vdt / jnumpy

Writing Python C extensions in Julia within 5 minutes.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

JNumPy: writing high-performance C extensions for Python in minutes

Install JNumPy


  • Python >= 3.7

You can install the Python package jnumpy with the following command:

pip install julia-numpy.

Note that JNumPy will install julia in JNUMPY_HOME for you, if there is no Julia installation available.

  1. write and export julia functions in file example.jl

    module example
    using TyPython
    using TyPython.CPython
    @export_py function mat_mul(a::StridedArray, b::StridedArray)::StridedArray
        return a * b
    function init()
        @export_pymodule example begin
            jl_mat_mul = Pyfunc(jl_mat_mul)
  2. initialize and import the julia functions in Python

    from jnumpy import init_jl, exec_julia, include_src
    import jnumpy as np
    include_src("example.jl", __file__)
    from example import jl_mat_mul
    x = np.array([[1,2],[3,4]])
    y = np.array([[4,3],[2,1]])
    jl_mat_mul(x, y)
    # array([[ 8,  5],
    #       [20, 13]])

Environment Variables


    The home directory for JNumPy-specific settings. The default value is ~/.jnumpy. JNumPy runs julia in a default environment ($JNUMPY_HOME/envs/default). In case that you don't have a julia executable, JNumPy installs julia into $JNUMPY_HOME using jill.py.


    The path of the julia executable in use.


    Command-line options when launching julia. If you want to use a custom environment, you could set --project=<dir>. TYPY_JL_OPTS is the same as those arguments passed to julia.


There are several examples presented in the demo directory. Those examples are standalone Python packages created using JNumPy, and can be imported if you have JNumPy installed.

  • demo/basic: a tiny Python package to give an example of how to use JNumPy.

  • demo/kmeans: a tiny Python package wrapping ParallelKMeans.jl. It produces a 10x performance gain against Scikit-Learn.

  • demo/fft: a tiny Python package wrapping FFTW.jl. It allows users to access FFT plans for accelerating FFTs.


Open-source contributions are kindly accepted and appreciated including bug reports, documentations, pull requests, and general suggestions.


Writing Python C extensions in Julia within 5 minutes.

License:MIT License


Language:Julia 81.6%Language:Python 18.3%Language:Shell 0.1%