Open
0 of 1 issue completed
Description
Currently, node IDs are "global" in the JobShopGraph (i.e., each node has a different ID independently of its type). However, most GNN frameworks, such as PyG, expect IDs to be unique only among nodes of the same type. This discrepancy necessitates the use of an observation wrapper to map global embeddings to local ones and partition the edge index matrix into multiple matrices.
To avoid this overhead, we could use a new JobShopGraph class to keep track of local IDs. A Node ID will be a tuple of NodeType and local ID (networkx still expects unique global IDs).
After this, we should also need to add new RL environments with a more useful observation space.
This introduces breaking changes, so a new major version is needed.