Skip to content
Homomorphic machine learning
Haskell C++ Shell Java C R Other
Branch: master
Clone or download
mikeizbicki Merge pull request #79 from feuerbach/master
Prevent haddock from crashing
Latest commit c6c0f5e May 29, 2016
Type Name Latest commit message Commit time
Failed to load latest commit information.
bench/allknn updated travis with generic install script Jul 22, 2015
cbits added earth mover distance; streamlined parallelism; updated tests Jan 2, 2015
datasets @ ea5a06d
executables removed shuffling of data set Aug 7, 2015
install updated travis with generic install script Jul 22, 2015
src/HLearn Prevent haddock from crashing May 8, 2016
subhask @ b5433b9
test updated travis with generic install script Jul 22, 2015
.travis.yml updated travis with generic install script Jul 22, 2015
LICENSE Changed license to BSD3 Apr 24, 2013 Fixed links to Wikipedia Apr 28, 2016
stack.yaml Make stack use subhask submodule. May 5, 2016


HLearn is a high performance machine learning library written in Haskell. For example, it currently has the fastest nearest neighbor implementation for arbitrary metric spaces (see this blog post).

HLearn is also a research project. The research goal is to discover the "best possible" interface for machine learning. This involves two competing demands: The library should be as fast as low-level libraries written in C/C++/Fortran/Assembly; but it should be as flexible as libraries written in high level languages like Python/R/Matlab. Julia is making amazing progress in this direction, but HLearn is more ambitious. In particular, HLearn's goal is to be faster than the low level languages and more flexible than the high level languages.

To achieve this goal, HLearn uses a very different interface than standard learning libraries. The H in HLearn stands for three separate concepts that are fundamental to HLearn's design:

  1. The H stands for Haskell. Machine learning is about estimating functions from data, so it makes sense that a functional programming language would be well suited for machine learning. But Functional programming languages are not widely used in machine learning because they traditionally lack strong support for the fast numerical computations required for learning algorithms. HLearn uses the SubHask library to get this fast numeric support in Haskell. The two libraries are being developed in tandem with each other.
  1. The H stands for Homomorphisms. Homomorphisms are a fundamental concept in abstract algebra, and HLearn exploits the algebraic structures inherrent in learning systems. The following table gives a brief overview of what these structures give us:

    Structure What we get
    Monoid parallel batch training
    Monoid online training
    Monoid fast cross-validation
    Abelian group "untraining" of data points
    Abelian group more fast cross-validation
    R-Module weighted data points
    Vector space fractionally weighted data points
    Functor fast simple preprocessing of data
    Monad fast complex preprocessing of data
  2. The H stands for the History monad. One of the most difficult tasks of developing a new learning algorithm is debugging the optimization procedure. There has previously been essentially no work on making this debugging process easier, and the History monad tries to solve this problem. It lets you thread debugging information throughout the optimization code without modifying the original code. Furthermore, there is no runtime overhead associated with this technique.

The downside of HLearn's ambition is that it currently does not implement many of the popular machine learning techniques.

More Documentation

Due to the rapid pace of development, HLearn's documentation is sparse. That said, the examples folder is a good place to start. The haddock documentation embedded within the code is decent; but unfortunately, hackage is unable to compile the haddocks because it uses an older version of GHC.

HLearn has several academic papers:

There are also a number of blog posts on my personal website. Unfortunately, they are mostly out of date with the latest version of HLearn. They might help you understand some of the main concepts in HLearn, but the code they use won't work at all.


I'd love to have you contribute, and I'd be happy to help you get started! Just create an issue to let me know you're interested and we can work something out.

You can’t perform that action at this time.