[ENH] direct/recursive forecasting

aeon/forecasting/_multiple.py

@@ -0,0 +1,53 @@
+"""Container for direct forecasting."""
+
+import numpy as np
+
+from aeon.forecasting.base import BaseForecaster
+
+
+def direct_forecasting(
+    forecaster: BaseForecaster, y: np.ndarray, steps_ahead: int, exog=None
+):
+    """Forecast steps_ahead points from X using forecaster.
+
+    Parameters
+    ----------
+    forecaster : BaseForecaster class or object.
+        Forecaster object with forecast method.
+    y : 1D np.ndarray
+        Time series to train forecasters on.
+    steps_ahead : int
+        Number of steps ahead to forecast.
+    window_size : int
+    Fits a different forecaster to each horizon
+
+    Returns
+    -------
+    np.ndarray
+        Length steps_ahead array of forecasts.
+    """
+    if not isinstance(forecaster, BaseForecaster):
+        raise ValueError("Forecaster must be a BaseForecaster object.")
+    preds = np.zeros(steps_ahead)
+    for i in range(1, steps_ahead + 1):
+        f = forecaster.__class__(horizon=i, **forecaster.params)
+        preds[i - 1] = f.forecast(y, exog)
+    return preds
+
+
+def recursive_forecasting(
+    forecaster: BaseForecaster, y: np.ndarray, steps_ahead: int, window=1, exog=None
+):
+    """Forecast steps_ahead points from X using forecaster."""
+    if not isinstance(forecaster, BaseForecaster):
+        raise ValueError("Forecaster must be a BaseForecaster object.")
+    if hasattr(forecaster, "window"):
+        forecaster = forecaster.__class__(horizon=1, **forecaster.params)
+    forecaster.fit(y, exog)
+    preds = np.zeros(steps_ahead)
+    y = y[-window:]
+    for i in range(1, steps_ahead + 1):
+        preds[i - 1] = forecaster.predict(y)
+        y[:-1] = y[1:]
+        y[window - 1] = preds[i - 1]
+    return preds