graphtoolbox.models.baseline

Reusable non-graph forecasting baselines used by the examples.

graphtoolbox.models.baseline.rmse(prediction, target) → float

Root mean squared error.

graphtoolbox.models.baseline.mape(prediction, target, eps: float = 1e-06) → float

Mean absolute percentage error, in percent.

graphtoolbox.models.baseline.smape(prediction, target, eps: float = 1e-08) → float

Symmetric mean absolute percentage error, in percent.

graphtoolbox.models.baseline.arimax_forecast(train: DataFrame, test: DataFrame, features: Sequence[str], target: str, slot_col: str, order=(2, 0, 2)) → ndarray

Fit one regression-with-ARMA-errors model per forecast slot.

class graphtoolbox.models.baseline.CovariateSeq2Seq(n_features: int, hidden_channels: int = 64, horizon: int = 48)

Bases: Module

Small LSTM encoder-decoder used as a covariate-aware baseline.

forward(history, future_covariates)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

graphtoolbox.models.baseline.lstm_forecast(train: DataFrame, test: DataFrame, features: Sequence[str], target: str, step: int = 48, hist_len: int = 336, hidden: int = 64, epochs: int = 40, batch: int = 64, lr: float = 0.001, seed: int = 42, device: str = 'cpu') → ndarray

Train a day-ahead seq2seq LSTM with known future covariates.

graphtoolbox.models.baseline.chronos2_forecast(full: DataFrame, t0: int, n_test: int, features: Sequence[str], target: str, step: int = 48, ctx: int = 1024, model_id: str = 'amazon/chronos-2', batch_size: int = 64) → ndarray

Run Chronos-2 zero-shot with past and known-future covariates.

graphtoolbox.models.baseline.xgboost_forecast(train: DataFrame, test: DataFrame, features: Sequence[str], target: str, **model_kwargs) → ndarray

Fit an XGBoost regressor and return its test forecast.

graphtoolbox.models.baseline.gam_forecast_by_slot(train: DataFrame, test: DataFrame, features: Sequence[str], target: str, slot_col: str, formula, max_iter: int = 1000) → ndarray

Fit one supplied pygam formula per forecast slot.

graphtoolbox.models.baseline.chronos_bolt_forecast(series, test_start: int, n_test: int, step: int = 48, context: int = 1024, model_id: str = 'amazon/chronos-bolt-small', batch_size: int = 32) → ndarray

Run a rolling zero-shot Chronos-Bolt forecast.

graphtoolbox.models.baseline.fit_tabular_baseline(model_type: str, x_train, y_train, x_test, x_validation=None, y_validation=None)

Fit a tabular baseline used for top-level reconciliation.

Parameters are already scaled by the caller. The return value is (model, train_prediction, validation_prediction, test_prediction).