Skip to content
This repository has been archived by the owner on Jul 22, 2024. It is now read-only.
/ mimkl Public archive

Matrix-Induced Multiple Kernel Learning

License

Notifications You must be signed in to change notification settings

IBM/mimkl

Repository files navigation

Build Status CII Best Practices

mimkl

A C++ library for Matrix-Induced Multiple Kernel Learning (MIMKL) with python bindings. The library implements unsupervised and supervised multiple kernel learning algorithms and the most common kernel functions as well as matrix-inducing versions thereof. The python wrapper exposes classes with interfaces that allow a seamless integration with scikit-learn.

The implementation has been used to generate all the results for the paper "Pathway-Induced Multiple Kernel Learning".

Requirements

  • C++14 capable C++ compiler
  • cmake (>3.0.2)
  • Python

pymimkl

pymimkl is the python package for matrix induced multiple kernel learning. It uses bindings from compiled mimkl (C++) (see Building) and provides scikit-learn like classifiers.

Installation

Install directly from git. This builds the mimkl project and the pymimkl bindings and might take some time.

pip install git+ssh://git@github.com/IBM/mimkl.git

Optimize the build by passing additional C++ flags. For example to enable parallelism (compiler must support OpenMP):

MIMKL_CXX_FLAGS="-fopenmp" pip install git+ssh://git@github.com/IBM/mimkl.git

Enable architecture optimizations (i386):

MIMKL_CXX_FLAGS="-march=native" pip install git+ssh://git@github.com/IBM/mimkl.git

Enable both:

MIMKL_CXX_FLAGS="-march=native -fopenmp" pip install git+ssh://git@github.com/IBM/mimkl.git

Docker

Pull the image from DockerHub:

docker pull tsenit/mimkl

Or directly build it from source:

docker-compose -f docker/docker-compose.yml build

Run the container:

docker run -it tsenit/mimkl /bin/bash

Development

Clone the repository:

git clone https://github.com/IBM/mimkl

Intialize and update the submodules

cd mimkl
git submodule init
git submodule update

Install pymimkl with pip.

pip install .
# to persist the created build_ folder for faster rebuilding and C++ testing:
# pip install -e .

Same as before to optimize the build.

MIMKL_CXX_FLAGS="-march=native -fopenmp" pip install .

run tests with

python setup.py test

Building

Clone the repository:

git clone https://github.com/IBM/mimkl

Intialize and update the submodules

cd mimkl
git submodule init
git submodule update

Create a build folder (in source build):

mkdir build
cd build

Generate the building files:

cmake ..
# unfortuantely, cmake is not aware of virtual environments. The fix is
# cmake -DPYTHON_EXECUTABLE=$(command -v python) ..
# https://github.com/pybind/pybind11/issues/99

Compile the code:

make

Test it:

make test
# make test CTEST_OUTPUT_ON_FAILURE=TRUE

Take note that the python tests have additional requirements (e.g. scipy).

References

Publications

EasyMKL: a scalable multiple kernel learning algorithm

Unsupervised multiple kernel learning for heterogeneous data integration ("UMKLKNN" im mimkl)

Code

eigen http://eigen.tuxfamily.org/index.php?title=Main_Page

dlib http://dlib.net

pybind11 https://github.com/pybind/pybind11

Uses of mimkl

PIMKL: Pathway-Induced Multiple Kernel Learning https://rdcu.be/bBN6U
for which mimkl was developed.

PIMKL algorithm

About

Matrix-Induced Multiple Kernel Learning

Resources

License

Stars

Watchers

Forks

Packages

No packages published