Nbeats

polars_ts.dl.nbeats

N-BEATS: Neural Basis Expansion Analysis for Time Series.

Implements the N-BEATS architecture (Oreshkin et al., ICLR 2020) with both generic and interpretable (trend + seasonality) stack types.

References

Oreshkin et al. (2020). N-BEATS: Neural basis expansion analysis for interpretable time series forecasting. ICLR.

_GenericBlock

Bases: Module

Generic N-BEATS block with learnable basis.

_TrendBlock

Bases: Module

Interpretable trend block using polynomial basis.

_SeasonalityBlock

Bases: Module

Interpretable seasonality block using Fourier basis.

_NBEATSNet

Bases: Module

Full N-BEATS network.

NBEATSForecaster

N-BEATS time series forecaster.

Parameters

h Forecast horizon. input_size Lookback window size. hidden_size Hidden layer size in each block. n_stacks Number of stacks (only used with default stack_types). n_blocks Number of blocks per stack. stack_types List of stack types: "generic", "trend", "seasonality". If None, uses n_stacks generic stacks. max_epochs Maximum training epochs. lr Learning rate. batch_size Training batch size. id_col, time_col, target_col Column names.

fit(df)

Train the N-BEATS model on historical data.

Parameters

df Panel DataFrame with historical observations.

Returns

NBEATSForecaster Self, for chaining.

predict(df)

Generate forecasts for each series.

Parameters

df Panel DataFrame (uses last input_size observations per series).

Returns

pl.DataFrame Columns: [id_col, time_col, y_hat].