The R version is available on CRAN and https://github.com/neurodata/r-mgc. The MATLAB version is available at https://github.com/neurodata/mgc-matlab.

mgcpy

mgcpy is a Python package containing tools for independence testing using multiscale graph correlation and other statistical tests, that is capable of dealing with high dimensional and multivariate data.

• Overview
• Documentation
• System Requirements
• Installation Guide
• Setting up the development environment
• License
• Issues

Overview

mgcpy aims to be a comprehensive independence testing package including commonly used independence tests and additional functionality such as two sample independence testing and a novel random forest-based independence test. These tests are not only included to benchmark MGC but to have a convenient location for users if they would prefer to utilize those tests instead. The package utilizes a simple class structure to enhance usability while also allowing easy extension of the package for developers. The package can be installed on all major platforms (e.g. BSD, GNU/Linux, OS X, Windows)from Python Package Index (PyPI) and GitHub.

Documenation

The official documentation with usage is at: https://mgc.neurodata.io/ ReadTheDocs: https://mgcpy.readthedocs.io/en/latest/

System Requirements

Hardware requirements

mgcpy package requires only a standard computer with enough RAM to support the in-memory operations.

Software requirements

OS Requirements

This package is supported for macOS and Linux. The package has been tested on the following systems:

• macOS: Mojave (10.14.1)
• Linux: Ubuntu 16.04

Python Dependencies

mgcpy mainly depends on the Python scientific stack.

numpy
scipy
Cython
scikit-learn
pandas
seaborn

Installation Guide:

Install from PyPi

pip3 install mgcpy

Install from Github

git clone https://github.com/neurodata/mgcpy
cd mgcpy
python3 setup.py install

• sudo, if required
• python3 setup.py build_ext --inplace # for cython, if you want to test in-place, first execute this

Setting up the development environment:

• To build image and run from scratch:

  • Install docker
  • Build the docker image, docker build -t mgcpy:latest .
    • This takes 10-15 mins to build
  • Launch the container to go into mgcpy's dev env, docker run -it --rm --name mgcpy-env mgcpy:latest

• Pull image from Dockerhub and run:

  • docker pull tpsatish95/mgcpy:latest or docker pull tpsatish95/mgcpy:development
  • docker run -it --rm -p 8888:8888 --name mgcpy-env tpsatish95/mgcpy:latest or docker run -it --rm -p 8888:8888 --name mgcpy-env tpsatish95/mgcpy:development

• To run demo notebooks (from within Docker):

  • cd demos
  • jupyter notebook --ip 0.0.0.0 --no-browser --allow-root
  • Then copy the url it generates, it looks something like this: http://(0de284ecf0cd or 127.0.0.1):8888/?token=e5a2541812d85e20026b1d04983dc8380055f2d16c28a6ad
  • Edit this: (0de284ecf0cd or 127.0.0.1) to: 127.0.0.1, in the above link and open it in your browser
  • Then open mgc.ipynb

• To mount/load local files into docker container:

  • Do docker run -it --rm -v :/root/workspace/ -p 8888:8888 --name mgcpy-env tpsatish95/mgcpy:latest, replace `` with your local dir path.
  • Do cd ../workspace when you are inside the container to view the mounted files. The mgcpy package code will be in /root/code directory.

MGC Algorithm's Flow

Power Curves

• Recreated Figure 2 in https://arxiv.org/abs/1609.05148, with the addition of MDMR and Fast MGC

License

This project is covered under the Apache 2.0 License.