-
Notifications
You must be signed in to change notification settings - Fork 26
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Experiments with end-to-end training #12
Conversation
…into endtoend_gnn
…into endtoend_gnn
…into endtoend_gnn
This is now ready to be merged, we have a baseline end-to-end training for elements to PFCandidates that seems to work reasonably well. The following plots are done on 1k testing events not seen during training. Confusion matrix of true vs predicted candidate pdgid (0 - no candidate): Number of true vs predicted candidates per event: True vs predicted pt of 1000 candidates in one testing event: |
* start adding gnn * add gnn to benchmarks * Update run_training.sh * update graph_data and EdgeNet to include edge_attr and benchmarking * add notebook for plotting * added first end-to-end training example * up * up * up * added end2end training examples * up * up * up * up * up * up * cmdline args * added sequential conv * added cls accuracy monitoring * up * remove additional edges * add act * dataset location * elem id encoding, fix norm * fix nans * add ntest * added num pred and true plotting: * added npy file saving * switch to relu * pfnet7 same setup as others * up * loss coefs configurable * added model to predict only id * fix bugs with relabeling * fix plot title * cosmetic * up * up * added sinkhorn loss * fixes * added reordering code * fix printout, reweighting * added class weighting * update readme * dropout configurable, simplify cross-check model * fix weight application * update weights Co-authored-by: Javier Duarte <jduarte@ucsd.edu> Former-commit-id: 779431e
graph_data.py
to prepare a more coherent set of input elementsx
ordered by block id, with the output candidatesy_candidates
padded within each block to have the same length.