---

name: time-series-forecasting description: "Use Lazy Predict LazyForecaster for time-series model benchmarking, exogenous forecasting, seasonal detection, forecasting metrics, horizon strategies, ensembles, diagnostics, plotting, and optional forecasting dependencies." disable-model-invocation: true metadata: disco-role: operating license: MIT

Time-series Forecasting

Use this sub-skill when the task asks for Lazy Predict forecasting with LazyForecaster, including baseline and sklearn lag-feature forecasters, statistical or deep optional forecasters, exogenous variables, time-series metrics, horizon strategies, ensembles, diagnostics, or plotting.

Start here

1. Confirm the user has ordered numeric observations split into y_train and y_test. Lazy Predict forecasts a future horizon of len(y_test).

2. Start with a bounded CPU-safe model list such as ['Naive', 'Ridge_TS'].

3. Add optional statistical, boosting, deep-learning, or foundation models only when dependencies and runtime budget are available.

4. Run the bundled smoke helper when the environment or basic contract is in doubt:

   python scripts/smoke_forecasting.py --predictions --exogenous
   

Main API

Read references/api-reference.md for the verified LazyForecaster signature, model categories, helper functions, and result columns.

Typical bounded workflow:

from lazypredict.TimeSeriesForecasting import LazyForecaster

fcst = LazyForecaster(
    forecasters=['Naive', 'Ridge_TS'],
    predictions=True,
    n_lags=5,
    verbose=0,
    ignore_warnings=True,
)
scores, predictions = fcst.fit(y_train, y_test)

For exogenous features, pass matrices aligned with the train and forecast periods:

scores, predictions = fcst.fit(y_train, y_test, X_train, X_test)

Workflows

Read references/workflows.md for quick starts, exogenous variables, model subsets, cross-validation, custom metrics, horizon strategies, ensembles, diagnostics, visualization, save/load, and forecasting from loaded models.

Troubleshooting

Read references/troubleshooting.md for too-short series, lag/window errors, exogenous shape mismatches, missing optional forecasting dependencies, invalid sort/tuning metrics, TimesFM local weight issues, GPU fallback, and plotting failures.

Route elsewhere

• Use supervised-benchmarking for LazyClassifier or LazyRegressor tabular supervised tasks.
• Use advanced-workflows for detailed tuning, search-space, SHAP, and optional advanced-dependency choices.
• Use cli-and-integrations for package install checks, MLflow integration, Spark, Dask/PySpark, and the supervised CSV CLI. The CLI does not provide the LazyForecaster workflow.