wucg/neural_ode_processes_for_network_dynamics-master

Neural ODE Processes for Network Dynamics (NDP4ND), a new class of stochastic processes governed by stochastic data-adaptive network dynamics, is to overcome the fundamental challenge of learning accurate network dynamics with sparse, irregularly-sampled, partial, and noisy observations.

Running Environment and Dependency Packages

We have tested the code on the hardware environment with Intel(R) Xeon(R) Silver 4310 [email protected] * 48 and NVIDIA GeForce RTX 3090 * 3.

The running environment and dependency packages are listed below.

Python 3.10.8

    pickle==4.0
    
    numpy==1.23.5
    
    scipy==1.9.3
    
    networkx==2.8.4
    
    torch==1.13.1
    
    torch_scatter==2.1.0+pt113cu117
    
    sklearn==1.2.0
    
    pandas==1.5.2
    
    seaborn==0.12.2

fastdtw==0.3.4

Script Description for Running Code

Training Phase: Train the NDP4ND for each network dynamics scenario

bash runme_train.sh

Test on all scenarios:

bash runme_test.sh

Test with various sparsities:

bash runme_test_testSparsity.sh

Test with various noises:

bash runme_test_testNoise.sh

get the infomation of testing results:

python stat_info_of_tesing_results.py

---

NOTE:

• The generated training and testing sets can be found in ./data/DynamicsData/

• Trained models can be found in ./saved_models/

• Testing results can be found in ./results/

---