PNN

This repository contains the code to replicate the experiments in the paper Precision Neural Networks: Joint Graph And Relational Learning (link).

Description

Precision Neural Networks (PNNs) are graph convolutional neural networks operating on the graph defined by the precision matrix. Their weights are optimized jointly with the estimation of a sparse precision matrix, leading to task-aware dependency structures.

Requirements

• Python 3.11.13
• pip install -r requirements.txt

Datasets

The file data_preprocess.ipynb contains the code to preprocess the real datasets. Below the preliminary steps to download the datasets.

ABIDE:

Following instructions at http://preprocessed-connectomes-project.org/abide/, download the file Phenotypic_V1_0b_preprocessed1.csv and place it at the path data/datasets/abide/Phenotypic_V1_0b_preprocessed1.csv

ADNI

From the website https://ida.loni.usc.edu/, download the files ADSP_PHC_T1_FS_DATADIC and ADSP_PHC_T1_FS and place them in the folder data/datasets/adni

Experiments

To replicate experiments in the paper, run the following scripts. The dataset and the experiment parameters should be specified inside the scripts. The folder best_params contains the best hyperparameter configuration obtained for each model and experimental setting on each dataset, and it is automatically loaded when running the experiment.

• synth_exp.py to run the experiments on synthetic data
• real_exp.py to run the experiments for PNNs and VNN on real data
• pca.py to run the experiments for PCA on real data