This repository contains a Python 2.7 implementation of the normalized FSIC (NFSIC) test as described in our paper

An Adaptive Test of Independence with Analytic Kernel Embeddings
Wittawat Jitkrittum, Zoltán Szabó, Arthur Gretton
arXiv, 2016. 

1. Make sure that you have a complete Scipy stack installed. One way to guarantee this is to install it using Anaconda with Python 2.7, which is also the environment we used to develop this package. Make sure to use Python 2.7.
2. Clone or download this repository. You will get a folder with name fsic-test.
3. Add the path to the folder to Python's search path i.e., to PYTHONPATH global variable. See, for instance, this page on stackoverflow on how to do this in Linux. See here for Windows.
4. Check that indeed the package is in the search path by openning a new Python shell, and issuing import fsic (fsic is the name of our Python package). If there is no import error, the installation is completed.

We rely on the following Python packages during development. Please make sure that you use the packages with the specified version numbers or newer.

numpy==1.11.0
matplotlib==1.5.1
scipy==0.18.0
theano==0.8.

Note that theano is not enabled in Anaconda by default. See this page for how to install it.

To get started, check demo_nfsic.ipynb which will guide you through from the beginning. There are many Jupyter notebooks in ipynb folder. Be sure to check them if you would like to explore more.

MIT license.

If you have questions or comments about anything regarding this work or code, please do not hesitate to contact Wittawat Jitkrittum.