krishnaswamylab / compass Goto Github PK

"Software package for investigation of graph expressivity"

Latest: for classification (and usage example), see PDEClassifier.
for wrapped networks with batch normalization, use HeatBlock, WaveBlock, and LinearBlock (with skip connection).
use ChebyPolyLayer for a message passing class.
Use get_cheby_coefs_heat and get_cheby_coefs_wave for computing the Chebyshev polynomial coefficients for the PDEs used by ChebyPolyLayer.
For usage example, see this classifier that uses the ChebyPolyLayer here. [DEPRECATED]

cross_validate.py for cross validation
prepare_dataset.py for creating k-fold cross validation dataset, and also adding edge_weights as 1's (because they are missing from data). Now only works on data with existing node features.
make_tuning_configs.py for creating the configuration yaml files and dSQ array job file for hyperparameter tuning.