lecrapaud 0.11.6__tar.gz → 0.12.1__tar.gz

{lecrapaud-0.11.6 → lecrapaud-0.12.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: lecrapaud
-Version: 0.11.6
+Version: 0.12.1
 Summary: Framework for machine and deep learning, with regression, classification and time series analysis
 License: Apache License
 Author: Pierre H. Gallet

{lecrapaud-0.11.6 → lecrapaud-0.12.1}/lecrapaud/api.py

@@ -128,7 +128,7 @@ class ExperimentEngine:
                 value = normalize_models_idx(value)
             setattr(self, key, value)
 
-    def train(self, data):
+    def train(self, data, best_params=None):
         logger.info("Running training...")
 
         data_eng = self.feature_engineering(data)
@@ -143,7 +143,7 @@ class ExperimentEngine:
         std_data, reshaped_data = self.preprocess_model(train, val, test)
         logger.info("Model preprocessing done.")
 
-        self.model_selection(std_data, reshaped_data)
+        self.model_selection(std_data, reshaped_data, best_params=best_params)
         logger.info("Model selection done.")
 
     def predict(self, new_data, verbose: int = 0):
@@ -321,7 +321,7 @@ class ExperimentEngine:
             data, scaled_data, reshaped_data = app.inference()
             return data, scaled_data, reshaped_data
 
-    def model_selection(self, data, reshaped_data):
+    def model_selection(self, data, reshaped_data, best_params=None):
         for target_number in self.target_numbers:
             app = ModelSelectionEngine(
                 data=data,
@@ -342,6 +342,7 @@ class ExperimentEngine:
                 perform_crossval=self.perform_crossval,
                 plot=self.plot,
                 preserve_model=self.preserve_model,
+                best_params=best_params[target_number],
             )
 
     def get_scores(self, target_number: int):

{lecrapaud-0.11.6 → lecrapaud-0.12.1}/lecrapaud/model_selection.py

@@ -721,6 +721,7 @@ class ModelSelectionEngine:
         plot=True,
         clean_dir=False,  # TODO: This has been unused because now feature_selection is in the target directory
         preserve_model=True,
+        best_params=None,
     ):
         """
         Selects the best models based on a target variable, optionally performing hyperparameter optimization
@@ -866,29 +867,33 @@ class ModelSelectionEngine:
             start = time.time()
             # Tuning hyperparameters
             if perform_hyperopt:
-                best_params = self.hyperoptimize(x_train, y_train, x_val, y_val, model)
-
-                # save best params
-                best_params_file = f"{self.target_dir}/best_params.json"
-                try:
-                    with open(best_params_file, "r") as f:
-                        json_dict = json.load(f)
-                except FileNotFoundError:
-                    json_dict = {}
-
-                json_dict[model.model_name] = serialize_for_json(best_params)
-                with open(best_params_file, "w") as f:
-                    json.dump(json_dict, f, indent=4)
+                model_best_params = self.hyperoptimize(
+                    x_train, y_train, x_val, y_val, model
+                )
+            elif best_params:
+                model_best_params = best_params[model_name]
             else:
                 try:
                     with open(f"{self.target_dir}/best_params.json") as f:
                         json_dict = json.load(f)
-                        best_params = json_dict[model_name]
+                        model_best_params = json_dict[model_name]
                 except Exception:
                     raise FileNotFoundError(
-                        f"Could not find {model_name} in current data. Try to run an hyperoptimization by setting `perform_hyperopt` to true"
+                        f"Could not find {model_name} in current data. Try to run an hyperoptimization by setting `perform_hyperopt` to true, pass `best_params`"
                     )
 
+            # save best params
+            best_params_file = f"{self.target_dir}/best_params.json"
+            try:
+                with open(best_params_file, "r") as f:
+                    json_dict = json.load(f)
+            except FileNotFoundError:
+                json_dict = {}
+
+            json_dict[model.model_name] = serialize_for_json(model_best_params)
+            with open(best_params_file, "w") as f:
+                json.dump(json_dict, f, indent=4)
+
             # Perform cross-validation of the best model on k-folds of train + val set
             if perform_crossval:
                 x_train_val = pd.concat([x_train, x_val, x_test], axis=0)
@@ -928,7 +933,7 @@ class ModelSelectionEngine:
                     # Train the model and get the score
                     if recurrent:
                         cross_validation_score, _, _ = self.train_model(
-                            params=best_params,
+                            params=model_best_params,
                             x_train=x_train_val[train_index],
                             y_train=y_train_val[train_index],
                             x_val=x_train_val[val_index],
@@ -937,7 +942,7 @@ class ModelSelectionEngine:
                         )
                     else:
                         cross_validation_score, _, _ = self.train_model(
-                            params=best_params,
+                            params=model_best_params,
                             x_train=x_train_val.iloc[train_index],
                             y_train=y_train_val.iloc[train_index],
                             x_val=x_train_val.iloc[val_index],
@@ -958,7 +963,7 @@ class ModelSelectionEngine:
 
                 # Retrain on entire training set, but keep score on cross-validation folds
                 best_score, best_model, best_pred = self.train_model(
-                    params=best_params,
+                    params=model_best_params,
                     x_train=pd.concat([x_train, x_val], axis=0),
                     y_train=pd.concat([y_train, y_val], axis=0),
                     x_val=x_test,
@@ -970,7 +975,7 @@ class ModelSelectionEngine:
                 # Evaluate on validation set
                 self.type_name = "validation"
                 best_score, best_model, best_pred = self.train_model(
-                    params=best_params,
+                    params=model_best_params,
                     x_train=pd.concat([x_train, x_val], axis=0),
                     y_train=pd.concat([y_train, y_val], axis=0),
                     x_val=x_test,
@@ -1016,7 +1021,7 @@ class ModelSelectionEngine:
             # Save model training metadata
             stop = time.time()
             training_time = stop - start
-            model_training.best_params = best_params
+            model_training.best_params = model_best_params
             model_training.model_path = model_path
             model_training.training_time = training_time
             model_training.save()
@@ -1138,7 +1143,7 @@ class ModelSelectionEngine:
         try:
             results = tuner.fit()
 
-            best_result = results.get_best_result(self.metric, "max")
+            best_result = results.get_best_result(self.metric, "min")
             best_params = best_result.config
             best_score = best_result.metrics
 

{lecrapaud-0.11.6 → lecrapaud-0.12.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "lecrapaud"
-version = "0.11.6"
+version = "0.12.1"
 description = "Framework for machine and deep learning, with regression, classification and time series analysis"
 authors = [
     {name = "Pierre H. Gallet"}