shpangit/corex_torch

Corex but with torch package

READ ME

CorEx implementation with the S&P500 example.

Tested on python 3.9.0

It is a re-implementation of https://github.com/gregversteeg/CorEx.
The references are :

Discovering Structure in High-Dimensional Data Through Correlation Explanation Greg Ver Steeg and Aram Galstyan, NIPS 2014, http://arxiv.org/abs/1406.1222

Some theoretical developments are described here: Maximally Informative Hierarchical Representions of High-Dimensional Data Greg Ver Steeg and Aram Galstyan, AISTATS 2015, http://arxiv.org/abs/1410.7404

The folder is organised with following files and folder:

• corex.py: original numpy code for CorEx algorithm from https://github.com/gregversteeg/CorEx.
• corex_torch.py: CorEx algorithm with torch implementation.
• corex_visualization.py: plotly based tools to show results in form of graph.
• data: folder containing the dataset of S&P500 components.
• Corex example on SP500.ipynb: Jupyter notebook with an example of using CorEx for clustering the assets.
• requirements.txt: list of dependecy for pip installation

corex.py and corex_torch.py have same structure.