PyIVA

pyiva is an implementation of the independent vector analysis (IVA) algorithm using a multivariate Laplace prior. It uses the same score function as found in the original IVA paper. IVA is an extension of independent component analysis (ICA) to multiple statistically-dependent datasets, and can be used for dimensionality reduction and data fusion.

Primary author: Austin Kim (austinkim1004 at gmail.edu) Code maintainer: Zois Boukouvalas (zoisb at umd.edu) Edited and packaged by Daniel C. Elton

This code is based on the Matlab code iva_laplace.m by Matthew Anderson (matt.anderson at umbc.edu)

Installation

to install, run python setup.py install

Example usage

    from pyiva.iva_laplace import iva_laplace

    W = iva_laplace(X)

Documentation on iva_laplace()

Required arguments:

• X : numpy array of shape (N, K, T) containing data observations from K data sets. Here X{k}=A{k}S{k}, where A{k} is an N x N unknown invertible mixing matrix and S{k} is N x T matrix with the nth row corresponding to T samples of the nth source in the kth dataset. For IVA it is assumed that each source is statistically independent of all the sources within a dataset and exactly dependent on at most one source in each of the other datasets. The data, X, is a 3-dimensional matrix of dimensions N x K x T. The latter enforces the assumption of an equal number of samples in each dataset.

Optional keyword arguments:

• A : [], true mixing matrices A, automatically sets verbose
• whiten : Boolean, default = True
• verbose : Boolean, default = False : enables print statements
• W_init : [], ... % initial estimates for demixing matrices in W
• maxIter : 2*512, ... % max number of iterations
• terminationCriterion : string, default = 'ChangeInCost' : criterion for terminating iterations, either 'ChangeInCost' or 'ChangeInW'
• termThreshold : float, default = 1e-6, : termination threshold
• alpha0 : float, default = 0.1 : initial step size scaling

Output:

• W : the estimated demixing matrices so that ideally W{k}A{k} = P*D{k} where P is any arbitrary permutation matrix and D{k} is any diagonal invertible (scaling) matrix. Note P is common to all datasets; this is to indicate that the local permutation ambiguity between dependent sources across datasets should ideally be resolved by IVA.

During runtime the following are reported:

• cost - the cost for each iteration
• isi - joint inter-symbol-interference is available if user supplies true mixing matrices for computing a performance metric

Citation

If you use pyIVA, please cite

@misc{PyIVA,
author = {Zois Boukouvalas and Austin Kim and Daniel C. Elton},
title = {{PyIVA}},
howpublished = {\url{https://github.com/zoisboukouvalas/pyiva}}
}

@ARTICLE{2018arXiv181100628B,
author = {{Boukouvalas}, Z. and {Elton}, D.~C. and {Chung}, P.~W. and {Fuge}, M.~D.},
title = "{Independent Vector Analysis for Data Fusion Prior to Molecular Property Prediction with Machine Learning}",
journal = {ArXiv e-prints},
archivePrefix = "arXiv",
eprint = {1811.00628},
primaryClass = "stat.ML",
keywords = {Statistics - Machine Learning, Condensed Matter - Materials Science, Computer Science - Machine Learning},
year = 2018,
month = nov,
adsurl = {http://adsabs.harvard.edu/abs/2018arXiv181100628B},
adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}

References

[1] T. Kim, I. Lee, & T.-W. Lee, "Independent Vector Analysis: Definition and Algorithms," Proc. of 40th Asilomar Conference on Signals, Systems, and Computers, 2006, 1393-1396

[2] T. Kim, T. Eltoft, & T.-W. Lee, "Independent Vector Analysis: an extension of ICA to multivariate components," Lecture Notes in Computer Science: Independent Component Analysis and Blind Signal Separation, Independent Component Analysis and Blind Signal Separation, Springer Berlin / Heidelberg, 2006, 3889, 165-172

[3] T. Kim, H. T. Attias, S.-Y. Lee, & T.-W. Lee, "Blind Source Separation Exploiting Higher-Order Frequency Dependencies," IEEE Trans. Audio Speech Lang. Process., 2007, 15, 70-79