Select the optimal order of latent variables for PCA/ICA/PPCA/PICA/CCA.

This tool-box contains a order selection algorithms for the linear admixture models (see this link). Most existing methods (based on information-theoretic criteria, see this document) rely on the assumption that samples are independent and identitically distributed. The proposed algorithm used here is not based on the likelihood function, and therefore has been shown to produce more consistent results.

For more detail, please see our paper in IEEE TMI:
Seghouane, Shokouhi, "Consistent Esitmaiton of Dimensionality for Data Driven Methods in fMRI Analysis", IEEE Transactions on Medical Imaging, 2018.

Please make sure to cite the our paper.

This repository contains code for two types of algorithms:

• single-vector factor analysis (e.g., PCA) Python and Matlab
• double-vector analysis (e.g., CCA) Matlab

• numpy, scipy

For experiments on real FMRI data:

• nipype (This automatically installs some pre-req modules, e.g., nibabel)
• nilearn

Note: Installing nipype on Windows Anaconda is a bit tricky. This is because pip isn't able to directly install traits. The solution for me was to download the traits wheel from from here and then run

pip install <location of trait wheel file>.whl

• Matlab

For direction-of-arrival simulations, I use doa-tools. You can find a forked version here in case @morriswmz changes the original.

NS, 2018