Hexagonal neural network experiments — CLI, Streamlit workbench, and benchmark suites.
Hexagonal Neural Networks (HexNets) generalize MLPs into a rotationally symmetric, multi-directional structure. Six orientations (H_0 \ldots H_5) share one weight matrix. Full write-up: docs/latex/main.pdf.
make install
hexnet --help
make streamlit-run
make unit-test| Task | Where |
|---|---|
| CLI flags & commands | .cursor/cli/CLI_PATTERNS.md |
| Run directories | .cursor/architecture/FILE_STRUCTURE.md |
| Benchmarks (Roman I–III) | docs/math/benchmark-families.md |
| Streamlit workbench | hexnets.tennisgazelle.com · .cursor/web/hexnets_web.md |
| Architecture depth | .cursor/architecture/ARCHITECTURE.md |
| Agents | AGENTS.md · heartbeat.md |
Layout: src/ (application), runs/ / reference/ / figures/ (runtime). Legacy notebook code: src/obsolete/ (not imported).
| Audience | Location |
|---|---|
| Humans (you are here) | README.md, docs/ |
| Agents | AGENTS.md → .cursor/ |
| Contributing | CONTRIBUTING.md |