https://github.com/arminarj/deepgcca-pytorch

An implementation of Deep Generalized Canonical Correlation Analysis (DGCCA or Deep GCCA) with pytorch.
https://github.com/arminarj/deepgcca-pytorch

cca correlation deepgcca dgcca multimodal multiview multiview-learning statistics

Last synced: 10 months ago
JSON representation

An implementation of Deep Generalized Canonical Correlation Analysis (DGCCA or Deep GCCA) with pytorch.

• Host: GitHub
• URL: https://github.com/arminarj/deepgcca-pytorch
• Owner: arminarj
• License: mit
• Created: 2020-05-23T21:07:03.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2020-06-11T11:09:32.000Z (about 6 years ago)
• Last Synced: 2025-04-09T11:52:48.972Z (over 1 year ago)
• Topics: cca, correlation, deepgcca, dgcca, multimodal, multiview, multiview-learning, statistics
• Language: Jupyter Notebook
• Homepage:
• Size: 2.62 MB
• Stars: 48
• Watchers: 2
• Forks: 13
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# DGCCA-pytorch:

A Pytorch Implementation of Deep Generalized Canonical Correlation Analysis as described in:

Adrian Benton, Huda Khayrallah, Biman Gujral, Dee Ann Reisinger, Sheng Zhang, and Raman Arora. Deep Generalized Canonical Correlation Analysis. The 4th Workshop on Representation Learning for NLP. 2019

[(Paper-link)](https://www.aclweb.org/anthology/W19-4301/)

# Deep Generalized Canonical Correlation Analysis:

*Generalized Canonical Correlation Analysis (GCCA)* is a method which corresponds to solving an optimization problom objective to find the best linear shared space called ***G*** for the *J* view of a data

**DeepGCCA** is a non-linear version of GCCA which uses neural networks as the feature extractor functions instead of linear transformers. ***DGCCA*** is some how exention of ***DeepCCA*** for more than two views though it has a different objective function.

![](./img/DGCCA.jpg)

![](./img/GCCA-DGCCA-Benton.jpg)

 - figures source [Deep Generalized Canonical Correlation Analysis - Arxiv 1702.02519](https://arxiv.org/abs/1702.02519)

## Pseudocode algorithm:

Pseudocode algorithm based on the paper,

![](./img/psuedocode.jpg)

# Example:

Synthatic Data: [(synth data generator)](/synth_data.py)

![](./img/Synth-data.jpg)

DGCCA Latent space for views:

![](./img/Lantent-space-views.jpg)

 - figures source [Deep Generalized Canonical Correlation Analysis - Arxiv 1702.02519](https://arxiv.org/abs/1702.02519)

# Prerequest:

- Python 3.6>=

- Pytorch 1.4 >= (should also work with >=1.0)

- Numpy

- Scipy

- Seanborn

# Other Implementations:

 - [Theano Implementation](https://bitbucket.org/adrianbenton/dgcca-py3/src/master/) By Adiran Benton.

 

 

 ## Notes:

 

 ### check list:

 

 - cuda test

 - Varient Batch sizes test

 

 ### to do: 

 

 - Nan gradient/update rules (Famous issue of Deep CCA - based models, like [DeepCCA Nan outputs](https://github.com/Michaelvll/DeepCCA))

 - More numerical stabilization for varient Architectures

 

 Warmest thanks to Mr. Adrian Benton for his kind helps.