CCA-Zoo: A collection of Regularized, Deep Learning based, Kernel, and Probabilistic CCA methods in a scikit-learn style framework

Abstract

cca-zoo is a Python package that implements many variants in a simple API with standardised outputs. We would like to highlight the unique benefits our package brings to the community in comparison to other established Python packages containing implementations of CCA. Firstly, cca-zoo contains a number of regularised CCA and PLS for high dimensional data that have previously only been available in installable packages in R. Native Python implementation will give Python users convenient access to these powerful models for both application and the development of new algorithms. Secondly,cca-zoo contains several deep CCA variants written in PyTorch (Paszke et al., 2019). We adopted a modular style allowing users to apply their desired neural network architectures for each view for their own training pipeline. Thirdly, cca-zoo contains generative models including probabilistic and deep variational CCA. This class of variations can be used to model the multiview data generation process and even generate new synthetic samples. Finally, cca-zoo provides data simulation utilities to synthesize data containing specified correlation structures as well as the paired MNIST data commonly used as a toy dataset in deep multiview learning.