autogluon.timeseries 1.2.1b20250425__py3-none-any.whl → 1.2.1b20250426__py3-none-any.whl

autogluon/timeseries/learner.py

@@ -29,15 +29,13 @@ class TimeSeriesLearner(AbstractLearner):
         known_covariates_names: Optional[List[str]] = None,
         trainer_type: Type[TimeSeriesTrainer] = TimeSeriesTrainer,
         eval_metric: Union[str, TimeSeriesScorer, None] = None,
-        eval_metric_seasonal_period: Optional[int] = None,
         prediction_length: int = 1,
         cache_predictions: bool = True,
         ensemble_model_type: Optional[Type] = None,
         **kwargs,
     ):
         super().__init__(path_context=path_context)
-        self.eval_metric: TimeSeriesScorer = check_get_evaluation_metric(eval_metric)
-        self.eval_metric_seasonal_period = eval_metric_seasonal_period
+        self.eval_metric = check_get_evaluation_metric(eval_metric, prediction_length=prediction_length)
         self.trainer_type = trainer_type
         self.target = target
         self.known_covariates_names = [] if known_covariates_names is None else known_covariates_names
@@ -82,7 +80,6 @@ class TimeSeriesLearner(AbstractLearner):
                 path=self.model_context,
                 prediction_length=self.prediction_length,
                 eval_metric=self.eval_metric,
-                eval_metric_seasonal_period=self.eval_metric_seasonal_period,
                 target=self.target,
                 quantile_levels=self.quantile_levels,
                 verbosity=kwargs.get("verbosity", 2),

autogluon/timeseries/metrics/__init__.py

@@ -1,5 +1,7 @@
 from pprint import pformat
-from typing import Type, Union
+from typing import Any, Dict, Optional, Sequence, Type, Union
+
+import numpy as np
 
 from .abstract import TimeSeriesScorer
 from .point import MAE, MAPE, MASE, MSE, RMSE, RMSLE, RMSSE, SMAPE, WAPE, WCD
@@ -22,7 +24,7 @@ __all__ = [
 
 DEFAULT_METRIC_NAME = "WQL"
 
-AVAILABLE_METRICS = {
+AVAILABLE_METRICS: Dict[str, Type[TimeSeriesScorer]] = {
     "MASE": MASE,
     "MAPE": MAPE,
     "SMAPE": SMAPE,
@@ -42,33 +44,59 @@ DEPRECATED_METRICS = {
 }
 
 # Experimental metrics that are not yet user facing
-EXPERIMENTAL_METRICS = {
+EXPERIMENTAL_METRICS: Dict[str, Type[TimeSeriesScorer]] = {
     "WCD": WCD,
 }
 
 
 def check_get_evaluation_metric(
-    eval_metric: Union[str, TimeSeriesScorer, Type[TimeSeriesScorer], None] = None
+    eval_metric: Union[str, TimeSeriesScorer, Type[TimeSeriesScorer], None],
+    prediction_length: int,
+    seasonal_period: Optional[int] = None,
+    horizon_weight: Optional[Sequence[float] | np.ndarray] = None,
 ) -> TimeSeriesScorer:
+    """Factory method for TimeSeriesScorer objects.
+
+    Returns
+    -------
+    scorer :
+        A `TimeSeriesScorer` object based on the provided `eval_metric`.
+
+        `scorer.prediction_length` is always set to the `prediction_length` provided to this method.
+
+        If `seasonal_period` is not `None`, then `scorer.seasonal_period` is set to this value. Otherwise the original
+        value of `seasonal_period` is kept.
+
+        If `horizon_weight` is not `None`, then `scorer.horizon_weight` is set to this value. Otherwise the original
+        value of `horizon_weight` is kept.
+    """
     scorer: TimeSeriesScorer
+    metric_kwargs: Dict[str, Any] = dict(
+        prediction_length=prediction_length, seasonal_period=seasonal_period, horizon_weight=horizon_weight
+    )
     if isinstance(eval_metric, TimeSeriesScorer):
         scorer = eval_metric
+        scorer.prediction_length = prediction_length
+        if seasonal_period is not None:
+            scorer.seasonal_period = seasonal_period
+        if horizon_weight is not None:
+            scorer.horizon_weight = scorer.check_get_horizon_weight(horizon_weight, prediction_length=prediction_length)
     elif isinstance(eval_metric, type) and issubclass(eval_metric, TimeSeriesScorer):
         # e.g., user passed `eval_metric=CustomMetric` instead of `eval_metric=CustomMetric()`
-        scorer = eval_metric()
+        scorer = eval_metric(**metric_kwargs)
     elif isinstance(eval_metric, str):
         metric_name = DEPRECATED_METRICS.get(eval_metric, eval_metric).upper()
         if metric_name in AVAILABLE_METRICS:
-            scorer = AVAILABLE_METRICS[metric_name]()
+            scorer = AVAILABLE_METRICS[metric_name](**metric_kwargs)
         elif metric_name in EXPERIMENTAL_METRICS:
-            scorer = EXPERIMENTAL_METRICS[metric_name]()
+            scorer = EXPERIMENTAL_METRICS[metric_name](**metric_kwargs)
         else:
             raise ValueError(
                 f"Time series metric {eval_metric} not supported. Available metrics are:\n"
                 f"{pformat(sorted(AVAILABLE_METRICS.keys()))}"
             )
     elif eval_metric is None:
-        scorer = AVAILABLE_METRICS[DEFAULT_METRIC_NAME]()
+        scorer = AVAILABLE_METRICS[DEFAULT_METRIC_NAME](**metric_kwargs)
     else:
         raise ValueError(
             f"eval_metric must be of type str, TimeSeriesScorer or None "

autogluon/timeseries/metrics/abstract.py

@@ -1,6 +1,8 @@
-from typing import Optional, Tuple, Union
+import warnings
+from typing import Optional, Sequence, Tuple, Union, overload
 
 import numpy as np
+import numpy.typing as npt
 import pandas as pd
 
 from autogluon.timeseries import TimeSeriesDataFrame
@@ -15,6 +17,18 @@ class TimeSeriesScorer:
 
     Follows the design of ``autogluon.core.metrics.Scorer``.
 
+    Parameters
+    ----------
+    prediction_length : int, default = 1
+        The length of the forecast horizon. The predictions provided to the `TimeSeriesScorer` are expected to contain
+        a forecast for this many time steps for each time series.
+    seasonal_period : int or None, default = None
+        Seasonal period used to compute some evaluation metrics such as mean absolute scaled error (MASE). Defaults to
+        `None`, in which case the seasonal period is computed based on the data frequency.
+    horizon_weight : Sequence[float], np.ndarray or None, default = None
+        Weight assigned to each time step in the forecast horizon when computing the metric. If provided, the
+        `horizon_weight` will be stored as a numpy array of shape `[1, prediction_length]`.
+
     Attributes
     ----------
     greater_is_better_internal : bool, default = False
@@ -40,6 +54,18 @@ class TimeSeriesScorer:
     needs_quantile: bool = False
     equivalent_tabular_regression_metric: Optional[str] = None
 
+    def __init__(
+        self,
+        prediction_length: int = 1,
+        seasonal_period: Optional[int] = None,
+        horizon_weight: Optional[Sequence[float]] = None,
+    ):
+        self.prediction_length = int(prediction_length)
+        if self.prediction_length < 1:
+            raise ValueError(f"prediction_length must be >= 1 (received {prediction_length})")
+        self.seasonal_period = seasonal_period
+        self.horizon_weight = self.check_get_horizon_weight(horizon_weight, prediction_length=prediction_length)
+
     @property
     def sign(self) -> int:
         return 1 if self.greater_is_better_internal else -1
@@ -66,18 +92,25 @@ class TimeSeriesScorer:
         self,
         data: TimeSeriesDataFrame,
         predictions: TimeSeriesDataFrame,
-        prediction_length: int = 1,
         target: str = "target",
-        seasonal_period: Optional[int] = None,
         **kwargs,
     ) -> float:
-        seasonal_period = get_seasonality(data.freq) if seasonal_period is None else seasonal_period
+        seasonal_period = get_seasonality(data.freq) if self.seasonal_period is None else self.seasonal_period
+
+        if "prediction_length" in kwargs:
+            warnings.warn(
+                "Passing `prediction_length` to `TimeSeriesScorer.__call__` is deprecated and will be removed in v2.0. "
+                "Please set the `eval_metric.prediction_length` attribute instead.",
+                category=FutureWarning,
+            )
+            self.prediction_length = kwargs["prediction_length"]
+            self.horizon_weight = self.check_get_horizon_weight(self.horizon_weight, self.prediction_length)
 
-        data_past = data.slice_by_timestep(None, -prediction_length)
-        data_future = data.slice_by_timestep(-prediction_length, None)
+        data_past = data.slice_by_timestep(None, -self.prediction_length)
+        data_future = data.slice_by_timestep(-self.prediction_length, None)
 
         assert not predictions.isna().any().any(), "Predictions contain NaN values."
-        assert (predictions.num_timesteps_per_item() == prediction_length).all()
+        assert (predictions.num_timesteps_per_item() == self.prediction_length).all()
         assert data_future.index.equals(predictions.index), "Prediction and data indices do not match."
 
         try:
@@ -200,3 +233,40 @@ class TimeSeriesScorer:
         q_pred = pd.DataFrame(predictions[quantile_columns])
         quantile_levels = np.array(quantile_columns, dtype=float)
         return y_true, q_pred, quantile_levels
+
+    @overload
+    @staticmethod
+    def check_get_horizon_weight(horizon_weight: None, prediction_length: int) -> None: ...
+    @overload
+    @staticmethod
+    def check_get_horizon_weight(
+        horizon_weight: Sequence[float] | np.ndarray, prediction_length: int
+    ) -> npt.NDArray[np.float64]: ...
+
+    @staticmethod
+    def check_get_horizon_weight(
+        horizon_weight: Sequence[float] | np.ndarray | None, prediction_length: int
+    ) -> Optional[npt.NDArray[np.float64]]:
+        """Convert horizon_weight to a non-negative numpy array that sums up to prediction_length.
+        Raises an exception if horizon_weight has an invalid shape or contains invalid values.
+
+        Returns
+        -------
+        horizon_weight:
+            None if the input is None, otherwise a numpy array of shape [1, prediction_length].
+        """
+        if horizon_weight is None:
+            return None
+        horizon_weight_np = np.ravel(horizon_weight).astype(np.float64)
+        if horizon_weight_np.shape != (prediction_length,):
+            raise ValueError(
+                f"horizon_weight must have length equal to {prediction_length=} (got {len(horizon_weight)=})"
+            )
+        if not (horizon_weight_np >= 0).all():
+            raise ValueError(f"All values in horizon_weight must be >= 0 (got {horizon_weight})")
+        if not horizon_weight_np.sum() > 0:
+            raise ValueError(f"At least some values in horizon_weight must be > 0 (got {horizon_weight})")
+        if not np.isfinite(horizon_weight_np).all():
+            raise ValueError(f"All horizon_weight values must be finite (got {horizon_weight})")
+        horizon_weight_np = horizon_weight_np * prediction_length / horizon_weight_np.sum()
+        return horizon_weight_np.reshape([1, prediction_length])

autogluon/timeseries/metrics/point.py

@@ -1,6 +1,6 @@
 import logging
 import warnings
-from typing import Optional
+from typing import Optional, Sequence
 
 import numpy as np
 import pandas as pd
@@ -38,10 +38,18 @@ class RMSE(TimeSeriesScorer):
     equivalent_tabular_regression_metric = "root_mean_squared_error"
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
-        return np.sqrt(self._safemean((y_true - y_pred) ** 2))
+        y_true, y_pred = y_true.to_numpy(), y_pred.to_numpy()
+        errors = ((y_true - y_pred) ** 2).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return np.sqrt(self._safemean(errors))
 
 
 class MSE(TimeSeriesScorer):
@@ -69,10 +77,18 @@ class MSE(TimeSeriesScorer):
     equivalent_tabular_regression_metric = "mean_squared_error"
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
-        return self._safemean((y_true - y_pred) ** 2)
+        y_true, y_pred = y_true.to_numpy(), y_pred.to_numpy()
+        errors = ((y_true - y_pred) ** 2).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return self._safemean(errors)
 
 
 class MAE(TimeSeriesScorer):
@@ -98,10 +114,18 @@ class MAE(TimeSeriesScorer):
     equivalent_tabular_regression_metric = "mean_absolute_error"
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
-        return self._safemean((y_true - y_pred).abs())
+        y_true, y_pred = y_true.to_numpy(), y_pred.to_numpy()
+        errors = np.abs(y_true - y_pred).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return self._safemean(errors)
 
 
 class WAPE(TimeSeriesScorer):
@@ -119,6 +143,7 @@ class WAPE(TimeSeriesScorer):
     - not sensitive to outliers
     - prefers models that accurately estimate the median
 
+    If `self.horizon_weight` is provided, both the errors and the target time series in the denominator will be re-weighted.
 
     References
     ----------
@@ -129,10 +154,19 @@ class WAPE(TimeSeriesScorer):
     equivalent_tabular_regression_metric = "mean_absolute_error"
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
-        return (y_true - y_pred).abs().sum() / y_true.abs().sum()
+        y_true, y_pred = y_true.to_numpy(), y_pred.to_numpy()
+        errors = np.abs(y_true - y_pred).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+            y_true = y_true.reshape([-1, self.prediction_length]) * self.horizon_weight
+        return np.nansum(errors) / np.nansum(np.abs(y_true))
 
 
 class SMAPE(TimeSeriesScorer):
@@ -158,10 +192,18 @@ class SMAPE(TimeSeriesScorer):
     equivalent_tabular_regression_metric = "symmetric_mean_absolute_percentage_error"
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
-        return self._safemean(2 * ((y_true - y_pred).abs() / (y_true.abs() + y_pred.abs())))
+        y_true, y_pred = y_true.to_numpy(), y_pred.to_numpy()
+        errors = (np.abs(y_true - y_pred) / (np.abs(y_true) + np.abs(y_pred))).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return 2 * self._safemean(errors)
 
 
 class MAPE(TimeSeriesScorer):
@@ -187,10 +229,18 @@ class MAPE(TimeSeriesScorer):
     equivalent_tabular_regression_metric = "mean_absolute_percentage_error"
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
-        return self._safemean((y_true - y_pred).abs() / y_true.abs())
+        y_true, y_pred = y_true.to_numpy(), y_pred.to_numpy()
+        errors = (np.abs(y_true - y_pred) / np.abs(y_true)).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return self._safemean(errors)
 
 
 class MASE(TimeSeriesScorer):
@@ -226,7 +276,15 @@ class MASE(TimeSeriesScorer):
     optimized_by_median = True
     equivalent_tabular_regression_metric = "mean_absolute_error"
 
-    def __init__(self):
+    def __init__(
+        self,
+        prediction_length: int = 1,
+        seasonal_period: Optional[int] = None,
+        horizon_weight: Optional[Sequence[float]] = None,
+    ):
+        super().__init__(
+            prediction_length=prediction_length, seasonal_period=seasonal_period, horizon_weight=horizon_weight
+        )
         self._past_abs_seasonal_error: Optional[pd.Series] = None
 
     def save_past_metrics(
@@ -240,16 +298,22 @@ class MASE(TimeSeriesScorer):
         self._past_abs_seasonal_error = None
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
-        y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
         if self._past_abs_seasonal_error is None:
             raise AssertionError("Call `save_past_metrics` before `compute_metric`")
 
-        num_items = len(self._past_abs_seasonal_error)
-        # Reshape abs errors into [num_items, prediction_length] to normalize per item without groupby
-        abs_errors = np.abs(y_true.to_numpy() - y_pred.to_numpy()).reshape([num_items, -1])
-        return self._safemean(abs_errors / self._past_abs_seasonal_error.values[:, None])
+        y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
+        y_true, y_pred = y_true.to_numpy(), y_pred.to_numpy()
+
+        errors = np.abs(y_true - y_pred).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return self._safemean(errors / self._past_abs_seasonal_error.to_numpy()[:, None])
 
 
 class RMSSE(TimeSeriesScorer):
@@ -286,7 +350,15 @@ class RMSSE(TimeSeriesScorer):
 
     equivalent_tabular_regression_metric = "root_mean_squared_error"
 
-    def __init__(self):
+    def __init__(
+        self,
+        prediction_length: int = 1,
+        seasonal_period: Optional[int] = None,
+        horizon_weight: Optional[Sequence[float]] = None,
+    ):
+        super().__init__(
+            prediction_length=prediction_length, seasonal_period=seasonal_period, horizon_weight=horizon_weight
+        )
         self._past_squared_seasonal_error: Optional[pd.Series] = None
 
     def save_past_metrics(
@@ -300,16 +372,21 @@ class RMSSE(TimeSeriesScorer):
         self._past_squared_seasonal_error = None
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
-        y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
         if self._past_squared_seasonal_error is None:
             raise AssertionError("Call `save_past_metrics` before `compute_metric`")
 
-        num_items = len(self._past_squared_seasonal_error)
-        # Reshape squared errors into [num_items, prediction_length] to normalize per item without groupby
-        squared_errors = ((y_true.to_numpy() - y_pred.to_numpy()) ** 2.0).reshape([num_items, -1])
-        return np.sqrt(self._safemean(squared_errors / self._past_squared_seasonal_error.values[:, None]))
+        y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
+        y_true, y_pred = y_true.to_numpy(), y_pred.to_numpy()
+        errors = ((y_true - y_pred) ** 2).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return np.sqrt(self._safemean(errors / self._past_squared_seasonal_error.to_numpy()[:, None]))
 
 
 class RMSLE(TimeSeriesScorer):
@@ -336,20 +413,26 @@ class RMSLE(TimeSeriesScorer):
     """
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
+        y_true, y_pred = y_true.to_numpy(), y_pred.to_numpy()
         y_pred = np.clip(y_pred, a_min=0.0, a_max=None)
 
-        return np.sqrt(np.power(np.log1p(y_pred) - np.log1p(y_true), 2).mean())
+        errors = np.power(np.log1p(y_pred) - np.log1p(y_true), 2).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return np.sqrt(self._safemean(errors))
 
     def __call__(
         self,
         data: TimeSeriesDataFrame,
         predictions: TimeSeriesDataFrame,
-        prediction_length: int = 1,
         target: str = "target",
-        seasonal_period: Optional[int] = None,
         **kwargs,
     ) -> float:
         if (data[target] < 0).any():
@@ -357,9 +440,7 @@ class RMSLE(TimeSeriesScorer):
         return super().__call__(
             data=data,
             predictions=predictions,
-            prediction_length=prediction_length,
             target=target,
-            seasonal_period=seasonal_period,
             **kwargs,
         )
 
@@ -382,35 +463,43 @@ class WCD(TimeSeriesScorer):
     Parameters
     ----------
     alpha : float, default = 0.5
-        Values > 0.5 correspond put a stronger penalty on underpredictions (when cumulative forecast is below the
+        Values > 0.5 put a stronger penalty on underpredictions (when cumulative forecast is below the
         cumulative actual value). Values < 0.5 put a stronger penalty on overpredictions.
     """
 
-    def __init__(self, alpha: float = 0.5):
+    def __init__(
+        self,
+        alpha: float = 0.5,
+        prediction_length: int = 1,
+        seasonal_period: Optional[int] = None,
+        horizon_weight: Optional[Sequence[float]] = None,
+    ):
+        super().__init__(
+            prediction_length=prediction_length, seasonal_period=seasonal_period, horizon_weight=horizon_weight
+        )
         assert 0 < alpha < 1, "alpha must be in (0, 1)"
         self.alpha = alpha
-        self.num_items: Optional[int] = None
         warnings.warn(
             f"{self.name} is an experimental metric. Its behavior may change in the future version of AutoGluon."
         )
 
-    def save_past_metrics(
-        self, data_past: TimeSeriesDataFrame, target: str = "target", seasonal_period: int = 1, **kwargs
-    ) -> None:
-        self.num_items = data_past.num_items
-
     def _fast_cumsum(self, y: np.ndarray) -> np.ndarray:
-        """Compute the cumulative sum for each consecutive `prediction_length` items in the array."""
-        assert self.num_items is not None, "Make sure to call `save_past_metrics` before `compute_metric`"
-        y = y.reshape(self.num_items, -1)
+        """Compute the cumulative sum for each consecutive `self.prediction_length` items in the array."""
+        y = y.reshape(-1, self.prediction_length)
         return np.nancumsum(y, axis=1).ravel()
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         y_true, y_pred = self._get_point_forecast_score_inputs(data_future, predictions, target=target)
         cumsum_true = self._fast_cumsum(y_true.to_numpy())
         cumsum_pred = self._fast_cumsum(y_pred.to_numpy())
         diffs = cumsum_pred - cumsum_true
-        error = diffs * np.where(diffs < 0, -self.alpha, (1 - self.alpha))
-        return 2 * self._safemean(error)
+        errors = (diffs * np.where(diffs < 0, -self.alpha, (1 - self.alpha))).reshape([-1, self.prediction_length])
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return 2 * self._safemean(errors)

autogluon/timeseries/metrics/quantile.py

@@ -1,4 +1,4 @@
-from typing import Optional
+from typing import Optional, Sequence
 
 import numpy as np
 import pandas as pd
@@ -25,6 +25,7 @@ class WQL(TimeSeriesScorer):
     - scale-dependent (time series with large absolute value contribute more to the loss)
     - equivalent to WAPE if ``quantile_levels = [0.5]``
 
+    If `horizon_weight` is provided, both the errors and the target time series in the denominator will be re-weighted.
 
     References
     ----------
@@ -34,16 +35,25 @@ class WQL(TimeSeriesScorer):
     needs_quantile = True
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         y_true, q_pred, quantile_levels = self._get_quantile_forecast_score_inputs(data_future, predictions, target)
-        values_true = y_true.values[:, None]  # shape [N, 1]
-        values_pred = q_pred.values  # shape [N, len(quantile_levels)]
+        y_true = y_true.to_numpy()[:, None]  # shape [N, 1]
+        q_pred = q_pred.to_numpy()  # shape [N, len(quantile_levels)]
 
-        return 2 * np.mean(
-            np.nansum(np.abs((values_true - values_pred) * ((values_true <= values_pred) - quantile_levels)), axis=0)
-            / np.nansum(np.abs(values_true))
+        errors = (
+            np.abs((q_pred - y_true) * ((y_true <= q_pred) - quantile_levels))
+            .mean(axis=1)
+            .reshape([-1, self.prediction_length])
         )
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+            y_true = y_true.reshape([-1, self.prediction_length]) * self.horizon_weight
+        return 2 * np.nansum(errors) / np.nansum(np.abs(y_true))
 
 
 class SQL(TimeSeriesScorer):
@@ -79,7 +89,15 @@ class SQL(TimeSeriesScorer):
 
     needs_quantile = True
 
-    def __init__(self):
+    def __init__(
+        self,
+        prediction_length: int = 1,
+        seasonal_period: Optional[int] = None,
+        horizon_weight: Optional[Sequence[float]] = None,
+    ):
+        super().__init__(
+            prediction_length=prediction_length, seasonal_period=seasonal_period, horizon_weight=horizon_weight
+        )
         self._past_abs_seasonal_error: Optional[pd.Series] = None
 
     def save_past_metrics(
@@ -93,17 +111,24 @@ class SQL(TimeSeriesScorer):
         self._past_abs_seasonal_error = None
 
     def compute_metric(
-        self, data_future: TimeSeriesDataFrame, predictions: TimeSeriesDataFrame, target: str = "target", **kwargs
+        self,
+        data_future: TimeSeriesDataFrame,
+        predictions: TimeSeriesDataFrame,
+        target: str = "target",
+        **kwargs,
     ) -> float:
         if self._past_abs_seasonal_error is None:
             raise AssertionError("Call `save_past_metrics` before `compute_metric`")
 
         y_true, q_pred, quantile_levels = self._get_quantile_forecast_score_inputs(data_future, predictions, target)
-        q_pred = q_pred.values
-        values_true = y_true.values[:, None]  # shape [N, 1]
-
-        ql = np.abs((q_pred - values_true) * ((values_true <= q_pred) - quantile_levels)).mean(axis=1)
-        num_items = len(self._past_abs_seasonal_error)
-        # Reshape quantile losses values into [num_items, prediction_length] to normalize per item without groupby
-        quantile_losses = ql.reshape([num_items, -1])
-        return 2 * self._safemean(quantile_losses / self._past_abs_seasonal_error.values[:, None])
+        q_pred = q_pred.to_numpy()
+        y_true = y_true.to_numpy()[:, None]  # shape [N, 1]
+
+        errors = (
+            np.abs((q_pred - y_true) * ((y_true <= q_pred) - quantile_levels))
+            .mean(axis=1)
+            .reshape([-1, self.prediction_length])
+        )
+        if self.horizon_weight is not None:
+            errors *= self.horizon_weight
+        return 2 * self._safemean(errors / self._past_abs_seasonal_error.to_numpy()[:, None])

autogluon/timeseries/models/abstract/abstract_timeseries_model.py

@@ -57,9 +57,6 @@ class TimeSeriesModelBase(ModelBase, ABC):
         Metric by which predictions will be ultimately evaluated on future test data. This only impacts
         ``model.score()``, as eval_metric is not used during training. Available metrics can be found in
         ``autogluon.timeseries.metrics``.
-    eval_metric_seasonal_period : int, optional
-        Seasonal period used to compute some evaluation metrics such as mean absolute scaled error (MASE). Defaults to
-        ``None``, in which case the seasonal period is computed based on the data frequency.
     hyperparameters : dict, default = None
         Hyperparameters that will be used by the model (can be search spaces instead of fixed values).
         If None, model defaults are used. This is identical to passing an empty dictionary.
@@ -88,7 +85,6 @@ class TimeSeriesModelBase(ModelBase, ABC):
         target: str = "target",
         quantile_levels: Sequence[float] = (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9),
         eval_metric: Union[str, TimeSeriesScorer, None] = None,
-        eval_metric_seasonal_period: Optional[int] = None,
     ):
         self.name = name or re.sub(r"Model$", "", self.__class__.__name__)
 
@@ -103,8 +99,7 @@ class TimeSeriesModelBase(ModelBase, ABC):
 
         self.path = os.path.join(self.path_root, self.name)
 
-        self.eval_metric: TimeSeriesScorer = check_get_evaluation_metric(eval_metric)
-        self.eval_metric_seasonal_period = eval_metric_seasonal_period
+        self.eval_metric = check_get_evaluation_metric(eval_metric, prediction_length=prediction_length)
         self.target: str = target
         self.covariate_metadata = covariate_metadata or CovariateMetadata()
 
@@ -393,7 +388,6 @@ class AbstractTimeSeriesModel(TimeSeriesModelBase, TimeSeriesTunable, ABC):
         target: str = "target",
         quantile_levels: Sequence[float] = (0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9),
         eval_metric: Union[str, TimeSeriesScorer, None] = None,
-        eval_metric_seasonal_period: Optional[int] = None,
     ):
         # TODO: make freq a required argument in AbstractTimeSeriesModel
         super().__init__(
@@ -406,7 +400,6 @@ class AbstractTimeSeriesModel(TimeSeriesModelBase, TimeSeriesTunable, ABC):
             target=target,
             quantile_levels=quantile_levels,
             eval_metric=eval_metric,
-            eval_metric_seasonal_period=eval_metric_seasonal_period,
         )
         self.target_scaler: Optional[TargetScaler]
         self.covariate_scaler: Optional[CovariateScaler]
@@ -700,19 +693,15 @@ class AbstractTimeSeriesModel(TimeSeriesModelBase, TimeSeriesTunable, ABC):
         self,
         data: TimeSeriesDataFrame,
         predictions: TimeSeriesDataFrame,
-        metric: Optional[str] = None,
     ) -> float:
         """Compute the score measuring how well the predictions align with the data."""
-        eval_metric = self.eval_metric if metric is None else check_get_evaluation_metric(metric)
-        return eval_metric.score(
+        return self.eval_metric.score(
             data=data,
             predictions=predictions,
-            prediction_length=self.prediction_length,
             target=self.target,
-            seasonal_period=self.eval_metric_seasonal_period,
         )
 
-    def score(self, data: TimeSeriesDataFrame, metric: Optional[str] = None) -> float:
+    def score(self, data: TimeSeriesDataFrame) -> float:
         """Return the evaluation scores for given metric and dataset. The last
         `self.prediction_length` time steps of each time series in the input data set
         will be held out and used for computing the evaluation score. Time series
@@ -722,9 +711,6 @@ class AbstractTimeSeriesModel(TimeSeriesModelBase, TimeSeriesTunable, ABC):
         ----------
         data: TimeSeriesDataFrame
             Dataset used for scoring.
-        metric: str
-            String identifier of evaluation metric to use, from one of
-            `autogluon.timeseries.utils.metric_utils.AVAILABLE_METRICS`.
 
         Returns
         -------
@@ -736,7 +722,7 @@ class AbstractTimeSeriesModel(TimeSeriesModelBase, TimeSeriesTunable, ABC):
             prediction_length=self.prediction_length, known_covariates_names=self.covariate_metadata.known_covariates
         )
         predictions = self.predict(past_data, known_covariates=known_covariates)
-        return self._score_with_predictions(data=data, predictions=predictions, metric=metric)
+        return self._score_with_predictions(data=data, predictions=predictions)
 
     def score_and_cache_oof(
         self,

autogluon/timeseries/models/ensemble/greedy.py

@@ -28,7 +28,6 @@ class TimeSeriesEnsembleSelection(EnsembleSelection):
         random_state: Optional[np.random.RandomState] = None,
         prediction_length: int = 1,
         target: str = "target",
-        eval_metric_seasonal_period: int = 1,
         **kwargs,
     ):
         super().__init__(
@@ -43,7 +42,6 @@ class TimeSeriesEnsembleSelection(EnsembleSelection):
         )
         self.prediction_length = prediction_length
         self.target = target
-        self.eval_metric_seasonal_period = eval_metric_seasonal_period
         self.metric: TimeSeriesScorer
 
         self.dummy_pred_per_window = []
@@ -79,6 +77,10 @@ class TimeSeriesEnsembleSelection(EnsembleSelection):
         self.scorer_per_window = []
         self.data_future_per_window = []
 
+        seasonal_period = self.metric.seasonal_period
+        if seasonal_period is None:
+            seasonal_period = get_seasonality(labels[0].freq)
+
         for window_idx, data in enumerate(labels):
             dummy_pred = copy.deepcopy(predictions[0][window_idx])
             # This should never happen; sanity check to make sure that all predictions have the same index
@@ -90,7 +92,7 @@ class TimeSeriesEnsembleSelection(EnsembleSelection):
             # Split the observed time series once to avoid repeated computations inside the evaluator
             data_past = data.slice_by_timestep(None, -self.prediction_length)
             data_future = data.slice_by_timestep(-self.prediction_length, None)
-            scorer.save_past_metrics(data_past, target=self.target, seasonal_period=self.eval_metric_seasonal_period)
+            scorer.save_past_metrics(data_past, target=self.target, seasonal_period=seasonal_period)
             self.scorer_per_window.append(scorer)
             self.data_future_per_window.append(data_future)
 
@@ -122,7 +124,9 @@ class TimeSeriesEnsembleSelection(EnsembleSelection):
             dummy_pred[list(dummy_pred.columns)] = y_pred_proba[window_idx]
             # We use scorer.compute_metric instead of scorer.score to avoid repeated calls to scorer.save_past_metrics
             metric_value = self.scorer_per_window[window_idx].compute_metric(
-                data_future, dummy_pred, target=self.target
+                data_future,
+                dummy_pred,
+                target=self.target,
             )
             total_score += metric.sign * metric_value
         avg_score = total_score / len(self.data_future_per_window)
@@ -162,14 +166,11 @@ class GreedyEnsemble(AbstractWeightedTimeSeriesEnsembleModel):
         model_scores: Optional[Dict[str, float]] = None,
         time_limit: Optional[float] = None,
     ):
-        if self.eval_metric_seasonal_period is None:
-            self.eval_metric_seasonal_period = get_seasonality(self.freq)
         ensemble_selection = TimeSeriesEnsembleSelection(
             ensemble_size=self.get_hyperparameters()["ensemble_size"],
             metric=self.eval_metric,
             prediction_length=self.prediction_length,
             target=self.target,
-            eval_metric_seasonal_period=self.eval_metric_seasonal_period,
         )
         ensemble_selection.fit(
             predictions=list(predictions_per_window.values()),

autogluon/timeseries/models/presets.py

@@ -183,7 +183,6 @@ def get_preset_models(
     prediction_length: int,
     path: str,
     eval_metric: Union[str, TimeSeriesScorer],
-    eval_metric_seasonal_period: Optional[int],
     hyperparameters: Union[str, Dict, None],
     hyperparameter_tune: bool,
     covariate_metadata: CovariateMetadata,
@@ -260,7 +259,6 @@ def get_preset_models(
                 freq=freq,
                 prediction_length=prediction_length,
                 eval_metric=eval_metric,
-                eval_metric_seasonal_period=eval_metric_seasonal_period,
                 covariate_metadata=covariate_metadata,
                 hyperparameters=model_hps,
                 **kwargs,

autogluon/timeseries/predictor.py

@@ -93,6 +93,14 @@ class TimeSeriesPredictor:
     eval_metric_seasonal_period : int, optional
         Seasonal period used to compute some evaluation metrics such as mean absolute scaled error (MASE). Defaults to
         ``None``, in which case the seasonal period is computed based on the data frequency.
+    horizon_weight : List[float], optional
+        Weight assigned to each time step in the forecast horizon when computing the `eval_metric`. If provided, this
+        must be a list with `prediction_length` non-negative values, where at least some values are greater than zero.
+        AutoGluon will automatically normalize the weights so that they sum up to `prediction_length`. By default, all
+        time steps in the forecast horizon have the same weight, which is equivalent to setting `horizon_weight = [1] * prediction_length`.
+
+        This parameter only affects model selection and ensemble construction; it has no effect on the loss function of
+        the individual forecasting models.
     known_covariates_names: List[str], optional
         Names of the covariates that are known in advance for all time steps in the forecast horizon. These are also
         known as dynamic features, exogenous variables, additional regressors or related time series. Examples of such
@@ -107,7 +115,7 @@ class TimeSeriesPredictor:
         List of increasing decimals that specifies which quantiles should be estimated when making distributional
         forecasts. Defaults to ``[0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]``.
     path : str or pathlib.Path, optional
-        Path to the directory where models and intermediate outputs will be saved. Defaults to a timestamped folder
+        Path to the local directory where models and intermediate outputs will be saved. Defaults to a timestamped folder
         ``AutogluonModels/ag-[TIMESTAMP]`` that will be created in the working directory.
     verbosity : int, default = 2
         Verbosity levels range from 0 to 4 and control how much information is printed to stdout. Higher levels
@@ -144,6 +152,7 @@ class TimeSeriesPredictor:
         freq: Optional[str] = None,
         eval_metric: Union[str, TimeSeriesScorer, None] = None,
         eval_metric_seasonal_period: Optional[int] = None,
+        horizon_weight: Optional[List[float]] = None,
         path: Optional[Union[str, Path]] = None,
         verbosity: int = 2,
         log_to_file: bool = True,
@@ -156,6 +165,11 @@ class TimeSeriesPredictor:
         self.verbosity = verbosity
         set_logger_verbosity(self.verbosity, logger=logger)
         self.path = setup_outputdir(path)
+        if self.path.lower().startswith("s3://"):
+            logger.warning(
+                "Warning: S3 paths are not supported for the `path` argument in TimeSeriesPredictor. "
+                "Use a local path and upload the trained predictor to S3 manually if needed"
+            )
         self._setup_log_to_file(log_to_file=log_to_file, log_file_path=log_file_path)
 
         self.cache_predictions = cache_predictions
@@ -187,15 +201,18 @@ class TimeSeriesPredictor:
             if std_freq != str(self.freq):
                 logger.info(f"Frequency '{self.freq}' stored as '{std_freq}'")
             self.freq = std_freq
-        self.eval_metric = check_get_evaluation_metric(eval_metric)
-        self.eval_metric_seasonal_period = eval_metric_seasonal_period
+        self.eval_metric: TimeSeriesScorer = check_get_evaluation_metric(
+            eval_metric,
+            prediction_length=prediction_length,
+            seasonal_period=eval_metric_seasonal_period,
+            horizon_weight=horizon_weight,
+        )
         if quantile_levels is None:
             quantile_levels = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]
         self.quantile_levels = sorted(quantile_levels)
         self._learner: TimeSeriesLearner = self._learner_type(
             path_context=self.path,
-            eval_metric=eval_metric,
-            eval_metric_seasonal_period=eval_metric_seasonal_period,
+            eval_metric=self.eval_metric,
             target=self.target,
             known_covariates_names=self.known_covariates_names,
             prediction_length=self.prediction_length,
@@ -494,33 +511,22 @@ class TimeSeriesPredictor:
 
             Available presets:
 
-            - ``"fast_training"``: fit simple statistical models (``ETS``, ``Theta``, ``Naive``, ``SeasonalNaive``) + fast tree-based models ``RecursiveTabular``
-              and ``DirectTabular``. These models are fast to train but may not be very accurate.
-            - ``"medium_quality"``: all models mentioned above + deep learning model ``TemporalFusionTransformer`` + Chronos-Bolt (small). Produces good forecasts with reasonable training time.
-            - ``"high_quality"``: All ML models available in AutoGluon + additional statistical models (``NPTS``, ``AutoETS``,
-              ``DynamicOptimizedTheta``). Much more accurate than ``medium_quality``, but takes longer to train.
+            - ``"fast_training"``: Simple statistical and tree-based ML models. These models are fast to train but may not be very accurate.
+            - ``"medium_quality"``: Same models as above, plus deep learning models ``TemporalFusionTransformer`` and Chronos-Bolt (small). Produces good forecasts with reasonable training time.
+            - ``"high_quality"``: A mix of multiple DL, ML and statistical forecasting models available in AutoGluon that offers the best forecast accuracy. Much more accurate than ``medium_quality``, but takes longer to train.
             - ``"best_quality"``: Same models as in ``"high_quality"``, but performs validation with multiple backtests. Usually better than ``high_quality``, but takes even longer to train.
 
-            Available presets with the new, faster `Chronos-Bolt <https://github.com/amazon-science/chronos-forecasting>`_ model:
+            Available presets with the `Chronos-Bolt <https://github.com/amazon-science/chronos-forecasting>`_ model:
 
             - ``"bolt_{model_size}"``: where model size is one of ``tiny,mini,small,base``. Uses the Chronos-Bolt pretrained model for zero-shot forecasting.
               See the documentation for ``ChronosModel`` or see `Hugging Face <https://huggingface.co/collections/amazon/chronos-models-65f1791d630a8d57cb718444>`_ for more information.
 
-            Available presets with the original `Chronos <https://github.com/amazon-science/chronos-forecasting>`_ model.
-            Note that as of v1.2 we recommend using the new, faster Chronos-Bolt models instead of the original Chronos models.
+            Exact definitions of these presets can be found in the source code
+            [`1 <https://github.com/autogluon/autogluon/blob/stable/timeseries/src/autogluon/timeseries/configs/presets_configs.py>`_,
+            `2 <https://github.com/autogluon/autogluon/blob/stable/timeseries/src/autogluon/timeseries/models/presets.py>`_].
 
-            - ``"chronos_{model_size}"``: where model size is one of ``tiny,mini,small,base,large``. Uses the Chronos pretrained model for zero-shot forecasting.
-              See the documentation for ``ChronosModel`` or see `Hugging Face <https://huggingface.co/collections/amazon/chronos-models-65f1791d630a8d57cb718444>`_ for more information.
-              Note that a GPU is required for model sizes ``small``, ``base`` and ``large``.
-            - ``"chronos"``: alias for ``"chronos_small"``.
-            - ``"chronos_ensemble"``: builds an ensemble of seasonal naive, tree-based and deep learning models with fast inference
-              and ``"chronos_small"``.
-            - ``"chronos_large_ensemble"``: builds an ensemble of seasonal naive, tree-based and deep learning models
-              with fast inference and ``"chronos_large"``.
-
-            Details for these presets can be found in ``autogluon/timeseries/configs/presets_configs.py``. If not
-            provided, user-provided values for ``hyperparameters`` and ``hyperparameter_tune_kwargs`` will be used
-            (defaulting to their default values specified below).
+            If no `presets` are selected, user-provided values for `hyperparameters` will be used (defaulting to their
+            default values specified below).
         hyperparameters : str or dict, optional
             Determines what models are trained and what hyperparameters are used by each model.
 
@@ -684,7 +690,8 @@ class TimeSeriesPredictor:
             target=self.target,
             known_covariates_names=self.known_covariates_names,
             eval_metric=self.eval_metric,
-            eval_metric_seasonal_period=self.eval_metric_seasonal_period,
+            eval_metric_seasonal_period=self.eval_metric.seasonal_period,
+            horizon_weight=self.eval_metric.horizon_weight,
             quantile_levels=self.quantile_levels,
             freq=self.freq,
             time_limit=time_limit,
@@ -1500,7 +1507,8 @@ class TimeSeriesPredictor:
             target=self.target,
             prediction_length=self.prediction_length,
             eval_metric=self.eval_metric.name,
-            eval_metric_seasonal_period=self.eval_metric_seasonal_period,
+            eval_metric_seasonal_period=self.eval_metric.seasonal_period,
+            horizon_weight=self.eval_metric.horizon_weight,
             quantile_levels=self.quantile_levels,
         )
         return simulation_dict

autogluon/timeseries/trainer.py

@@ -46,7 +46,6 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
         path: str,
         prediction_length: int = 1,
         eval_metric: Union[str, TimeSeriesScorer, None] = None,
-        eval_metric_seasonal_period: Optional[int] = None,
         save_data: bool = True,
         skip_model_selection: bool = False,
         enable_ensemble: bool = True,
@@ -86,8 +85,7 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
         #: self.refit_single_full() and self.refit_full().
         self.model_refit_map = {}
 
-        self.eval_metric: TimeSeriesScorer = check_get_evaluation_metric(eval_metric)
-        self.eval_metric_seasonal_period = eval_metric_seasonal_period
+        self.eval_metric = check_get_evaluation_metric(eval_metric, prediction_length=prediction_length)
         if val_splitter is None:
             val_splitter = ExpandingWindowSplitter(prediction_length=self.prediction_length)
         assert isinstance(val_splitter, AbstractWindowSplitter), "val_splitter must be of type AbstractWindowSplitter"
@@ -577,7 +575,6 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
         ensemble = self.ensemble_model_type(
             name=self._get_ensemble_model_name(),
             eval_metric=self.eval_metric,
-            eval_metric_seasonal_period=self.eval_metric_seasonal_period,
             target=self.target,
             prediction_length=self.prediction_length,
             path=self.path,
@@ -791,6 +788,17 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
             raise ValueError(f"Model {model_name} failed to predict. Please check the model's logs.")
         return predictions
 
+    def _get_eval_metric(self, metric: Union[str, TimeSeriesScorer, None]) -> TimeSeriesScorer:
+        if metric is None:
+            return self.eval_metric
+        else:
+            return check_get_evaluation_metric(
+                metric,
+                prediction_length=self.prediction_length,
+                seasonal_period=self.eval_metric.seasonal_period,
+                horizon_weight=self.eval_metric.horizon_weight,
+            )
+
     def _score_with_predictions(
         self,
         data: TimeSeriesDataFrame,
@@ -798,13 +806,11 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
         metric: Union[str, TimeSeriesScorer, None] = None,
     ) -> float:
         """Compute the score measuring how well the predictions align with the data."""
-        eval_metric = self.eval_metric if metric is None else check_get_evaluation_metric(metric)
-        return eval_metric.score(
+        return self._get_eval_metric(metric).score(
             data=data,
             predictions=predictions,
             prediction_length=self.prediction_length,
             target=self.target,
-            seasonal_period=self.eval_metric_seasonal_period,
         )
 
     def score(
@@ -814,7 +820,7 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
         metric: Union[str, TimeSeriesScorer, None] = None,
         use_cache: bool = True,
     ) -> float:
-        eval_metric = self.eval_metric if metric is None else check_get_evaluation_metric(metric)
+        eval_metric = self._get_eval_metric(metric)
         scores_dict = self.evaluate(data=data, model=model, metrics=[eval_metric], use_cache=use_cache)
         return scores_dict[eval_metric.name]
 
@@ -833,7 +839,7 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
         metrics_ = [metrics] if not isinstance(metrics, list) else metrics
         scores_dict = {}
         for metric in metrics_:
-            eval_metric = self.eval_metric if metric is None else check_get_evaluation_metric(metric)
+            eval_metric = self._get_eval_metric(metric)
             scores_dict[eval_metric.name] = self._score_with_predictions(
                 data=data, predictions=predictions, metric=eval_metric
             )
@@ -855,7 +861,7 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
         confidence_level: float = 0.99,
     ) -> pd.DataFrame:
         assert method in ["naive", "permutation"], f"Invalid feature importance method {method}."
-        metric = check_get_evaluation_metric(metric) if metric is not None else self.eval_metric
+        eval_metric = self._get_eval_metric(metric)
 
         logger.info("Computing feature importance")
 
@@ -906,7 +912,9 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
             else:
                 data_sample = data
 
-            base_score = self.evaluate(data=data_sample, model=model, metrics=metric, use_cache=False)[metric.name]
+            base_score = self.evaluate(data=data_sample, model=model, metrics=eval_metric, use_cache=False)[
+                eval_metric.name
+            ]
 
             for feature in features:
                 # override importance for unused features
@@ -914,9 +922,9 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
                     continue
                 else:
                     data_sample_replaced = importance_transform.transform(data_sample, feature_name=feature)
-                    score = self.evaluate(data=data_sample_replaced, model=model, metrics=metric, use_cache=False)[
-                        metric.name
-                    ]
+                    score = self.evaluate(
+                        data=data_sample_replaced, model=model, metrics=eval_metric, use_cache=False
+                    )[eval_metric.name]
 
                     importance = base_score - score
                     if relative_scores:
@@ -1266,7 +1274,6 @@ class TimeSeriesTrainer(AbstractTrainer[TimeSeriesModelBase]):
         return get_preset_models(
             path=self.path,
             eval_metric=self.eval_metric,
-            eval_metric_seasonal_period=self.eval_metric_seasonal_period,
             prediction_length=self.prediction_length,
             freq=freq,
             hyperparameters=hyperparameters,

autogluon/timeseries/version.py

@@ -1,4 +1,4 @@
 """This is the autogluon version file."""
 
-__version__ = "1.2.1b20250425"
+__version__ = "1.2.1b20250426"
 __lite__ = False

{autogluon.timeseries-1.2.1b20250425.dist-info → autogluon.timeseries-1.2.1b20250426.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: autogluon.timeseries
-Version: 1.2.1b20250425
+Version: 1.2.1b20250426
 Summary: Fast and Accurate ML in 3 Lines of Code
 Home-page: https://github.com/autogluon/autogluon
 Author: AutoGluon Community
@@ -55,10 +55,10 @@ Requires-Dist: fugue>=0.9.0
 Requires-Dist: tqdm<5,>=4.38
 Requires-Dist: orjson~=3.9
 Requires-Dist: tensorboard<3,>=2.9
-Requires-Dist: autogluon.core[raytune]==1.2.1b20250425
-Requires-Dist: autogluon.common==1.2.1b20250425
-Requires-Dist: autogluon.features==1.2.1b20250425
-Requires-Dist: autogluon.tabular[catboost,lightgbm,xgboost]==1.2.1b20250425
+Requires-Dist: autogluon.core[raytune]==1.2.1b20250426
+Requires-Dist: autogluon.common==1.2.1b20250426
+Requires-Dist: autogluon.features==1.2.1b20250426
+Requires-Dist: autogluon.tabular[catboost,lightgbm,xgboost]==1.2.1b20250426
 Provides-Extra: all
 Provides-Extra: chronos-onnx
 Requires-Dist: optimum[onnxruntime]<1.23,>=1.17; extra == "chronos-onnx"

{autogluon.timeseries-1.2.1b20250425.dist-info → autogluon.timeseries-1.2.1b20250426.dist-info}/RECORD

@@ -1,25 +1,25 @@
-autogluon.timeseries-1.2.1b20250425-py3.9-nspkg.pth,sha256=cQGwpuGPqg1GXscIwt-7PmME1OnSpD-7ixkikJ31WAY,554
+autogluon.timeseries-1.2.1b20250426-py3.9-nspkg.pth,sha256=cQGwpuGPqg1GXscIwt-7PmME1OnSpD-7ixkikJ31WAY,554
 autogluon/timeseries/__init__.py,sha256=_CrLLc1fkjen7UzWoO0Os8WZoHOgvZbHKy46I8v_4k4,304
 autogluon/timeseries/evaluator.py,sha256=l642tYfTHsl8WVIq_vV6qhgAFVFr9UuZD7gLra3A_Kc,250
-autogluon/timeseries/learner.py,sha256=7dqSHKCIX2osjv9cmWWLwaGvdrPvla0HTnsR75bdenY,14112
-autogluon/timeseries/predictor.py,sha256=Ur5TV0idIssnUzVdZcVsnrNWAmX6JGxbLZZnpT8EGmQ,87980
+autogluon/timeseries/learner.py,sha256=pIn4YSOk0aqCWyBpIlwnAsFnG4h7PLXk8guFH3wFS-w,13923
+autogluon/timeseries/predictor.py,sha256=Dz-LJVU5sjlFCOqHTeYPt77DuGavdAXB0DkclpM55rY,88173
 autogluon/timeseries/regressor.py,sha256=xw5VPrXS-NQ_Ts4ppDjoNV0TdqUYjW4VINUtb_BZdiI,11868
 autogluon/timeseries/splitter.py,sha256=yzPca9p2bWV-_VJAptUyyzQsxu-uixAdpMoGQtDzMD4,3205
-autogluon/timeseries/trainer.py,sha256=zrO3ARI_h14pYy7_GEGO7nqd9rONiDmx5hhCRMyzwls,58115
-autogluon/timeseries/version.py,sha256=zTyiK3ClhRyp8BZjIMLB2n9YsA7GaR9qnPDLREdyRQs,91
+autogluon/timeseries/trainer.py,sha256=57OyqlTAVahDPxF5GmdDljIr1RbjnIUL_d5TbrkTJ2c,58075
+autogluon/timeseries/version.py,sha256=OsP4ql15uiax-GUPsVJ1sGZMaTOWMBXWWMnLOe5tRrY,91
 autogluon/timeseries/configs/__init__.py,sha256=BTtHIPCYeGjqgOcvqb8qPD4VNX-ICKOg6wnkew1cPOE,98
 autogluon/timeseries/configs/presets_configs.py,sha256=cLat8ecLlWrI-SC5KLBDCX2SbVXaucemy2pjxJAtSY0,2543
 autogluon/timeseries/dataset/__init__.py,sha256=UvnhAN5tjgxXTHoZMQDy64YMDj4Xxa68yY7NP4vAw0o,81
 autogluon/timeseries/dataset/ts_dataframe.py,sha256=W3VE65lFyWmqMQ3XHN4Jhrqf_dO1EOLneNL2QDvVxeY,48120
-autogluon/timeseries/metrics/__init__.py,sha256=dJCrZ2cHwqhqNctwQjwG-FHgGUmzIFT-D0z72f4RAVM,2104
-autogluon/timeseries/metrics/abstract.py,sha256=CHUZB6xt9oF9yijSOjgGtjLuKo2X0mT6dQDuwg4ZzpU,8192
-autogluon/timeseries/metrics/point.py,sha256=2nlieQcPBCI9hXMT3v0Oe802ykZDuzvEtDpunzt0IVA,15785
-autogluon/timeseries/metrics/quantile.py,sha256=wvFeDMvRf1mFurhvVr_7g13Kg-hKIRoW4y9t2no_e7A,3969
+autogluon/timeseries/metrics/__init__.py,sha256=6v6aFjC_zbSTcSjU2hr0_LQf3MhCArMgKow3IfU2g-M,3477
+autogluon/timeseries/metrics/abstract.py,sha256=Z4ThftPBgLl9AgPQoJWcjHpVaOWWr6X3s3WwpvpAUOg,11818
+autogluon/timeseries/metrics/point.py,sha256=xllyGh11otbmUVHyIaceROPR3qyllWPQ9xlSmIGI3EI,18306
+autogluon/timeseries/metrics/quantile.py,sha256=vhmETtjPsIfVlvtILNAT6F2PtIDNPrOroy-U1FQbgw8,4632
 autogluon/timeseries/metrics/utils.py,sha256=HuDe1BNe8yJU4f_DKM913nNrUueoRaw6zhxm1-S20s0,910
 autogluon/timeseries/models/__init__.py,sha256=MYD9JJ-wUDE5B6jW6E6LU2eXQ6vflfQBvqQJkdzJa3A,1189
-autogluon/timeseries/models/presets.py,sha256=BdSTW91-flgqhVNuZIvqEf7wUj1iB6BPger4tJaoAZQ,12322
+autogluon/timeseries/models/presets.py,sha256=HEACiRpnY6dcff7W44gnM0x1KRgr2bNf5D6zcaHgHxo,12201
 autogluon/timeseries/models/abstract/__init__.py,sha256=Htfkjjc3vo92RvyM8rIlQ0PLWt3jcrCKZES07UvCMV0,146
-autogluon/timeseries/models/abstract/abstract_timeseries_model.py,sha256=Mj0bx45A2zy9Vzhcd7xjct3KUJGnMKLTozTMvtdsViw,33059
+autogluon/timeseries/models/abstract/abstract_timeseries_model.py,sha256=94TG7tsdfENP41QATr4IeMofaFt8ySjrrrH4MxZZ3Xc,32104
 autogluon/timeseries/models/abstract/model_trial.py,sha256=ENPg_7nsdxIvaNM0o0UShZ3x8jFlRmwRc5m0fGPC0TM,3720
 autogluon/timeseries/models/abstract/tunable.py,sha256=SFl4vjkb6BfFFaRPVdftnnLYlIyCThutLHxiiAlV6tY,7168
 autogluon/timeseries/models/autogluon_tabular/__init__.py,sha256=r9i6jWcyeLHYClkcMSKRVsfrkBUMxpDrTATNTBc_qgQ,136
@@ -36,7 +36,7 @@ autogluon/timeseries/models/chronos/pipeline/utils.py,sha256=dtDX5Pyu95bGv7qmqgf
 autogluon/timeseries/models/ensemble/__init__.py,sha256=_BivnZaOWJiIvu93IQy0mrLdCZKT2NHHSqkf31hwF2s,158
 autogluon/timeseries/models/ensemble/abstract.py,sha256=ie-BKD4JIkQQoKqtf6sYI5Aix7dSgywFsSdeGPxoElk,5821
 autogluon/timeseries/models/ensemble/basic.py,sha256=BRPWg_Wgfb87iInFSoTRE75BRHaovRR5HFRvzxET_wU,3423
-autogluon/timeseries/models/ensemble/greedy.py,sha256=2MVLTPvJ9Khuqri1gwQlo0RmKFeWK4qFkEcLH1Dh41E,7362
+autogluon/timeseries/models/ensemble/greedy.py,sha256=oW2d3-cce1Xck3NOtTh_8uHnjmc-2hGntPGoJQHUibE,7213
 autogluon/timeseries/models/gluonts/__init__.py,sha256=asC1PTj4j9xMbilvk1IT1julnpeoKbv5ZNuAR6-DFgA,361
 autogluon/timeseries/models/gluonts/abstract_gluonts.py,sha256=35T8rty6sPGiaSFNpiVNmeseo1_qpn664UcWo92W5eI,32906
 autogluon/timeseries/models/gluonts/torch/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -60,11 +60,11 @@ autogluon/timeseries/utils/datetime/base.py,sha256=3NdsH3NDq4cVAOSoy3XpaNixyNlbj
 autogluon/timeseries/utils/datetime/lags.py,sha256=gQDk5_zmsY5DUWDUpSaCKYkQ9nHKKY-LsywJQRAoYSk,5988
 autogluon/timeseries/utils/datetime/seasonality.py,sha256=YK_2k8hvYIMW-sJPnjGWRtCnvIOthwA2hATB3nwVoD4,834
 autogluon/timeseries/utils/datetime/time_features.py,sha256=MjLi3zQ00uWWJtXH9oGX2GJkTbvjdSiuabSa4kcVuxE,2672
-autogluon.timeseries-1.2.1b20250425.dist-info/LICENSE,sha256=CeipvOyAZxBGUsFoaFqwkx54aPnIKEtm9a5u2uXxEws,10142
-autogluon.timeseries-1.2.1b20250425.dist-info/METADATA,sha256=FQJvTO-OW1GKRFf4RMRuyNUgzGJscd3nXb7it95OrGM,12737
-autogluon.timeseries-1.2.1b20250425.dist-info/NOTICE,sha256=7nPQuj8Kp-uXsU0S5so3-2dNU5EctS5hDXvvzzehd7E,114
-autogluon.timeseries-1.2.1b20250425.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
-autogluon.timeseries-1.2.1b20250425.dist-info/namespace_packages.txt,sha256=giERA4R78OkJf2ijn5slgjURlhRPzfLr7waIcGkzYAo,10
-autogluon.timeseries-1.2.1b20250425.dist-info/top_level.txt,sha256=giERA4R78OkJf2ijn5slgjURlhRPzfLr7waIcGkzYAo,10
-autogluon.timeseries-1.2.1b20250425.dist-info/zip-safe,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
-autogluon.timeseries-1.2.1b20250425.dist-info/RECORD,,
+autogluon.timeseries-1.2.1b20250426.dist-info/LICENSE,sha256=CeipvOyAZxBGUsFoaFqwkx54aPnIKEtm9a5u2uXxEws,10142
+autogluon.timeseries-1.2.1b20250426.dist-info/METADATA,sha256=Wbji_e1nnEoqojfqovvTcoocs8z-u88kylcXK9PnXkA,12737
+autogluon.timeseries-1.2.1b20250426.dist-info/NOTICE,sha256=7nPQuj8Kp-uXsU0S5so3-2dNU5EctS5hDXvvzzehd7E,114
+autogluon.timeseries-1.2.1b20250426.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
+autogluon.timeseries-1.2.1b20250426.dist-info/namespace_packages.txt,sha256=giERA4R78OkJf2ijn5slgjURlhRPzfLr7waIcGkzYAo,10
+autogluon.timeseries-1.2.1b20250426.dist-info/top_level.txt,sha256=giERA4R78OkJf2ijn5slgjURlhRPzfLr7waIcGkzYAo,10
+autogluon.timeseries-1.2.1b20250426.dist-info/zip-safe,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
+autogluon.timeseries-1.2.1b20250426.dist-info/RECORD,,