title	tsseg
emoji	📈
colorFrom	indigo
colorTo	blue
sdk	gradio
sdk_version	6.6.0
python_version	3.10
app_file	app.py
pinned	false

tsseg-app — Interactive Time Series Segmentation

An interactive web application for Time Series Segmentation, powered by tsseg and deployed on HuggingFace Spaces.

Features

• 30+ segmentation algorithms — Change Point Detection & State Detection
• Upload your own data — CSV with optional ground truth labels
• Built-in datasets — MoCap (CMU 86) and synthetic generators
• Dynamic parameter tuning — interactive sliders, dropdowns, and text inputs
• Multi-algorithm comparison — side-by-side visual comparison
• Standard metrics — F1, Covering, ARI, NMI, SMS, and more
• Export results — download predictions and scores as CSV

Quick Start

Online (HuggingFace Spaces)

👉 huggingface.co/spaces/fchavelli/tsseg

Local

git clone https://github.com/fchavelli/tsseg-app.git
cd tsseg-app
make install
make run

The app will be available at http://localhost:7860.

Architecture

tsseg-app/
├── app.py                  # Main Gradio application
├── requirements.txt        # Dependencies (for HF Spaces)
├── pyproject.toml          # Project metadata
├── Makefile                # Dev commands
├── src/
│   ├── algorithms.py       # Detector discovery & introspection
│   ├── data.py             # Data loading (synthetic, MoCap, CSV)
│   ├── metrics.py          # Evaluation metrics wrapper
│   ├── params.py           # Dynamic parameter rendering
│   ├── plotting.py         # Plotly visualization
│   ├── utils.py            # Helpers (state conversion, z-norm, etc.)
│   └── config/             # YAML configs with tunable parameter ranges
│       ├── pelt.yaml
│       ├── binseg.yaml
│       └── ...
├── assets/                 # Logo and static files
├── tests/                  # Test suite
└── .github/workflows/
    └── sync-hf.yml         # CI: auto-deploy to HuggingFace Spaces

Tabs

Tab	Description
Data	Load synthetic, MoCap, or upload CSV data
Segmentation	Choose algorithm, tune parameters, run detection
Comparison	Compare multiple algorithm runs side by side
Evaluation	Compute F1, ARI, SMS, etc. when ground truth is available
About	Full list of available algorithms and links

GPU Algorithms on CPU

Algorithms requiring PyTorch (TIRE, TGLAD, E2USD, Time2State, VQTSS) and TensorFlow (TSCP2) are included and run on CPU. They will be slower than on GPU but fully functional. The environment variable CUDA_VISIBLE_DEVICES="" is set automatically.

Development

make install    # Create conda env + install deps
make run        # Start the app locally
make test       # Run tests
make lint       # Check style with ruff

Deployment

The app auto-deploys to HuggingFace Spaces via GitHub Actions on push to main. Set the HF_TOKEN secret in your GitHub repo settings.

License

AGPLv3

Built with Gradio · Powered by tsseg