Developing a GNN to calssify doublets in the QA tracking algorithm.
Dataloader, generate the graph in eta-phi space from the samples in TrackML dataset (dataset are not included in this repo).
Main code to build up the model. First transform the graph from Dataloader and generate a k-nn graph in the embedding space. At the end tranform it back to get the edge scores.
Main training code, in each loop we need to the the HitiD matching in order to trace the edge scores.
Use for checking the strutuce of the training_PhaseIII.py, only contain the core of training without looping over the epochs.
For test run only, still developing from the old package.