-
AdaGramp can adaptively design a appropriate message passing architecture according to data set distribution for anomaly detection.
-
The framework of AdaGramp is as follows:
- Ensure you have installed CUDA 11.0 before installing other packages
1.Python environment: recommending using Conda package manager to install
conda create -n adagramp python=3.7
source activate adagramp
2.Python package:
torch == 1.13.1
torch-geometric == 2.3.0
torch-cluster == 1.6.1
torch-scatter == 2.1.1
torch-sparse == 0.6.17
torch-spline-conv == 1.2.2
faiss == 1.7.3
1.Performance test with the optimal message passing architecture designed by AdaGramp
run performance_test.py
2.Search new message passing architectures from scratch using AdaGramp
run hierarchical_random_search.py