API Reference¶

Complete API reference for all Yohou-Nixtla classes and functions. Use the search box to filter, or click any name to see full documentation.

Quick Reference¶

Wrap Nixtla's statsforecast library. Fast and interpretable, best for structured seasonal data.

Class	Model	Best For
`AutoARIMAForecaster`	AutoARIMA	General-purpose, auto model selection
`AutoETSForecaster`	AutoETS	Trend and seasonality, auto selection
`AutoCESForecaster`	AutoCES	Complex exponential smoothing
`AutoThetaForecaster`	AutoTheta	Automatic Theta model selection
`ARIMAForecaster`	ARIMA	Manual ARIMA order specification
`HoltWintersForecaster`	HoltWinters	Manually specified ETS components
`ThetaForecaster`	Theta	Manually specified Theta model
`NaiveForecaster`	Naive	Baseline: repeat last value
`SeasonalNaiveForecaster`	SeasonalNaive	Baseline: repeat last season
`CrostonForecaster`	Croston	Intermittent (sparse) demand

Wrap Nixtla's neuralforecast library. Suited for complex patterns and large datasets. Require pip install yohou-nixtla[neural].

Class	Model	Best For
`NBEATSForecaster`	N-BEATS	General deep learning baseline
`NHITSForecaster`	N-HiTS	Long-horizon forecasting
`MLPForecaster`	MLP	Simple feedforward baseline
`PatchTSTForecaster`	PatchTST	Transformer, channel-independent
`TimesNetForecaster`	TimesNet	CNN-based temporal modeling

Function	Module	Description
`infer_freq(y)`	`yohou_nixtla._conversion`	Map polars DataFrame time interval to Nixtla frequency alias
`yohou_to_nixtla(y, X)`	`yohou_nixtla._conversion`	Convert Yohou wide-format to Nixtla long-format
`nixtla_to_yohou(forecast_df, y_columns)`	`yohou_nixtla._conversion`	Convert Nixtla long-format predictions to Yohou wide-format

All forecasters inherit these parameters from BaseNixtlaForecaster:

Parameter	Type	Default	Description
`freq`	`str \\| None`	`None`	Frequency string. Auto-inferred from data if `None`.
`feature_transformer`	transformer or `None`	`None`	Applied to exogenous features before fitting and predicting.
`target_transformer`	transformer or `None`	`None`	Applied to target before fitting; inverse-applied after predicting.
`target_as_feature`	`"transformed" \\| "raw" \\| None`	`None`	Include lagged target values as additional features.

Stats forecasters additionally accept:

Parameter	Type	Default	Description
`season_length`	`int`	`1`	Seasonal period length.
`n_jobs`	`int`	`1`	Parallel jobs for multi-series fitting.

Neural forecasters additionally accept:

Parameter	Type	Default	Description
`input_size`	`int`	`24`	Lookback window (number of past steps).
`max_steps`	`int`	`100`	Maximum training steps.

All forecasters implement the BasePointForecaster lifecycle:

Method	Signature	Description
`fit`	`fit(y, X=None, forecasting_horizon=1)`	Train on historical data.
`predict`	`predict(forecasting_horizon, X=None)`	Generate point forecasts.
`observe`	`observe(y_new)`	Append new observations without retraining.
`rewind`	`rewind(y)`	Reset the internal observation state.

Attribute	Type	Description
`nixtla_forecaster_`	`StatsForecast \\| NeuralForecast`	The fitted Nixtla orchestrator.
`freq_`	`str`	The inferred or provided frequency string.
`y_columns_`	`list[str]`	Target column names from training data.
`instance_`	model instance	The constructed backend model instance.

Name	Type	Module	Description
`ARIMAForecaster`	Class	yohou_nixtla.stats	ARIMA forecaster via statsforecast.
`AutoARIMAForecaster`	Class	yohou_nixtla.stats	AutoARIMA forecaster via statsforecast.
`AutoCESForecaster`	Class	yohou_nixtla.stats	AutoCES (Complex Exponential Smoothing) forecaster via statsforecast.
`AutoETSForecaster`	Class	yohou_nixtla.stats	AutoETS forecaster via statsforecast.
`AutoThetaForecaster`	Class	yohou_nixtla.stats	AutoTheta forecaster via statsforecast.
`BaseStatsForecaster`	Class	yohou_nixtla.stats	Base class for statsforecast model wrappers in yohou.
`CrostonForecaster`	Class	yohou_nixtla.stats	Croston's method forecaster via statsforecast.
`HoltWintersForecaster`	Class	yohou_nixtla.stats	Holt-Winters (triple exponential smoothing) forecaster via statsforecast.
`NaiveForecaster`	Class	yohou_nixtla.stats	Naive forecaster via statsforecast.
`SeasonalNaiveForecaster`	Class	yohou_nixtla.stats	Seasonal Naive forecaster via statsforecast.
`ThetaForecaster`	Class	yohou_nixtla.stats	Theta forecaster via statsforecast.