JunZheCheng-ctrl/Transformer-IDS
Transformer-based Intrusion Detection System with Cross-Dataset Evaluation
What's novel
Transformer-based Intrusion Detection System with Cross-Dataset Evaluation
Code Analysis
15 files read · 4 roundsA collection of ad-hoc PyTorch training scripts for binary intrusion detection using a TabTransformer-style model, with cross-dataset evaluation via naive feature truncation, but the repository is structurally broken with compiled bytecode files, misnamed files, and a PNG saved as a .py file.
Strengths
The TabTransformer model implementation (FeatureGate + per-feature token embedding + TransformerEncoder + temperature scaling) is a reasonable architecture for tabular data, and the experiment scripts do cover multiple evaluation scenarios (within-dataset, transfer, imbalance, depth ablation).
Weaknesses
The repository is fundamentally non-functional: the core model class is in the wrong file, multiple .py files are actually compiled bytecode or PNG images, imports reference non-existent module paths, file names are misleading (imbalance experiment called 'cross_dataset', feature importance called 'loss_balanced'), and there are zero tests. The README describes a clean directory structure and 'AUC-oriented optimization' that don't match the actual code.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
20
Contributors
1
Files
39
Active weeks
2
Repository
Language
Python
Stars
1
Forks
0
License
—