Table 3 Table of model architecture parameters for training and inference

conv_encode_edge

True

Whether or not the edge information should be encoded in the convolution

Block

Res+

Graph backbone block to be chosen among the following types: {res+, res, dense, plain}

gcn_aggr

max

The aggregator of GENConv to be chosen among the following types: {mean, max, add, softmax, softmax_sg, power}

Norm

Batch

the type of normalization layer to be chosen among the following types: {batch, layer}

Backbone

rev

gcn backbone to be chosen among the following types: {deepergcn, weighttied, deq, rev}