PKBoost: Adaptive GBDT for Concept Drift, Built from scratch in Rust, PKBoost manages changing data distributions in fraud detection with a fraud rate of 0.2%. It shows less than 2% degradation under drift. In comparison, XGBoost experiences a 31.8% drop and LightGBM a 42.5% drop
Stars
67
Forks
4
Watchers
67
Open Issues
0
Overall repository health assessment
No package.json found
This might not be a Node.js project
56
commits
8
commits
2
commits
2
commits
1
commits
Merge pull request #8 from PKBoost-AI-Labs/feat/serialization-regression
ab8d434View on GitHubfeat: resolve merge conflicts and add serialization support for regression models
4b415fdView on GitHubfeat(perf): optimize histogram building and data gathering
8d6c198View on GitHubUltimate Squeeze: f32 SIMD Histograms, Double-Pass, ForkUnion, and AoS optimizations. Achieved 3.63M samples/sec (2.3x speedup).
caf0201View on GitHubperf: replace Rayon with ForkUnion for parallelism and thread allocation
ee2352fView on GitHub