The tensorflow-haskell package provides Haskell bindings to TensorFlow.
This is not an official Google product.
https://tensorflow.github.io/haskell/haddock/
TensorFlow.Core is a good place to start.
Neural network model for the MNIST dataset: code
Toy example of a linear regression model (full code):
import Control.Monad (replicateM, replicateM_)
import System.Random (randomIO)
import Test.HUnit (assertBool)
import qualified TensorFlow.Core as TF
import qualified TensorFlow.GenOps.Core as TF
import qualified TensorFlow.Minimize as TF
import qualified TensorFlow.Ops as TF hiding (initializedVariable)
import qualified TensorFlow.Variable as TF
main :: IO ()
main = do
-- Generate data where `y = x*3 + 8`.
xData <- replicateM 100 randomIO
let yData = [x*3 + 8 | x <- xData]
-- Fit linear regression model.
(w, b) <- fit xData yData
assertBool "w == 3" (abs (3 - w) < 0.001)
assertBool "b == 8" (abs (8 - b) < 0.001)
fit :: [Float] -> [Float] -> IO (Float, Float)
fit xData yData = TF.runSession $ do
-- Create tensorflow constants for x and y.
let x = TF.vector xData
y = TF.vector yData
-- Create scalar variables for slope and intercept.
w <- TF.initializedVariable 0
b <- TF.initializedVariable 0
-- Define the loss function.
let yHat = (x `TF.mul` TF.readValue w) `TF.add` TF.readValue b
loss = TF.square (yHat `TF.sub` y)
-- Optimize with gradient descent.
trainStep <- TF.minimizeWith (TF.gradientDescent 0.001) loss [w, b]
replicateM_ 1000 (TF.run trainStep)
-- Return the learned parameters.
(TF.Scalar w', TF.Scalar b') <- TF.run (TF.readValue w, TF.readValue b)
return (w', b')Note: building this repository with stack requires version 2.3.1 or newer.
Check your stack version with stack --version in a terminal.
As an expedient we use docker for building. Once you have docker working, the following commands will compile and run the tests.
git clone --recursive https://github.com/tensorflow/haskell.git tensorflow-haskell
cd tensorflow-haskell
docker build -t tensorflow/haskell:2.12.0 docker
# TODO: move the setup step to the docker script.
stack --docker setup
stack --docker test
There is also a demo application:
cd tensorflow-mnist
stack --docker build --exec Main
If you want to use GPU you can do:
IMAGE_NAME=tensorflow/haskell:2.12.0-gpu
docker build -t $IMAGE_NAME docker/gpu
# TODO: move the setup step to the docker script.
stack --docker --docker-image=$IMAGE_NAME setup
stack --docker --docker-image=$IMAGE_NAME test
See Nvidia docker 2 install instructions
stack --docker --docker-image=$IMAGE_NAME setup
stack --docker --docker-run-args "--runtime=nvidia" --docker-image=$IMAGE_NAME test
Stack needs to use nvidia-docker instead of the normal docker for GPU support. We must wrap 'docker' with a script. This script will shadow the normal docker command.
ln -s `pwd`/tools/nvidia-docker-wrapper.sh <somewhere in your path>/docker
stack --docker --docker-image=$IMAGE_NAME setup
stack --docker --docker-image=$IMAGE_NAME test
Run the install_macos_dependencies.sh
script in the tools/ directory. The script installs dependencies
via Homebrew and then downloads and installs the TensorFlow
library on your machine under /usr/local.
After running the script to install system dependencies, build the project with stack:
stack test
The stack.yaml file describes a NixOS environment containing the necessary
dependencies. To build, run:
$ stack --nix build
Xiaokui Shu (@subbyte) maintains separate instructions for installation on CentOS.
https://github.com/helq/tensorflow-haskell-deptyped is experimenting with using dependent types to statically validate tensor shapes. May be merged with this repository in the future.
Example:
{-# LANGUAGE DataKinds, ScopedTypeVariables #-}
import Data.Maybe (fromJust)
import Data.Vector.Sized (Vector, fromList)
import TensorFlow.DepTyped
test :: IO (Vector 8 Float)
test = runSession $ do
(x :: Placeholder "x" '[4,3] Float) <- placeholder
let elems1 = fromJust $ fromList [1,2,3,4,1,2]
elems2 = fromJust $ fromList [5,6,7,8]
(w :: Tensor '[3,2] '[] Build Float) = constant elems1
(b :: Tensor '[4,1] '[] Build Float) = constant elems2
y = (x `matMul` w) `add` b -- y shape: [4,2] (b shape is [4.1] but `add` broadcasts it to [4,2])
let (inputX :: TensorData "x" [4,3] Float) =
encodeTensorData . fromJust $ fromList [1,2,3,4,1,0,7,9,5,3,5,4]
runWithFeeds (feed x inputX :~~ NilFeedList) y
main :: IO ()
main = test >>= printThis project is licensed under the terms of the Apache 2.0 license.