Graph-spa - A Spatiotemporal Graph Learning Approach for Dynamic Prediction of Acute Respiratory Failure Using High-Dimensional ICU Data
This document provides an overview and descriptions of the various files and directories within the project.
- Figure_02_03_plots.ipynb: Contains code to generate Figure 02 and Figure 03.
- Figure_04_plot.ipynb: Contains code to generate Figure 04.
- Graph-spa_training.ipynb: Training code for Graph-spa model
- Graph-spa_training_and_extraction.ipynb: Training code for Graph-spa model with the extraction of features graphs
- layer_dev.py: contains classes for each layer of the model; upgraded version of layer.py from https://github.com/liuxz1011/TodyNet
- layer_dev_graph.py: contains classes for each layer of the model with the extraction of features graphs from the temporal graph pooling layer; an upgraded version of layer.py from https://github.com/liuxz1011/TodyNet
- loader.py: modified data loader file resourced from https://github.com/ratschlab/HIRID-ICU-Benchmark
- net_dev.py: model adapted from https://github.com/liuxz1011/TodyNet
- net_dev_graph.py: model adapted from https://github.com/liuxz1011/TodyNet with extraction of feature graphs
- varname_clean.csv: Clean variable names from HiRID: modified from https://github.com/ratschlab/HIRID-ICU-Benchmark
contains numpy matrix for plotting Figure 04
example of a best-performing saved model
We do not have the authorization to share the HiRID ICU Benchmark dataset. Please contact https://github.com/ratschlab/HIRID-ICU-Benchmark and https://physionet.org/content/hirid/1.1.1/