How joker383627/for_hep works
This repository contains a particle-physics analysis notebook collection and a PyTorch top-versus-QCD jet classifier. The classifier workflow reads ROOT jet data, prepares padded and standardized particle inputs, trains a particle-attention model, and evaluates classification performance. The model internals and training/evaluation behavior are supported by sampled source; notebook roles are inferred only from their filenames because their contents and the README were not provided.
Main entry points: models.py Class logits · train.py Evaluation metrics · preprocess.py Feature preparation · top_tagger.ipynb Tagger workflow
Components 16
Jet data preparation
- Feature preparation
top_tagger/preprocess.py - Jet dataset
top_tagger/preprocess.py - Data loaders
top_tagger/preprocess.py
Classifier model
- Particle transformer
top_tagger/models.py - Pairwise features
top_tagger/models.py - Particle attention
top_tagger/models.py - Class attention
top_tagger/models.py - Class logits
top_tagger/models.py
Training and evaluation
- Training loop
top_tagger/train.py - Evaluation metrics
top_tagger/train.py - Background rejection
top_tagger/train.py
Physics analysis
- Tagger workflow
top_tagger/top_tagger.ipynb - Data mining
data_analysis/data_mining.ipynb - SM vs SMEFT
data_analysis/sm_vs_smeft.ipynb
Other components
- Analyst
- ROOT jet data
Connections 15
- ROOT jet data → Feature preparation: reads ROOT
- Feature preparation → Jet dataset: builds tensors
- Jet dataset → Data loaders: splits batches
- Training loop → Data loaders: consumes batches
- Training loop → Class logits: trains model
- Training loop → Evaluation metrics: evaluates model
- Evaluation metrics → Class logits: scores batches
- Evaluation metrics → Background rejection: provides scores
- Class logits → Pairwise features: computes interactions
- Class logits → Particle attention: dispatches embeddings
- Class logits → Class attention: dispatches embeddings
- Class logits → Evaluation metrics: returns logits
- Analyst → Tagger workflow: uses
- Analyst → Data mining: uses
- Analyst → SM vs SMEFT: uses