fix merge conflict

tsururu/dataset.py

@@ -397,7 +397,6 @@ def __init__(
             elif column_type == "datetime":
                 data[column_name] = pd.to_datetime(data[column_name])
 
-        self.seq_data = data
         self.columns_and_features_params = columns_and_features_params
         self.history = history
         self.step = step
@@ -406,6 +405,7 @@ def __init__(
         self.date_column = columns_and_features_params["date"]["column"][0]
         self.delta = delta
         self.print_freq_period_info()
+        self.seq_data = data.sort_values(["id", "date"])
 
     def make_padded_test(
         self,

tsururu/models.py

@@ -46,6 +46,7 @@ def __init__(
         self.get_num_iterations = get_num_iterations
         if self.get_num_iterations:
             self.num_iterations = []
+        self.columns = None
 
     def initialize_validator(self):
         """Initialization of the sample generator for training the model
@@ -102,6 +103,10 @@ def __init__(
         super().__init__(get_num_iterations, validation_params, model_params)
 
     def fit(self, X: pd.DataFrame, y: NDArray[np.floating]) -> None:
+        # Initialize columns' order and reorder columns
+        self.columns = sorted(X.columns)
+        X = X[self.columns]
+
         # Initialize cv object
         cv = self.initialize_validator()
 
@@ -152,6 +157,9 @@ def fit(self, X: pd.DataFrame, y: NDArray[np.floating]) -> None:
         print(f"Std: {np.std(self.scores).round(4)}")
 
     def predict(self, X: pd.DataFrame) -> NDArray[np.floating]:
+        # Reorder columns
+        X = X[self.columns]
+
         models_preds = [model.predict(X) for model in self.models]
         y_pred = np.mean(models_preds, axis=0)
         return y_pred

tsururu/strategies.py

@@ -396,9 +396,12 @@ def _generate_X_y(
                     )
                     if role == "id":
                         self.id_feature_column = current_X
-                    final_X = pd.concat((final_X, current_X), axis=1)
+                    if not column_params["drop_raw_feature"]:
+                        final_X = pd.concat((final_X, current_X), axis=1)
             else:
                 for transformer_name, transformer_params in column_params["features"].items():
+                    assert not (role != "target" and transformer_params.get("transform_target")), "It is not possible to use transform_target=True with transformers for exogenous variables"
+
                     transformer_init_params = {
                         "transformer_name": transformer_name,
                         "transformer_params": {

tsururu/transformers.py

@@ -1,13 +1,17 @@
 from __future__ import annotations
 from typing import List, Union, Tuple, Optional
 from numpy.typing import NDArray
+import re
 
 import numpy as np
 import pandas as pd
 import holidays
 
 from .dataset import IndexSlicer
 
+LAG_TRANSFORMER_MASK = r"lag_\d+__"
+SEASON_TRANSFORMER_MASK = r"season_\w+__"
+
 date_attrs = {
     "y": "year",
     "m": "month",
@@ -81,10 +85,14 @@ def fit(
             Fitted transformer.
         """
         self.columns = raw_ts_X.columns[
-            np.hstack(
-                [raw_ts_X.columns.str.contains(raw_column_name) for raw_column_name in columns]
+            np.any(
+                [raw_ts_X.columns.str.contains(
+                    fr"{LAG_TRANSFORMER_MASK}{re.escape(raw_column_name)}$|{SEASON_TRANSFORMER_MASK}{re.escape(raw_column_name)}$|^{re.escape(raw_column_name)}$"
+                ) for raw_column_name in columns],
+                axis=0
             )
         ]
+
         self.id_column = id_column
         self.transform_train = transform_train
         self.transform_target = transform_target