A Virtual Function Placement (VFP) method using Double Deep Q-Learning (DDQL) powered by GATConv (a Graph Neural Network -GNN- layer).
GATConv paper: https://arxiv.org/abs/1710.10903
GATConv library: https://pytorch-geometric.readthedocs.io/en/latest/generated/torch_geometric.nn.conv.GATConv.html
This repository implements the VNF placement agent described here (https://ieeexplore.ieee.org/abstract/document/10207694/) using GNN.
This method employs ReNet as its learning environment. For more details, check here: https://github.com/mshokrnezhad/ReNet
IMPORTANT NOTE: GATConv returns a set of values for each input graph node. Therefore, it is extremely useful for methods attempting to comprehend the nodes based on their relationships in a graph. I require a GNN layer that returns values for the graph itself as a whole as opposed to its nodes. It may be possible via GATConv, but I do not have time for further investigation at this time.