mathflow: Dependently typed tensorflow modeler

[ bsd3, library, math ] [ Propose Tags ]


Automatic Flags

Use Python for test


Use -f <flag> to enable a flag, or -f -<flag> to disable that flag. More info


Maintainer's Corner

Package maintainers

For package maintainers and hackage trustees


  • No Candidates
Versions [RSS]
Dependencies base (>=4.7 && <5), process, singletons, template-haskell [details]
License BSD-3-Clause
Copyright 2017 Junji Hashimoto
Author Junji Hashimoto
Category Math
Home page
Source repo head: git clone
Uploaded by junjihashimoto at 2017-08-08T01:36:13Z
Reverse Dependencies 1 direct, 0 indirect [details]
Downloads 846 total (3 in the last 30 days)
Rating (no votes yet) [estimated by Bayesian average]
Your Rating
  • λ
  • λ
  • λ
Status Docs available [build log]
Last success reported on 2017-08-08 [all 1 reports]

Readme for mathflow-

[back to package description]

mathflow(Dependently typed tensorflow modeler)

Hackage version Build Status

This package provides a model of tensor-operations. The model is independent from tensorflow-binding of python and haskell, though this package generates python-code. tensor's dimensions and constraints are described by dependent types. The tensor-operations are based on tensorflow-api. Currently the model can be translated into python-code. To write this package, I refer to this neural network document and singletons.


Install tensorflow of python and this package.

> sudo apt install python3 python3-pip
> pip3 install -U pip
> pip3 install tensorflow
> git clone
> cd mathflow
> stack install


About model

Model has a type of Tensor (dimensions:[Nat]) value-type output-type.

  • dimensions are tensor-dimensions.
  • value-type is a value type like Integer or Float of tensorflow-data-types.
  • output-type is a type of code which this package generates. PyString-type is used for generating python-code.

This package makes tensorflow-graph from the mode. The model's endpoint is always a tensor-type.

At first write graph by using arithmetic operators like (+,-,,/), % (which is matrix multiply) and tensorflow-functions. Mathflow.{TF,TF.NN,TF.Train} packages define Tensorflow-functions.

A example is below.

testMatMul :: Tensor '[2,1] Int PyString
testMatMul = 
  let n1 = (Tensor "tf.constant([[2],[3]])") :: Tensor '[2,1] Int PyString
      n2 = (Tensor "tf.constant([[2,0],[0,1]])") :: Tensor '[2,2] Int PyString
      y = (n2 %* n1) :: Tensor '[2,1] Int PyString
  in y

Create model and run it

Write tensorflow-model.

testMatMul :: Tensor '[2,1] Int PyString
testMatMul = 
  let n1 = (Tensor "tf.constant([[2],[3]])") :: Tensor '[2,1] Int PyString
      n2 = (Tensor "tf.constant([[2,0],[0,1]])") :: Tensor '[2,2] Int PyString
      y = n2 %* n1 :: Tensor '[2,1] Int PyString
  in y

Run the model. This run function generates python-code and excecute the code by python.

main = do
  (retcode,stdout,stderr) <- run testMatMul
  print stdout