Added support for Graph Convolution Networks (GCN) as a base classification model in Graph Agreement Models (GAM).
To support this, this pull request introduces the following differences:
The data pipeline needed to be modified to contain the extra information required by GCN.
This implementation of GCN feeds data in a different way than previous models, so in order to keep the previous trainers clean and easy to read, this pull request adds a separated trainer for GCN.
To account for the way data is fed into GCN, we also have a separate implementation for adversarial training with GCNs, found in adversarial_sparse.py.
Added support for Graph Convolution Networks (GCN) as a base classification model in Graph Agreement Models (GAM).
To support this, this pull request introduces the following differences:
@kvis-google @dattias