ic_new

A modern rewrite of the Interactive Classifier — a tool for interactively classifying chant-manuscript neumes using a k-Nearest Neighbors model.

This project replaces the legacy Rodan job (Django + Celery + Gamera + Backbone.Marionette) at [/Rodan-lite/backend/django/code/jobs/interactive_classifier/] with a non-Django Python service, a React + Vite frontend, and no Gamera dependency.

Key differences from the legacy IC

• Input is a full page image plus a bounding-box annotation file (for example, MOTHRA JSON or YOLO), not page-level GameraXML. Neume crops are derived from those annotations rather than supplied as pre-cropped glyph images.
• Manual split (CCA on a single glyph) is deferred. Reintroduce only if real data shows crops that contain multiple neumes.
• Output stays as GameraXML so downstream MEI pipelines keep working.

Status

Layer	Path	Status
Algorithm core	core/ic_core/	In progress — scaffolded, partial impl
API	api/	In progress, open to future change
Frontend	frontend/	Not started

Repository layout

ic_new/
├── core/
│   ├── ic_core/        # Phase 1: algorithm core (uv-managed Python package)
│   └── tests/          # Pytest suite + fixtures
├── api/                # Phase 2: FastAPI service (planned)
├── frontend/           # Phase 3: React + Vite UI (planned)
└── docs/
    ├── CLAUDE.md       # Guidance for Claude Code working in this repo
    ├── migration_plan.md
    └── KNN_ALGORITHM.md

Quickstart (algorithm core)

The core package uses uv for environment and dependency management.

cd core/ic_core
uv sync                 # install dependencies
uv run pytest ../tests  # run tests
uv run ruff check .     # lint

Documentation

• docs/migration_plan.md — full migration strategy, phasing, and risks
• docs/KNN_ALGORITHM.md — algorithm spec and invariants
• docs/CLAUDE.md — architecture notes and conventions for AI-assisted development