m2u

Signed-off-by: oneJue <[email protected]>

ts_benchmark/baselines/duet/duet.py

@@ -215,7 +215,12 @@ def validate(self, valid_data_loader, criterion):
         self.model.train()
         return total_loss
 
-    def forecast_fit(self, train_valid_data: pd.DataFrame, train_ratio_in_tv: float) -> "ModelBase":
+    def forecast_fit(
+        self,
+        train_valid_data: pd.DataFrame,
+        covariates: Optional[Dict],
+        train_ratio_in_tv: float,
+    ) -> "ModelBase":
         """
         Train the model.
 

ts_benchmark/baselines/time_series_library/adapters_for_transformers.py

@@ -262,21 +262,22 @@ def validate(
     def forecast_fit(
         self,
         train_valid_data: pd.DataFrame,
-        covariates: Optional[Dict],
+        covariates: dict,
         train_ratio_in_tv: float,
     ) -> "ModelBase":
         """
         Train the model.
 
         :param train_data: Time series data used for training.
-        :param covariates: Additional external variables
+        :param covariates: Additional external variables.
         :param train_ratio_in_tv: Represents the splitting ratio of the training set validation set. If it is equal to 1, it means that the validation set is not partitioned.
         :return: The fitted model object.
         """
         exog_dim = -1
-        if covariates["exog"] is not None:
-            exog_dim = covariates["exog"].shape[-1]
-            train_valid_data = pd.concat([train_valid_data, covariates["exog"]], axis=1)
+        exog_data = covariates.get("exog")
+        if exog_data is not None:
+            exog_dim = exog_data.shape[-1]
+            train_valid_data = pd.concat([train_valid_data, exog_data], axis=1)
 
         if train_valid_data.shape[1] == 1:
             train_drop_last = False
@@ -396,36 +397,40 @@ def forecast_fit(
             adjust_learning_rate(optimizer, epoch + 1, config)
 
     def forecast(
-        self, horizon: int, covariates: Optional[Dict], train: pd.DataFrame
+        self, horizon: int, covariates: dict, series: pd.DataFrame
     ) -> np.ndarray:
         """
         Make predictions.
 
         :param horizon: The predicted length.
         :param covariates: Additional external variables
-        :param testdata: Time series data used for prediction.
+        :param series: Time series data used for prediction.
         :return: An array of predicted results.
         """
         exog_dim = -1
         if covariates["exog"] is not None:
             exog_dim = covariates["exog"].shape[-1]
-            train = pd.concat([train, covariates["exog"]], axis=1)
+            series = pd.concat([series, covariates["exog"]], axis=1)
+        if exog_dim != -1 and horizon != self.config.output_chunk_length:
+            raise ValueError(
+                f"Error: 'exog' is enabled during training, but horizon ({horizon}) != output_chunk_length ({self.config.output_chunk_length}) during forecast."
+            )
 
         if self.early_stopping.check_point is not None:
             self.model.load_state_dict(self.early_stopping.check_point)
 
         if self.config.norm:
-            train = pd.DataFrame(
-                self.scaler.transform(train.values),
-                columns=train.columns,
-                index=train.index,
+            series = pd.DataFrame(
+                self.scaler.transform(series.values),
+                columns=series.columns,
+                index=series.index,
             )
 
         if self.model is None:
             raise ValueError("Model not trained. Call the fit() function first.")
 
         config = self.config
-        train, test = split_time(train, len(train) - config.seq_len)
+        series, test = split_time(series, len(series) - config.seq_len)
 
         # Additional timestamp marks required to generate transformer class methods
         test = self.padding_data_for_forecast(test)
@@ -517,6 +522,10 @@ def batch_forecast(
             input_np = np.concatenate(
                 (input_np, input_data["covariates"]["exog"]), axis=2
             )
+        if exog_dim != -1 and horizon != self.config.output_chunk_length:
+            raise ValueError(
+                f"Error: 'exog' is enabled during training, but horizon ({horizon}) != output_chunk_length ({self.config.output_chunk_length}) during forecast."
+            )
 
         if self.config.norm:
             origin_shape = input_np.shape

ts_benchmark/evaluation/strategy/rolling_forecast.py

@@ -21,7 +21,7 @@ def __init__(
         self,
         series: pd.DataFrame,
         index_list: List[int],
-        covariates: Optional[Dict] = None,
+        covariates: dict,
     ):
         self.series = series
         self.index_list = index_list
@@ -294,12 +294,8 @@ def _eval_sample(
         all_rolling_predict = []
         for i, index in itertools.islice(enumerate(index_list), num_rollings):
             train, rest = split_time(series, index)
-            target_train, exog_train = split_channel(train)
-            test, _ = split_channel(
-                split_time(rest, horizon)[0].iloc[
-                    :, : target_train_valid_data.shape[-1]
-                ]
-            )
+            test, _ = split_channel(split_time(rest, horizon)[0], target_channel)
+            target_train, exog_train = split_channel(train, target_channel)
             covariates = {"exog": exog_train}
 
             start_inference_time = time.time()

ts_benchmark/models/model_base.py

@@ -50,7 +50,7 @@ def forecast_fit(
         self,
         train_data: pd.DataFrame,
         *,
-        covariates: Optional[Dict] = None,
+        covariates: dict,
         train_ratio_in_tv: float = 1.0,
         **kwargs
     ) -> "ModelBase":
@@ -75,9 +75,8 @@ def forecast(
         self,
         horizon: int,
         *,
-        covariates: Optional[Dict] = None,
         series: pd.DataFrame,
-        **kwargs
+        covariates: dict,
     ) -> np.ndarray:
         """
         Forecasting with the model

ts_benchmark/utils/data_processing.py

@@ -66,6 +66,7 @@ def split_channel(
 
     :param df: Input DataFrame to split.
     :param target_channel: Rules for selecting target columns. Can include:
+
         - Integers (positive/negative) for single column indices.
         - Lists/tuples of two integers representing slices (e.g., `[2,4]` selects columns 2-3).
         - If `None`, all columns are treated as target columns (exog becomes None).