openstef 3.2.70__tar.gz → 3.2.71__tar.gz

{openstef-3.2.70 → openstef-3.2.71}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: openstef
-Version: 3.2.70
+Version: 3.2.71
 Summary: Open short term energy forecaster
 Home-page: https://github.com/OpenSTEF/openstef
 Author: Alliander N.V

{openstef-3.2.70 → openstef-3.2.71}/openstef/model/objective.py

@@ -97,7 +97,12 @@ class RegressorObjective:
                 "stratification_min_max": self.model_type != MLModelType.ProLoaf,
                 "back_test": True,
             }
-        (self.train_data, self.validation_data, self.test_data,) = self.split_func(
+        (
+            self.train_data,
+            self.validation_data,
+            self.test_data,
+            self.operational_score_data,
+        ) = self.split_func(
             self.input_data,
             test_fraction=self.test_fraction,
             validation_fraction=self.validation_fraction,

{openstef-3.2.70 → openstef-3.2.71}/openstef/model_selection/model_selection.py

@@ -113,7 +113,7 @@ def split_data_train_validation_test(
     validation dataset. In an operational setting the following sequence is
     returned (when using stratification):
 
-    Test >> Train >> Validation
+    Train >> Validation (and the test is the Train and Validation combined.)
 
     For a back test (indicated with argument "back_test") the following sequence
     is returned:
@@ -141,6 +141,7 @@ def split_data_train_validation_test(
         - Test data.
 
     """
+    test_fraction = test_fraction if back_test else 0
     train_fraction = 1 - (test_fraction + validation_fraction)
     if train_fraction < 0:
         raise ValueError(
@@ -172,10 +173,18 @@ def split_data_train_validation_test(
         start_date_test = end_date - np.round(number_indices * test_fraction) * delta
         test_data = data_[start_date_test:]
         train_val_data = data_[:start_date_test]
+        operational_score_data = (
+            pd.DataFrame()
+        )  # Empty because a backtest is no operational setting.
     else:
         start_date_val = start_date + np.round(number_indices * test_fraction) * delta
-        test_data = data_[:start_date_val]
+        test_data = data_[
+            :start_date_val
+        ]  # Empty as all data is used for training in an operational setting.
         train_val_data = data_[start_date_val:]
+        operational_score_data = data_.copy(deep=True).reset_index(
+            drop=True
+        )  # Used to check wether a new operationally train model is better than the old one.
 
     if stratification_min_max and (
         len(set(train_val_data.index.date)) >= min_days_for_stratification
@@ -248,11 +257,7 @@ def split_data_train_validation_test(
     validation_data = validation_data.sort_index()
     test_data = test_data.sort_index()
 
-    return (
-        train_data,
-        validation_data,
-        test_data,
-    )
+    return (train_data, validation_data, test_data, operational_score_data)
 
 
 def backtest_split_default(
@@ -286,14 +291,14 @@ def backtest_split_default(
         for ifold in range(n_folds):
             test_data = data[data["random_fold"] == ifold].sort_index()
 
-            (train_data, validation_data, _,) = split_data_train_validation_test(
+            (train_data, validation_data, _, _) = split_data_train_validation_test(
                 data[data["random_fold"] != ifold].iloc[:, :-2],
                 test_fraction=0,
                 back_test=True,
                 stratification_min_max=stratification_min_max,
             )
 
-            yield train_data, validation_data, test_data.iloc[:, :-2]
+            yield train_data, validation_data, test_data.iloc[:, :-2], pd.DataFrame()
     else:
         yield split_data_train_validation_test(
             data,

{openstef-3.2.70 → openstef-3.2.71}/openstef/pipeline/train_create_forecast_backtest.py

@@ -82,10 +82,14 @@ def train_model_and_forecast_back_test(
     ) = zip(
         *(
             train_model_and_forecast_test_core(
-                pj, modelspecs, train_data, validation_data, test_data
+                pj,
+                modelspecs,
+                train_data,
+                validation_data,
+                test_data,
             )
             + (train_data, validation_data, test_data)
-            for train_data, validation_data, test_data in backtest_split_func(
+            for train_data, validation_data, test_data, _ in backtest_split_func(
                 data_with_features, n_folds, **backtest_split_args
             )
         )

{openstef-3.2.70 → openstef-3.2.71}/openstef/pipeline/train_model.py

@@ -176,7 +176,14 @@ def train_model_pipeline_core(
     logger = structlog.get_logger(__name__)
 
     # Call common pipeline
-    model, report, train_data, validation_data, test_data = train_pipeline_common(
+    (
+        model,
+        report,
+        train_data,
+        validation_data,
+        test_data,
+        operational_score_data,
+    ) = train_pipeline_common(
         pj,
         model_specs,
         input_data,
@@ -192,8 +199,8 @@ def train_model_pipeline_core(
             combined = combined.iloc[:, :-1]
 
         x_data, y_data = (
-            combined.iloc[:, 1:-1],
-            combined.iloc[:, 0],
+            operational_score_data.iloc[:, 1:-1],
+            operational_score_data.iloc[:, 0],
         )
 
         # Score method always returns R^2
@@ -260,7 +267,12 @@ def train_pipeline_common(
         horizons=horizons,
     )
 
-    train_data, validation_data, test_data = train_pipeline_step_split_data(
+    (
+        train_data,
+        validation_data,
+        test_data,
+        operational_score_data,
+    ) = train_pipeline_step_split_data(
         data_with_features=data_with_features,
         pj=pj,
         test_fraction=test_fraction,
@@ -284,7 +296,7 @@ def train_pipeline_common(
         validation_data["forecast"] = model.predict(validation_data.iloc[:, 1:-1])
         test_data["forecast"] = model.predict(test_data.iloc[:, 1:-1])
 
-    return model, report, train_data, validation_data, test_data
+    return model, report, train_data, validation_data, test_data, operational_score_data
 
 
 def train_pipeline_step_load_model(
@@ -515,7 +527,7 @@ def train_pipeline_step_split_data(
             required_arguments=["data", "test_fraction"]
         )
 
-    train_data, validation_data, test_data = split_func(
+    train_data, validation_data, test_data, operational_score_data = split_func(
         data_with_features, test_fraction, **split_args
     )
 
@@ -523,4 +535,4 @@ def train_pipeline_step_split_data(
     if not test_data_predefined.empty:
         test_data = test_data_predefined
 
-    return train_data, validation_data, test_data
+    return train_data, validation_data, test_data, operational_score_data

{openstef-3.2.70 → openstef-3.2.71}/openstef.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: openstef
-Version: 3.2.70
+Version: 3.2.71
 Summary: Open short term energy forecaster
 Home-page: https://github.com/OpenSTEF/openstef
 Author: Alliander N.V

{openstef-3.2.70 → openstef-3.2.71}/setup.py

@@ -29,7 +29,7 @@ def read_long_description_from_readme():
 
 setup(
     name="openstef",
-    version="3.2.70",
+    version="3.2.71",
     packages=find_packages(include=["openstef", "openstef.*"]),
     description="Open short term energy forecaster",
     long_description=read_long_description_from_readme(),