lecrapaud 0.9.3__tar.gz → 0.10.0__tar.gz

{lecrapaud-0.9.3 → lecrapaud-0.10.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: lecrapaud
-Version: 0.9.3
+Version: 0.10.0
 Summary: Framework for machine and deep learning, with regression, classification and time series analysis
 License: Apache License
 Author: Pierre H. Gallet

{lecrapaud-0.9.3 → lecrapaud-0.10.0}/lecrapaud/api.py

@@ -167,14 +167,10 @@ class ExperimentEngine:
                 else:
                     scores_reg.append(score)
 
-            # renaming pred column and concatenating with initial data
+            # renaming and concatenating with initial data
             if isinstance(y_pred, pd.DataFrame):
-                y_pred.rename(
-                    columns={"PRED": f"TARGET_{target_number}_PRED"}, inplace=True
-                )
-                new_data = pd.concat(
-                    [new_data, y_pred[f"TARGET_{target_number}_PRED"]], axis=1
-                )
+                y_pred = y_pred.add_prefix(f"TARGET_{target_number}_")
+                new_data = pd.concat([new_data, y_pred], axis=1)
 
             else:
                 y_pred.name = f"TARGET_{target_number}_PRED"
@@ -294,3 +290,8 @@ class ExperimentEngine:
 
     def get_feature_summary(self):
         return pd.read_csv(f"{self.experiment.path}/feature_summary.csv")
+
+    def get_threshold(self, target_number: int):
+        return joblib.load(
+            f"{self.experiment.path}/TARGET_{target_number}/thresholds.pkl"
+        )

{lecrapaud-0.9.3 → lecrapaud-0.10.0}/lecrapaud/experiment.py

@@ -39,7 +39,9 @@ def create_experiment(
 
     with get_db() as db:
         all_targets = Target.get_all(db=db)
-        targets = [target for target in all_targets if target.name in data.columns]
+        targets = [
+            target for target in all_targets if target.name in data.columns.str.upper()
+        ]
         experiment_name = f"{experiment_name}_{groups["number_of_groups"] if group_column else 'ng'}_{corr_threshold}_{percentile}_{max_features}_{dates['start_date'].date() if date_column else 'nd'}_{dates['end_date'].date() if date_column else 'nd'}"
 
         experiment_dir = f"{tmp_dir}/{experiment_name}"

{lecrapaud-0.9.3 → lecrapaud-0.10.0}/lecrapaud/feature_selection.py

@@ -47,7 +47,6 @@ from lecrapaud.db import (
     FeatureSelection,
     FeatureSelectionRank,
 )
-from lecrapaud.db.session import get_db
 from lecrapaud.search_space import all_models
 
 # Annoying Warnings
@@ -120,6 +119,7 @@ class FeatureSelectionEngine:
             target_id=target.id,
             experiment_id=self.experiment_id,
         )
+        feature_map = {f.name: f.id for f in Feature.get_all(limit=20000)}
 
         if feature_selection.best_features_path and os.path.exists(
             feature_selection.best_features_path
@@ -156,21 +156,30 @@ class FeatureSelectionEngine:
 
         if target_type == "classification" and self.X_categorical.shape[1] > 0:
             feat_scores = self.select_categorical_features(percentile=percentile)
-            with get_db() as db:
-                for row in feat_scores.itertuples(index=False):
-                    feature = Feature.find_by(name=row.features, db=db)
-                    FeatureSelectionRank.upsert(
-                        ["feature_selection_id", "feature_id", "method"],
-                        db=db,
-                        score=row.score,
-                        pvalue=row.pvalue,
-                        support=row.support,
-                        rank=row.rank,
-                        method=row.method,
-                        training_time=row.training_time,
-                        feature_selection_id=feature_selection.id,
-                        feature_id=feature.id,
-                    )
+            rows = []
+            for row in feat_scores.itertuples(index=False):
+                feature_id = feature_map.get(row.features)
+
+                rows.append(
+                    {
+                        "feature_selection_id": feature_selection.id,
+                        "feature_id": feature_id,
+                        "method": row.method,
+                        "score": row.score,
+                        "pvalue": row.pvalue,
+                        "support": row.support,
+                        "rank": row.rank,
+                        "training_time": row.training_time,
+                    }
+                )
+
+            if len(rows) == 0:
+                logger.warning(
+                    f"No categorical features selected for TARGET_{target_number}"
+                )
+
+            FeatureSelectionRank.bulk_upsert(rows=rows)
+
             categorical_features_selected = feat_scores[feat_scores["support"]][
                 "features"
             ].values.tolist()
@@ -238,30 +247,26 @@ class FeatureSelectionEngine:
 
         logger.info("Inserting feature selection results to db...")
         rows = []
-        with get_db() as db:
-            feature_map = {f.name: f.id for f in Feature.get_all(db=db, limit=20000)}
-            for row in feat_scores.itertuples(index=False):
-                feature_id = feature_map.get(row.features)
-                if not feature_id:
-                    continue  # or raise if feature must exist
-
-                rows.append(
-                    {
-                        "feature_selection_id": feature_selection.id,
-                        "feature_id": feature_id,
-                        "method": row.method,
-                        "score": row.score,
-                        "pvalue": None if pd.isna(row.pvalue) else row.pvalue,
-                        "support": row.support,
-                        "rank": row.rank,
-                        "training_time": row.training_time,
-                    }
-                )
+        for row in feat_scores.itertuples(index=False):
+            feature_id = feature_map.get(row.features)
+
+            rows.append(
+                {
+                    "feature_selection_id": feature_selection.id,
+                    "feature_id": feature_id,
+                    "method": row.method,
+                    "score": row.score,
+                    "pvalue": None if pd.isna(row.pvalue) else row.pvalue,
+                    "support": row.support,
+                    "rank": row.rank,
+                    "training_time": row.training_time,
+                }
+            )
 
-            if len(rows) == 0:
-                raise ValueError(f"No features selected for TARGET_{target_number}")
+        if len(rows) == 0:
+            logger.warning(f"No numerical features selected for TARGET_{target_number}")
 
-            FeatureSelectionRank.bulk_upsert(rows=rows, db=db)
+        FeatureSelectionRank.bulk_upsert(rows=rows)
 
         # Merge the results
         logger.info("Merging feature selection methods...")

{lecrapaud-0.9.3 → lecrapaud-0.10.0}/lecrapaud/model_selection.py

@@ -1537,6 +1537,9 @@ def apply_thresholds(
                     pd.Series(np.where(exceeded, col, -np.inf), index=pred_proba.index)
                 )
 
+        # For each row:
+        # 1. If any threshold is exceeded, take the class with highest probability among exceeded
+        # 2. If no threshold is exceeded, take the class with highest probability overall
         if class_predictions:
             preds_df = pd.concat(class_predictions, axis=1)
             probs_df = pd.concat(class_probabilities, axis=1)

{lecrapaud-0.9.3 → lecrapaud-0.10.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "lecrapaud"
-version = "0.9.3"
+version = "0.10.0"
 description = "Framework for machine and deep learning, with regression, classification and time series analysis"
 authors = [
     {name = "Pierre H. Gallet"}