--- title: ci-proto emoji: 📈 colorFrom: red colorTo: blue sdk: gradio sdk_version: 6.31.0 app_file: app.py short_description: Zero-shot probabilistic time-series forecasting python_version: "3.10" startup_duration_timeout: 30m --- # ci-proto — zero-shot probabilistic forecasting Interactive demo of [ContinuousIntelligence/ci-proto](https://huggingface.co/ContinuousIntelligence/ci-proto), a 167M-parameter time-series foundation model for zero-shot probabilistic forecasting. Give it a CSV with a timestamp column and a value column and it returns the median forecast plus nested quantile bands (0.1–0.9) for every future step. ## How it works - **CSV with a `split` column**: rows marked `history` are observed; the rows marked `forecast` that follow them set the horizon and its timestamps. - **Plain CSV**: every row is history. Pick a horizon; future timestamps continue the frequency inferred from the history. The bundled example is five days of hourly bicycle rentals in Washington, D.C. followed by the 24 hours to forecast. The forecasting logic mirrors the authors' `examples/forecast.py` reference implementation. ## Credits - The `continuousintelligence` package is vendored from [ContinuousIntelligence/ci-proto](https://huggingface.co/ContinuousIntelligence/ci-proto) (Apache-2.0; see `VENDOR_LICENCE` and `VENDOR_NOTICE`). - The bundled `bike_sharing.csv` example is from the UCI Bike Sharing dataset (Hadi Fanaee-T, 2013), licensed **CC BY 4.0**, DOI [10.24432/C5W894](https://doi.org/10.24432/C5W894). Please cite: Fanaee-T, H. & Gama, J. (2013). *Event labeling combining ensemble of weather and air quality datasets*, PAI.