Team Road Rage - Graph Neural Network - Pak Hay Kwok ([email protected]) and Qi Qi ([email protected])
This is the graph neural network submission for Team Road Rage
There are quite a few pre-processing steps, most of the small data files for Berlin have been uploaded to save having to re-run
From notebooks:
Run create_graph_dataset.ipynb - this will process all the raw images into graph data and store the data in 3 .h5 files (training, validation, testing)
Run train.ipynb - this will do the training and the submission
Prior to running the above notebooks, you will need to run two python scripts (and change the city names) in the src folder:
Run gnn_make_static_grid.py - this calculates the overall max volume for each pixel across all channels
Run gnn_make_nodes_edges.py - this will create the nodes and edges information for the road network
gnn_dataset.py - the graph dataset
models_graph_resnet.py - the gnn from last year https://github.com/mie-lab/traffic4cast-Graph-ResNet
models_graph_ensnet.py - the new gnn