EEG,ECoG phase series analysis using Torus Graph / 脳神経データ解析

Fit torus graph to multivariate circular data and view test statistics under null hypothesis that two nodes are independent.

Note: This model does not handle time series information.

Files

• torus_graph_model/
  • Torus graph model
• utils/
  • preprocessing for neuroscience time series data.
  • visualization of correlation matrix (and its variant)

Basic Usage

Sample from torus graph

python torus_graph_model/sample.py

###　Use your time series data

1. Place your time series data in CSV format.
2. Run

python scripts/inference.py <path_to_your_csv> #if it is a raw signal
python scripts/inference.py <path_to_your_csv> --phase #if it is already a circular data series

Use human EEG data from Chennu et al., 2016

python script/score_matching.py #naive matrix inversion or conditional models
python script/score_matching_admmpath.py #ADMM with regularization path and SMIC minimization


# confirmation using simulation data
python script/simulation.py

About this data

1. baseline
2. mild
3. moderate
4. recover

Optimization Method

Score matching estimator using

• without regularization (full model)
• without regularization (conditional model)
• $l_1$ regularization using gradient descent (not recommended)
• LASSO with ADMM
• LASSO with LARS and SMIC

Example Usage

1. Place your EEG/ECoG dataset to PATH_TO_DATA_DIR
2. Specify FILE_NAME to be your target time series data.
3. run python torus_graph_model/sample.py and wait.
4. check output/ for results.

Acknowledgement

This repository is partly a reimplementation of Klein et al, 2020