now including tensor and deep tensor cca and variational bayesian cca!
Note: for standard install use: pip install cca-zoo
For deep learning elements use: pip install cca-zoo[deep]
For probabilistic elements use: pip install cca-zoo[probabilistic]
This means that there is no need to install the large pytorch package or numpyro to run cca-zoo unless you wish to use deep learning
Available at https://cca-zoo.readthedocs.io/en/latest/
If this repository was helpful to you please do give a star.
In case this work is used as part of research I attach a DOI bibtex entry:
@software{james_chapman_2021_4586389,
author = {James Chapman},
title = {jameschapman19/cca\_zoo: v1.1.22},
month = mar,
year = 2021,
publisher = {Zenodo},
version = {v1.1.22},
doi = {10.5281/zenodo.4586389},
url = {https://doi.org/10.5281/zenodo.4586389}
}
A variety of linear CCA and PLS methods implemented where possible as the solutions to generalized eigenvalue problems and otherwise using alternating minimization methods for non-convex optimisation based on least squares
Solutions based on either alternating least squares or as the solution to genrralized eigenvalue problem
https://academic.oup.com/biomet/article-abstract/58/3/433/233349?redirectedFrom=fulltext
https://arxiv.org/pdf/1502.02330.pdf
Mai's sparse CCA
Witten's sparse CCA
Waiijenborg's elastic penalized CCA
A variety of Deep CCA and related methods. All allow for user to pass their own model architectures. Recently added solutions to DCCA using nob-linear orthogonal iterations (or alternating least squares)
https://ttic.uchicago.edu/~klivescu/papers/andrew_icml2013.pdf https://arxiv.org/pdf/1510.02054v1.pdf Using either Andrew's original Tracenorm Objective or Wang's alternating least squares solution
https://www.aclweb.org/anthology/W19-4301.pdf An alternative objective based on the linear GCCA solution. Can be extended to more than 2 views
https://arxiv.org/abs/1904.01775 An alternative objective based on the linear MCCA solution. Can be extended to more than 2 views
https://arxiv.org/pdf/2005.11914.pdf
http://proceedings.mlr.press/v37/wangb15.pdf
https://arxiv.org/pdf/1610.03454.pdf Wang's DVCCA and DVCCA Private
Piggybacking off sklearn's kernel methods which also allow callable kernel functions
I've translated my work building baselines for my own research into a python package for my own experience but also, I hope, to help others. With that in mind if you have either suggestions for improvements/additions do let me know using the issues tab. The intention is to give flexibility to build new algorithms and substitute model architectures but there is a tradeoff between robustness and flexibility.
I've added this section to give due credit to the repositories that helped me in addition to their copyright notices in the code where relevant.
Models can be tested on data from MNIST datasets provided by the torch package (https://pytorch.org/) and the UCI dataset provided by mvlearn package (https://mvlearn.github.io/)
MATLAB implementation https://github.com/anaston/PLS_CCA_framework
MATLAB implementation of SPLS by @jmmonteiro (https://github.com/jmmonteiro/spls)
Keras implementation of DCCA from @VahidooX's github page(https://github.com/VahidooX) The following are the other implementations of DCCA in MATLAB and C++. These codes are written by the authors of the original paper:
Torch implementation of DCCA from @MichaelVll & @Arminarj: https://github.com/Michaelvll/DeepCCA
C++ implementation of DCCA from Galen Andrew's website (https://homes.cs.washington.edu/~galen/)
MATLAB implementation of DCCA/DCCAE from Weiran Wang's website (http://ttic.uchicago.edu/~wwang5/dccae.html)
MATLAB implementation of TCCA from https://github.com/rciszek/mdr_tcca
Torch implementation of VAE (https://github.com/pytorch/examples/tree/master/vae)