src/delta_barth/analysis/forecast.py

@@ -1,6 +1,8 @@
 from __future__ import annotations
 
 import datetime
+# --- new: for calculating timedelta
+from dateutil.relativedelta import relativedelta
 import math
 from collections.abc import Mapping, Set
 from dataclasses import asdict
@@ -215,8 +217,16 @@ def _process_sales(
 
     features = ["jahr", "monat"]
     target = SALES_FEAT
-    current_year = datetime.datetime.now().year
+
-    first_year = cast(int, df_cust["jahr"].min())
+    # --- new: not necessary anymore
+    #current_year = datetime.datetime.now().year
+    #first_year = cast(int, df_cust["jahr"].min())
+
+    # --- new: dates und forecast
+    #last_date = pd.to_datetime(monthly_sum.index[-1], format="%m.%Y")
+    last_date = pd.to_datetime(datetime.now().strftime("%m.%Y"), format="%m.%Y")
+    future_dates = pd.date_range(start=last_date + pd.DateOffset(months=1), periods=6, freq="MS")
+    forecast = pd.DataFrame({"datum": future_dates.strftime("%m.%Y")}).set_index("datum")
 
     # Randomized Search
     kfold = KFold(n_splits=5, shuffle=True)
@@ -231,8 +241,9 @@ def _process_sales(
         "early_stopping_rounds": [20, 50],
     }
 
-    # --- new: best_estimator
+    # --- new: best_estimator (internal usage)
-    best_estimator: BestEstimatorXGBRegressor | None = None
+    best_estimator = None
+
     best_params: BestParametersXGBRegressor | None = None
     best_score_mae: float | None = float("inf")
     best_score_r2: float | None = None
@@ -240,20 +251,19 @@ def _process_sales(
     too_few_month_points: bool = True
     forecast: pd.DataFrame | None = None
 
-    # --- new: dates und forecast
-    #last_date = pd.to_datetime(monthly_sum.index[-1], format="%m.%Y")
-    last_date = pd.to_datetime(datetime.now().strftime("%m.%Y"), format="%m.%Y")
-    future_dates = pd.date_range(start=last_date + pd.DateOffset(months=1), periods=anzahl, freq="MS")
-    forecast = pd.DataFrame({"Datum": future_dates.strftime("%m.%Y")}).set_index("Datum")
-
     dates = monthly_sum.index
-    for index, i in enumerate(range(len(dates)-36, -1, -12)):
+    # --- new: use monthly basis for time windows
-        current_date = dates[i]
+    starting_date = datetime.now() - relativedelta(months=36)
-        split_date = dates[-anzahl]
+    #starting_date = dates.max() - relativedelta(months=36)
+    start_index = next((i for i, date in enumerate(dates) if date >= starting_date), len(dates) - 1)
+
+    for index, i in enumerate(range(start_index, -1, -12)):
+        start_date = dates[i]
+        split_date = dates[-6]
 
         train = cast(
             pd.DataFrame,
-            monthly_sum.loc[current_date:split_date].copy(),  # type: ignore
+            monthly_sum.loc[start_date:split_date].copy(),  # type: ignore
         )
         test = cast(
             pd.DataFrame,
@@ -287,12 +297,13 @@ def _process_sales(
                     best_params = cast(BestParametersXGBRegressor, rand.best_params_)
                     best_score_mae = error
                     best_score_r2 = cast(float, r2_score(y_test, y_pred))
-                    best_start_year = start_year
+                    # --- new: use store start_date in best_start_year
+                    best_start_year = start_date
 
     # --- new: use best_estimator to calculate future values and store them in forecast
     if best_estimator is not None:
         X_future = pd.DataFrame({"jahr": future_dates.year, "monat": future_dates.month}, index=future_dates)
-        y_future = rand.best_estimator_.predict(X_future)
+        y_future = best_estimator.predict(X_future)
         forecast["vorhersage"] = y_future  
         
     best_score_mae = best_score_mae if not math.isinf(best_score_mae) else None