---

name: components-and-exogenous description: "Configure NeuralProphet trend, seasonality, autoregression, regressors, events, holidays, and multi-series models." disable-model-invocation: true metadata: disco-role: operating license: MIT

Components and exogenous features

Use this sub-skill when the task is to change the forecast structure rather than merely run a basic fit: trend and changepoints, seasonalities, autoregression, lagged regressors, future-known regressors, custom events, holidays, conditional seasonality, or global/local multi-series modeling.

Load this when the user asks to

• Add or tune trend, changepoints, yearly/weekly/daily/custom seasonalities, additive versus multiplicative components.
• Use n_lags, ar_reg, ar_layers, lagged regressors, or sparse autoregression.
• Add future regressors that are known for forecast periods.
• Add custom events or country holidays with lower/upper windows and regularization.
• Fit global, local, or glocal components with an ID column for multiple time series.
• Explain why prediction fails after adding a regressor, event, or conditional seasonality column.

Start here

1. Read component-recipes.md for task-oriented recipes and dataframe requirements.
2. Read api-reference.md for verified constructor and add_* signatures.
3. Read troubleshooting.md for missing future regressor columns, invalid component names, holiday/country issues, lag alignment, and global/local mistakes.
4. Run scripts/smoke_components.py for a tiny CPU check that exercises future regressors, custom events, and multi-ID data.

Boundaries

Stay here for component design and exogenous columns. Route elsewhere for:

• Base dataframe validation, fit, predict, and future dataframe mechanics: ../core-forecasting/SKILL.md.
• Validation/test/CV, metrics, quantile regression, and conformal prediction: ../evaluation-and-uncertainty/SKILL.md.
• Plotting backends, save/load, CLI/version, logging/seeding, accelerators, and TorchProphet migration: ../operations-and-migration/SKILL.md.

Fast component checklist

• Add all components before calling fit; a fitted model is not meant to be reconfigured and fitted again casually.
• Every extra column configured for training must be present when fitting; every future-known regressor needed for future timestamps must be supplied to make_future_dataframe through regressors_df.
• Event workflows need an event table with event and ds, then history data expanded with create_df_with_events when appropriate.
• Multi-series/global models need an ID column and consistent component choices for global, local, or glocal behavior.
• Keep component names distinct from reserved columns and existing component/event/regressor names.