LDFA-H: Latent Dynamic Factor Analysis of High-dimensional time series

This repo consists of

1. Python package ldfa which implements LDFA-H [1]
2. Experimental data analyzed in [1]
3. Reproducible IPython notebooks for simulation and experimental data analysis in [1]

Install

Prerequisite

Package ldfa requires:

1. Python >= 3.5
2. numpy >= 1.8
3. matplotlib and scipy.

Git clone

Clone this repo through github:

git clone https://github.com/HeejongBong/ldfa.git

Python install

Install package ldfa using setup.py script:

python setup.py install

Experimental data

The data are available in /data/. The data file consists of v4_lfp_beta and pfc_lfp_beta which are the beta band-passed filtered LFP in PFC and V4, respectively. maps.mat contains the spatial arrangement of the 96 electrodes on each Utah array. The original data are collected by Khanna, Scott, and Smith (2020) [2].

Reproducible Ipython notebooks

The scripts are available in /example/. The script for simulation analysis is provided in Python notebook 3.1 CDFA versus existing methods in addressing noise auto-correlation.ipynb with an accessory MATLAB notebook 3.1 DKCCA on simulated data.ipynb. To run DKCCA, a separate installation of MATLAB package DKCCA is required [3].

References

[1] Bong, H., Liu, Z., Ren, Z., Smith, M. A., Ventura, V., & Kass, R. E. (2020). Latent Dynamic Factor Analysis of High-Dimensional Neural Recordings. Submitted to NeurIPS2020.

[2] Khanna, S. B., Scott, J. A., & Smith, M. A. (2020). Dynamic shifts of visual and saccade signals in prefrontal cortical regions 8Ar and FEF. Journal of neurophysiology. In press.

[3] Rodu, J., Klein, N., Brincat, S. L., Miller, E. K., & Kass, R. E. (2018). Detecting multivariate cross-correlation between brain regions. Journal of neurophysiology, 120(4), 1962-1972.