GNNs without dampings

[AgathaSchmidt]

Feature description

First, we want to create simple GNNs that can make predictions for all 400 counties without any changes in contact pattern. Each county is represented by a node and the edges between them represent the connection between nodes. Edge weights represent the number of individuals commuting from one node to another.

We create two typer of GNNs:

1. Simple GNN without edge weights. Computing hidden states based on edge weights is computationally expensive. We implement models, that do not explicitly use edge weights in their calculations.
2. GNNs with edge weights. These GNNs need more training time but have the potential to yield more accurate results thanks to more information provided.