studio/proxy
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
41 Commits 1 Branch 0 Tags
039df0757d6dbd4125c5549c606183376e9cb1c7
Commit Graph

3 Commits

Author SHA1 Message Date
Sienna Meridian Satterwhite
039df0757d feat: complete ensemble integration and remove legacy model code 
- Remove legacy KNN DDoS replay and scanner model file watcher
- Wire ensemble inference into detector check() paths
- Update config: remove model_path/k/poll_interval_secs, add observe_only
- Add cookie_weight sweep CLI command for hyperparameter exploration
- Update training pipeline: batch iterator, weight export improvements
- Retrain ensemble weights (scanner 99.73%, DDoS 99.99% val accuracy)
- Add unified audit log module
- Update dataset parsers with copyright headers and minor fixes

Signed-off-by: Sienna Meridian Satterwhite <sienna@sunbeam.pt>
 2026-03-10 23:38:22 +00:00
Sienna Meridian Satterwhite
385e9d4c59 chore: add SPDX copyright headers and update license year 
Add `// Copyright Sunbeam Studios 2026` and `// SPDX-License-Identifier:
Apache-2.0` headers to all source files missing them. Update LICENSE
copyright year, Dockerfile copyright header, and .dockerignore for new
project structure (lean4/, docs/, training artifacts).

Signed-off-by: Sienna Meridian Satterwhite <sienna@sunbeam.pt>
 2026-03-10 23:38:21 +00:00
Sienna Meridian Satterwhite
067d822244 feat(training): add burn MLP and CART tree trainers with weight export 
Behind the `training` feature flag (burn 0.20 + ndarray + autodiff).
Trains a single-hidden-layer MLP with Adam optimizer and weighted BCE
loss, plus a CART decision tree using Gini impurity. Exports trained
weights as Rust const arrays that compile directly into the binary.

Signed-off-by: Sienna Meridian Satterwhite <sienna@sunbeam.pt>
 2026-03-10 23:38:21 +00:00