FGM_Neighborhood

Functional Graphical Models via Neighborhood Selection Approach

ROC

This folder contains codes running multiple methods, namely FPCA-gX, FPCA-gY, FGLasso, PSKL, and FPCA-PSKL (for Model D). Threshold is set to $\epsilon=0$. Penalty parameter $\lambda_n$ varies. Output results are TPR and FPR for each methods. Runtime of each method under each run is also recorded. A script is given for plotting ROC for all methods under each setting.

ROC Threshold

This folder contains codes running FPCA-gX method. Threshold varies under each setting. Penalty parameter $\lambda_n$ varies under each threshold parameter. Output results are TPR and FPR for FPCA-gX method under all thresholds A script is given for plotting ROC for all thresholds under each setting.

SCV

This folder contains codes running FPCA-gX method. Threshold varies under each setting. Penalty parameter $\lambda_n$ varies under each threshold parameter. Output results are adjacency matrices under AND and OR schemes under the optimal choice of $\lambda_n, t_\epsilon$ pairs that maximizes SCV-BIC. Runtime of each run is also recorded. A script is given for computing precision and recall for the optimal adjacency matrices.

Real Data Analysis

This folder contains codes running FPCA-gX method. Two datasets: ABIDE for autism and ADHD. Both raw data and time series extracted from raw data can be found in Data folder. Time series are extracted by "ABIDE.read.data.R" and "ADHD.read.data.R" respectively.

ABIDE dataset:

"SCV.autism.R" selects and outputs the optimal adjacency matrices using SCV under FPCA-gX method. "spa.ctrl.autism.R" outputs an adjacency matrix with 2% connection density under FPCA-gX method.

ADHD dataset:

"SCV.ADHD.R" selects and outputs the optimal adjacency matrices using SCV under FPCA-gX method. "spa.ctrl.ADHD.R" outputs an adjacency matrix with 2% connection density under FPCA-gX method. "spa.ctrl.ADHD.FGLasso.R" outputs an adjacency matrix estimated by FGLasso method.