Skip to content

Chronos-2 pretrained forecasting model

Source: R/tabular_chronos.R
tabular_chronos.Rd

tabular_chronos() defines a pretrained time-series forecasting model that produces quantile (distributional) forecasts. The network has fixed pretrained weights, so no training is performed; the historical ("context") data is ingested at fit time and the model forecasts a fixed horizon. This function can fit quantile regression and regression models.

There are different ways to fit this model, and the method of estimation is chosen by setting the model engine. The engine-specific pages for this model are listed below.

¹ The default engine. ² Requires a parsnip extension package for quantile regression and regression.

More information on how parsnip is used for modeling is at https://www.tidymodels.org/.

Usage

tabular_chronos(mode = "unknown", engine = "brulee")

Arguments

mode

A single character string for the type of model. The possible values for this model are "quantile regression" (the natural mode, which returns a hardhat::quantile_pred()) and "regression" (which returns the median point forecast). The mode must be set before fitting; for "quantile regression" it is set with set_mode("quantile regression", quantile_levels = ...).

engine

A single character string specifying what computational engine to use for fitting. The only valid value is "brulee".

Details

Unlike the other models in this package, Chronos-2 is pretrained and has no tuning parameters. Forecast configuration is supplied through the engine with set_engine(), e.g. set_engine("brulee", prediction_length = 14). The available engine arguments mirror brulee::brulee_chronos(): prediction_length, id_column, timestamp_column, model_id, revision, device, and cache_dir. The quantile_levels are taken from the mode (via set_mode()) and forwarded to the fit automatically.

On first use the engine downloads the pretrained weights (about 500MB) and caches them locally.

The parsnip interface forecasts a single series. predict() returns one row per horizon step (.pred_quantile or .pred), which cannot unambiguously represent more than one series; supplying data with multiple id_column values is therefore an error. For multi-series forecasting, call brulee::brulee_chronos() directly, where the id column is retained in the output.

References

Ansari, A. F., Shchur, O., Küken, J., Auer, A., Han, B., Mercado, P., et al. (2025). "Chronos-2: From univariate to universal forecasting." arXiv preprint arXiv:2510.15821.

See also

fit(), set_engine(), update(), brulee engine details brulee::brulee_chronos()

Examples

show_engines("tabular_chronos")
#> # A tibble: 0 × 2
#> # ℹ 2 variables: engine <chr>, mode <chr>

# Quantile (distributional) forecast
tabular_chronos() |>
  set_engine("brulee", prediction_length = 14) |>
  set_mode("quantile regression", quantile_levels = (1:9) / 10)
#> ! parsnip could not locate an implementation for `tabular_chronos`
#>   quantile regression model specifications using the `brulee` engine.
#> ℹ The parsnip extension package tabby implements support for this
#>   specification.
#> ℹ Please install (if needed) and load to continue.
#> tabular chronos Model Specification (quantile regression)
#> 
#> Engine-Specific Arguments:
#>   prediction_length = 14
#> 
#> Computational engine: brulee 
#> 
#> Quantile levels: 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, and 0.9.

# Median point forecast
tabular_chronos() |>
  set_engine("brulee", prediction_length = 14) |>
  set_mode("regression")
#> ! parsnip could not locate an implementation for `tabular_chronos`
#>   regression model specifications using the `brulee` engine.
#> ℹ The parsnip extension package tabby implements support for this
#>   specification.
#> ℹ Please install (if needed) and load to continue.
#> tabular chronos Model Specification (regression)
#> 
#> Engine-Specific Arguments:
#>   prediction_length = 14
#> 
#> Computational engine: brulee 
#>