Linearly constraint Bayesian Matrix Factorization for Blind Source Separation

This repository provides the implementation for the paper Linearly constraint Bayesian Matrix Factorization for Blind Source Separation (Mikkel N. Schmidt). Majority of the code are translated from the Matlab implementation that is provided by Mikkel N. Schmidt

Installation and preparation

1. Clone this repo:

   git clone https://github.com/lyn1874/Linear_Constraint_Bayesian_NMF.git
   cd Linear_Constraint_Bayesian_NMF

2. Requirement:

   python3/3.7.7  
   matplotlib/3.2.1-python-3.7.7  
   scipy/1.4.1-python-3.7.7  
   pandas/1.0.3-python-3.7.7
   

Train the model

1. Train the model:

   ./run.sh dataset N mu_prior infinity
   Args:
   dataset: mnist 
   N: number of components, int
   mu_prior: the mean of the prior distribution for component matrix A and mixing coeffients B
   infinity: bool variable. If True, then the variance of the prior distribution for A and B are infinitely large (non-informative prior)	

Citation

If you use this code for your research, please cite the paper:

@inproceedings{NIPS2009_371bce7d,
 author = {Schmidt, Mikkel},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {Y. Bengio and D. Schuurmans and J. Lafferty and C. Williams and A. Culotta},
 pages = {},
 publisher = {Curran Associates, Inc.},
 title = {Linearly constrained Bayesian matrix factorization for blind source separation},
 url = {https://proceedings.neurips.cc/paper/2009/file/371bce7dc83817b7893bcdeed13799b5-Paper.pdf},
 volume = {22},
 year = {2009}
}