Web ML

Small self-hosted web application for uploading datasets, training classical and neural-network models, visualizing training progress, and inspecting experiments.

Key pieces

• Backend entry: backend/app.py — minimal runner that loads the app.
• HTTP endpoints and Socket.IO handlers: backend/endpoints.py.
• ML wrapper and model helpers: ModelWrapper (backend/ml/wrapper.py).
• DB models: backend/models.py.
• Shared extensions (db/socket/login): backend/extensions.py.
• Frontend: frontend/ (Vite + React + TypeScript).
• Frontend login/register pages: frontend/src/pages/login.tsx, frontend/src/pages/register.tsx.
• Training UI components: frontend/src/components/train-model-dialog.tsx, frontend/src/components/training-visualizer.tsx.

Requirements

• Python 3.10+ (backend)
• Node 18+ / npm or pnpm (frontend)
• Packages listed in requirements.txt and frontend/package.json

Quickstart — Backend

1. Create and activate a virtual environment and install dependencies:

# bash
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

3. Run the backend (includes Socket.IO):

# bash
python app.py

This will start the server on port 5050 by default. The Flask app and Socket.IO handlers are wired in backend/endpoints.py and extensions come from backend/extensions.py.

Quickstart — Frontend

1. Install dependencies and run dev server:

# bash
cd frontend
npm install
npm run dev

2. The frontend dev server (Vite) proxies /api to the backend using frontend/vite.config.ts.

Running tests

• Frontend unit tests use Jest. From the frontend directory:

# bash
cd frontend
npm test

• There are no automated backend tests included by default.

API highlights

• Dataset endpoints and upload are implemented in backend/endpoints.py (e.g., /api/upload, /api/datasets).
• Model train endpoint (non-MLP streaming/non-streaming) is at /api/train/<model_id> in backend/endpoints.py. For MLP streaming training the Socket.IO events (start_training, pause_training, resume_training, early_stop_training) are handled in the same file.
• Model serialization and evaluation helper is ModelWrapper.

Development notes

• The code stores models and metadata in SQLite at instance/app.db. Models are pickled into the DB via ModelWrapper.to_db_record.
• Frontend components expect the backend API at http://localhost:5050 (see frontend/vite.config.ts proxy).
• CORS is configured in the backend to allow the dev frontend origin (see backend/endpoints.py and backend/extensions.py).
• Large files are stored under uploads/ (configured via app.config['UPLOAD_FOLDER'] in backend/endpoints.py).

Where to start when changing code

• Add or modify endpoints in backend/endpoints.py or refactor into blueprints (recommended).
• Update ML logic in ModelWrapper if adding new model types or metrics.
• Edit UI behavior in the corresponding frontend component files under frontend/src/components/.

Reset / maintenance

• To reset uploads and DB for a clean slate:
• Stop the backend.
• Remove instance/app.db and contents of uploads/.
• Restart backend to recreate (some initialization logic runs on boot in backend/endpoints.py).

OR, send a POST request to the api/reset/both?key=supersecretresetkey endpoint

Help / debugging

• Backend logs are printed to stdout when running python backend/app.py.
• The frontend dev server prints proxy activity (see frontend/vite.config.ts configure hooks).
• If Socket.IO connections fail in the browser, ensure the backend is reachable and socket.io-client version matches server compatibility.

Contributing

• Follow the existing code structure. Consider splitting large files (for example split backend/endpoints.py into blueprints) and reuse shared extensions in backend/extensions.py.