Hugging Face
Constellation publishes open datasets on Hugging Face. The first is the demo fleet: the same satellites, ground stations, and links the API loads with load_demo_fleet, so you can study the data offline and then forecast it with the API.
What Hugging Face is
Hugging Face is a public website where companies and researchers publish datasets and machine learning models, much like an app store for data. Each dataset has a page that explains what it holds, a viewer that shows the rows in the browser, and download links. Viewing and downloading public datasets needs no account.
If you have not used it before, you only need two things from it: the dataset page and the Files and versions tab.
| Link | What you find there |
|---|---|
| Constellation on Hugging Face | Our organization page, with everything we publish |
| Demo fleet dataset | The demo fleet as CSV and JSON files, with column documentation |
What the demo fleet dataset holds
A snapshot of each entity's newest values: 6 ground stations, 39 satellites, 21 ground-antenna links, and 4 traffic demand pools.
| File | Rows | What it holds |
|---|---|---|
links.csv | 21 | One ground antenna to one satellite: pointing, geometry, RF configuration, and observed SNR |
ground_stations.csv | 6 | Station name, WGS84 latitude and longitude, and the demand pool it serves |
satellites.csv | 39 | NORAD catalog ID and name |
demand_pools.csv | 4 | Latest ten-minute offered traffic per pool |
topology.json | 70 entities | The same snapshot in the shape GET /topology returns |
The dataset page documents every column and its unit. Demand in the demo fleet is synthetic, not measured traffic. One link, to PROBA-3 at about 48,800 km, sits at -12.5 dB SNR and is kept as an example of a link that does not close.
Download it
In the browser. Open the dataset page, choose Files and versions, and download any file. The CSV files open in Excel, Google Sheets, or any spreadsheet.
In Python. Install the datasets library with pip install datasets, then load a table by name:
from datasets import load_dataset
links = load_dataset("constellationspace/demo-fleet", "links", split="train")
print(links.to_pandas().sort_values("snr_db").head())
The table names are links, ground_stations, satellites, and demand_pools.
Forecast the same fleet
The dataset is a static snapshot. To forecast the fleet, load it into your account: ask an AI agent to call load_demo_fleet, which writes two hours of SNR telemetry and seven days of demand history ending at the present. Then request predictions as in the quick start.
License
The dataset is published under CC BY 4.0. Credit "Constellation Space" when you use it.