Tutorial "A tour of Python profiling and optimization"

Tue August 28, 14:00, A. Ingargiola, EuroScipy 2018

To run on the cloud (no installation) use

WARNING During the EuroScipy 2018 Tutorial a local installation is preferable.

Index

Tips

Optimization Zen and Numba vs Cython comparison

Notebooks:

Part I: Optimizing Python code.ipynb
Part II: Parallelism with Joblib.ipynb

Requirements

You will need python 3.6 or later with the following packages:

cython line_profiler numba numexpr numpy ipykernel pandas joblib matplotlib seaborn

available through PyPI or conda/conda-forge. You will also need:

randomgen

currently only available from PyPI.

NOTE for Windows: You need a cython installation, including a working compiler. For this, you need to download Visual Studio 2015 (several GB). If you know/find an easier solution I would like to know. Thanks.

Installation

These instructions require conda. You also need to add conda-forge to the conda channels.

Cloning (reproducible)

Linux and macOS users can use the following command:

$ conda env create -f environment.yml

to create the py36_optimize environment for this tutorial. The yml file is:

environment_macos.yml for macOS
binder/environment.yml for Linux

From scratch (non-reproducible)

This creates the environment from scratch, using the most updated packages (non-reproducible):

$ conda create -n py36_optimize python=3.6 cython line_profiler numba numexpr numpy ipykernel pandas joblib matplotlib seaborn
$ conda activate py36_optimize
$ pip install randomgen  # no conda package yet

Installing the IPython kernel

After installing the environment you need to make it available to Jupyter:

$ conda activate py36_optimize
$ python -m ipykernel install --name py36_optimize --display-name "Python 3.6 (py36_optimize)"