prepare fix for joblib support in C# environment, related to #23

src/delta_barth/__init__.py

@@ -0,0 +1,23 @@
+import multiprocessing
+import os
+import sys
+from pathlib import Path
+
+# os.environ["JOBLIB_DISABLE_SHARED_MEMORY"] = "1"
+
+interpreter = r"A:\Arbeitsaufgaben\Delta-Barth\cs-wrapper\dopt.DeltaBarth\python\python.exe"
+intp_pth = Path(interpreter).resolve()
+
+assert intp_pth.exists(), f"interpreter path seems not to exist: {intp_pth}"
+
+# multiprocessing.set_executable(str(intp_pth))
+# setattr(sys, "frozen", True) # !! causes termination
+# sys.executable = str(intp_pth)
+setattr(sys, "executable", str(intp_pth))
+setattr(sys, "_base_executable", str(intp_pth))
+# multiprocessing.set_start_method("spawn", force=True)
+
+
+target = Path(r"A:\Arbeitsaufgaben\Delta-Barth\cs-wrapper\dopt.DeltaBarth")
+file = target / "executed.txt"
+file.touch()

src/delta_barth/analysis/forecast.py

@@ -8,6 +8,11 @@ from dataclasses import asdict
 from datetime import datetime as Datetime
 from typing import TYPE_CHECKING, Final, TypeAlias, cast
 
+import joblib
+import joblib.externals
+import joblib.externals.loky
+import joblib.externals.loky.backend
+import joblib.externals.loky.backend.popen_loky_win32
 import numpy as np
 import pandas as pd
 import scipy.stats
@@ -291,18 +296,22 @@ def _process_sales(
         if len(train[train[SALES_FEAT] > 0]) >= (base_num_data_points_months + 10 * add_year):
             too_few_month_points = False
 
-            rand = RandomizedSearchCV(
-                XGBRegressor(),
-                params,
-                scoring="neg_mean_absolute_error",
-                cv=kfold,
-                n_jobs=-1,
-                n_iter=100,
-                verbose=0,
-            )
-            rand.fit(
-                X_train, y_train, eval_set=[(X_train, y_train), (X_test, y_test)], verbose=0
-            )
+            with joblib.parallel_config(backend="loky"):
+                rand = RandomizedSearchCV(
+                    XGBRegressor(),
+                    params,
+                    scoring="neg_mean_absolute_error",
+                    cv=kfold,
+                    n_jobs=-1,
+                    n_iter=100,
+                    verbose=0,
+                )
+                rand.fit(
+                    X_train,
+                    y_train,
+                    eval_set=[(X_train, y_train), (X_test, y_test)],
+                    verbose=0,
+                )
             y_pred = rand.best_estimator_.predict(X_test)  # type: ignore
 
             if len(np.unique(y_pred)) != 1: