Spaces:
Running
Real measured series as examples, and a 15-minute frequency option
The examples did not work: clicking one left the inputs empty and produced no forecast. The load_example_input endpoint errors while load_example succeeds, so the cached output existed but the inputs were never repopulated. This drops cache_examples/cache_mode and uses run_on_click=True, which loads the inputs and runs the model on click.
It also replaces the three placeholder CSVs with thirteen real measured series, ordered from easiest to hardest: aggregated national demand at one end, a single household at the other, with solar, wind, EV charging, a day-ahead price and a low-voltage feeder in between. They are drawn from the FETS dataset (10.5281/zenodo.19418721, CC BY) and Elia open data, and they make the interval behaviour visible: it stays tight on aggregated load and opens up on a single asset.
Adds 15-minute (15T) to the frequency map. Almost all European grid data is quarter-hourly and the dropdown had no option for it; TinyCast's seasonal scale factor already handles 15T correctly.