Numba tutorial materials for Scipy 2016
Switch branches/tags
Clone or download
gforsyth change jit failure example
just using basic string addition now. this should never work in LLVM
(one would think) so we don't need to keep checking if the examples
still work. Also, this flows a little more naturally from the previous
example, instead of an arbitrary one.
Latest commit 214cfd8 Apr 14, 2017

Numba: Tell those C++ bullies to get lost

This is the repository for the Scipy 2016 tutorial. The tutorial will be presented as a set of Jupyter notebooks with exercises sprinkled throughout.

  1. Installation
  2. Optional extras
  3. Check your installation

Installation Instructions

We strongly, strongly, strongly encourage you to use conda to install the required packages for this tutorial. There are non-Python dependencies required that make manual installation or installing with pip very involved.

Note also that this tutorial is written for Python 3.5. Most things will still work on Python 3.4. No guarantees of any kind are made that it will be compatible with Python 2.

Regarding matplotlib

This tutorial uses the Viridis colormap pretty much everywhere we can use a colormap. This colormap was first made available in matplotlib 1.5.0. Please upgrade if you have an earlier version installed.

Installing with conda

Option a) Create a new environment

Download the environment.yml file in the root of this repository, e.g.


and then create the environment with

conda env create -f environment.yml

This will create a conda environment named numbatutorial with all of the required packages.

You can activate the environment with

source activate numbatutorial

or on Windows:

activate numbatutorial

Option b) Install the required packages

conda install jupyter ipython numpy numba line_profiler matplotlib
pip install line_profiler

Note: Do not use conda to install line_profiler; the version available in conda default channels is out of date.

Installing with pip

To install (specifically) Numba using pip, you need to have LLVM 3.7 installed on your machine with both libraries and header files.


Ubuntu / Debian

You should be able to do a

sudo apt-get install llvm-3.7-dev

You may also need to install libedit-dev


You can follow instructions here for getting LLVM installed on Windows.


Install XCode which includes LLVM

Install llvmlite

If your llvm-config (or llvm-config.exe) file is in a non-standard location, set the LLVM_CONFIG environment variable to point at the llvm-config binary.


pip install llvmlite

If that installed successfully then you can continue to install the rest of the dependencies (which are must less fussy)

Install everything else

pip install numpy matplotlib jupyter ipython numba line_profiler 


pip install -r requirements.txt

Optional extras

No hands-on work requires these, but if you want to play with some of the examples. If you installed using either environments.yml or requirements.txt these are already installed.

conda install cython dask
pip install cython dask

We recommend you also install the Jupyter notebook extensions.

pip install --user

Once they are installed, start a notebook server

jupyter notebook

and (assuming port 8888) navigate to http://localhost:8888/nbextensions where you can choose which extensions to enable. One that is helpful (for us!) when using Numba in the notebook is the Skip-Traceback extension. You're welcome to enable whichever extensions you like (we're also fans of Codefolding and the Comment/Uncomment Hotkey).

Check Installation

Once you have downloaded all of the requires libraries/packages, you can run the script to confirm that everything is working as expected. Either download the file directly or clone this repository and then run


Video of the live tutorial

Check out the video of the live tutorial at SciPy 2016 (filmed Monday 11 July).