quinlan7 / dvgnn

This codes includes our works on Traffic Flow Prediction by Dynamic Causal Explanation Based Diffusion-Variational Graph Neural Network for Spatio-temporal Forecasting.

The code consists of two parts. The first part is dynamic causal graph generation module, and the second part is multi-step prediction module. We supply the demo of transportation and FMRI dataset.

##For the dynamic causal graph generation module, The setting are shown in configuration document：./graph_generation/configurations/Tdrive.conf

Requirements:

• tensorflow
• scipy
• numpy
• matplotlib
• pandas
• math
• seaborn
• sklearn
• argparse
• configparser
• time

Run the demo:

./graph_generation/Transportation or FMRI\main.py

Then we will get the parameter file after training, for example, Tdrive_normalization_parameter.npz. Run dynamic_graph_trans_.py to generate the dynamic transition matrix, such as dynamic_Tdrive_adj.npy file.

##For the second part, put the generative file from the first step to ./prediction/Transportation or FMRI/generated_adj (file directory)

Requirements:

• torch
• shutil
• numpy
• matplotlib
• pandas
• math
• tensorflow
• sklearn
• argparse
• csv
• time

Run the demo:

./prediction-code/Transportation or FMRI\main.py

If you have any questions, please feel free to email me, [email protected]!