📈 TiRex - timeseries forecasting 📊

Upload data or choose a preset, filter by name, then click Plot.

Data Selection

Or choose a preset:

Forecast Length Setting

1 512

Transpose data

Search / Filter

Select which timeseries to show

Forecast Plot

Instructions

How to format your data:

  • Your file must be a table (CSV, XLS, XLSX, or Parquet).
  • One row per timeseries. Each row is treated as a separate series.
  • If you want to name each series, put the name as the first value in every row:
    • Example (CSV):
      AAPL, 120.5, 121.0, 119.8, ...
      AMZN, 3300.0, 3310.5, 3295.2, ...
    • In that case, the first column is not numeric, so it will be used as the series name.
  • If you do not want named series, simply leave out the first column entirely and have all values numeric:
    • Example:
      120.5, 121.0, 119.8, ...
      3300.0, 3310.5, 3295.2, ...
    • Then every row will be auto-named “Series 0, Series 1, …” in order.
  • Consistency rule: Either all rows have a non-numeric first entry for the name, or none do. Do not mix.
  • The rest of the columns (after the optional name) must be numeric data points for that series.
  • You can filter by typing in the search box. Then check or uncheck individual names before plotting.
  • Use “Check All” / “Uncheck All” to quickly select or deselect every series.
  • Finally, click Plot Forecasts to view the quantile forecast for each selected series (for 64 steps ahead).

Citation

If you use TiRex in your research, please cite our work:

@inproceedings{auer:25tirex,
title = {{{TiRex}}: {{Zero-Shot Forecasting Across Long}} and {{Short Horizons}} with {{Enhanced In-Context Learning}}},
author = {Andreas Auer and Patrick Podest and Daniel Klotz and Sebastian B{\"o}ck and G{\"u}nter Klambauer and Sepp Hochreiter},
booktitle = {The Thirty-Ninth Annual Conference on Neural Information Processing Systems},
year = {2025},
url = {https://arxiv.org/abs/2505.23719},
}