OneHot`
Tests 2016-12-08 01:03:29 +03:00
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`includes: cbits/im2col.h`
Spelling: decend/decent -> descend/descent 2017-09-25 12:56:42 +03:00			`cbits/gradient_descent.h`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`cbits/pad.h`
			`c-sources: cbits/im2col.c`
Spelling: decend/decent -> descend/descent 2017-09-25 12:56:42 +03:00			`cbits/gradient_descent.c`
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`cbits/pad.c`
Initial commit 2016-06-23 15:12:57 +03:00
Cleanup imports and move examples to new project 2017-02-20 11:16:38 +03:00			`cc-options: -std=c99 -O3 -msse4.2 -Wall -Werror -DCABAL=1`
Initial commit 2016-06-23 15:12:57 +03:00
			`test-suite test`
			`type: exitcode-stdio-1.0`

			`main-is: test.hs`

			`ghc-options: -Wall -threaded -O2`

			`hs-source-dirs:`
			`test`

Add test files to other-modules in test suite 2017-04-12 08:08:41 +03:00
			`other-modules: Test.Hedgehog.Compat`
			`Test.Hedgehog.Hmatrix`
			`Test.Hedgehog.TypeLits`

			`Test.Grenade.Network`
			`Test.Grenade.Layers.Convolution`
			`Test.Grenade.Layers.FullyConnected`
			`Test.Grenade.Layers.Nonlinear`
			`Test.Grenade.Layers.PadCrop`
			`Test.Grenade.Layers.Pooling`
			`Test.Grenade.Layers.Internal.Convolution`
			`Test.Grenade.Layers.Internal.Pooling`
			`Test.Grenade.Layers.Internal.Reference`

			`Test.Grenade.Recurrent.Layers.LSTM`
			`Test.Grenade.Recurrent.Layers.LSTM.Reference`

Change the singleton type for shape. This allows us to more easily get a handle to the parameters of the shape, while still being pretty easy to cast to a KnownNat. We can now more easily add to the gradients test, and properly check the sizes of the head without silly hacks. 2017-04-12 04:59:40 +03:00			`if impl(ghc < 8.0)`
			`ghc-options: -fno-warn-incomplete-patterns`

Initial commit 2016-06-23 15:12:57 +03:00			`build-depends:`
			`base >= 4.8 && < 5`
			`, grenade`
Bump hedgehog and criterion 2017-12-14 08:05:37 +03:00			`, hedgehog >= 0.5 && < 0.6`
Initial commit 2016-06-23 15:12:57 +03:00			`, hmatrix`
			`, mtl`
Fix silent conversion bug in pooling layer Add quickcheck against reference implemenations for layer internals Add skolem layer generators. 2016-12-15 03:21:37 +03:00			`, singletons`
grenade.cabal: Tighten up constraints 2016-06-28 14:55:07 +03:00			`, text == 1.2.*`
Fix silent conversion bug in pooling layer Add quickcheck against reference implemenations for layer internals Add skolem layer generators. 2016-12-15 03:21:37 +03:00			`, typelits-witnesses`
Move tests to hedgehog. 2017-04-10 01:49:18 +03:00			`, transformers`
Fix silent conversion bug in pooling layer Add quickcheck against reference implemenations for layer internals Add skolem layer generators. 2016-12-15 03:21:37 +03:00			`, constraints`
			`, MonadRandom`
			`, random`
Remove primes on shape instantiations Add singletons for Shape and remove hacks on recurrent nets Add Recurrent Nets 2016-12-20 08:31:09 +03:00			`, ad`
			`, reflection`
Network scale autodiff test 2017-04-11 03:08:08 +03:00			`, vector`
Add some benchmarks 2016-12-12 02:20:40 +03:00

			`benchmark bench`
			`type: exitcode-stdio-1.0`

			`main-is: bench.hs`

			`ghc-options: -Wall -threaded -O2`

			`hs-source-dirs:`
			`bench`

			`build-depends:`
			`base >= 3 && < 5`
			`, bytestring == 0.10.*`
Bump hedgehog and criterion 2017-12-14 08:05:37 +03:00			`, criterion >= 1.1 && < 1.3`
Add some benchmarks 2016-12-12 02:20:40 +03:00			`, grenade`
			`, hmatrix`
Remove primes on shape instantiations Add singletons for Shape and remove hacks on recurrent nets Add Recurrent Nets 2016-12-20 08:31:09 +03:00
			`benchmark bench-lstm`
			`type: exitcode-stdio-1.0`

			`main-is: bench-lstm.hs`

			`ghc-options: -Wall -threaded -O2`

			`hs-source-dirs:`
			`bench`

			`build-depends:`
			`base >= 3 && < 5`
			`, bytestring == 0.10.*`
			`, criterion == 1.1.*`
			`, grenade`
			`, hmatrix`