lecrapaud 0.18.2__py3-none-any.whl → 0.18.4__py3-none-any.whl

lecrapaud/db/models/experiment.py

@@ -25,7 +25,7 @@ from lecrapaud.db.models.score import Score
 
 from lecrapaud.db.models.base import Base, with_db
 from lecrapaud.db.models.utils import create_association_table
-from lecrapaud.utils import logger
+from lecrapaud.utils import logger, contains_best
 
 # jointures
 lecrapaud_experiment_target_association = create_association_table(
@@ -237,13 +237,27 @@ class Experiment(Base):
         Returns:
             Experiment or None: The experiment with the best score or None if not found
         """
+        experiments = db.query(cls).filter(cls.name.ilike(f"%{name}%")).all()
+        if not experiments:
+            logger.error(f"No experiments found with the given name: {name}")
+            return None
+
+        experiments = [
+            exp
+            for exp in experiments
+            if all(
+                [contains_best(f"{exp.path}/{target.name}") for target in exp.targets]
+            )
+        ]
+        if not experiments:
+            logger.error(
+                f"No fully trained experiments found with the given name: {name}"
+            )
+            return None
+
         if metric == "both":
             # Calculate a combined score: average of normalized RMSE and LogLoss
             # This ensures we're comparing apples to apples by normalizing the scores
-            experiments = db.query(cls).filter(cls.name.ilike(f"%{name}%")).all()
-            if not experiments:
-                logger.error(f"No experiments found with the given name: {name}")
-                return None
 
             # Get all scores
             rmse_scores = [e.avg_rmse for e in experiments if e.avg_rmse is not None]
@@ -284,11 +298,6 @@ class Experiment(Base):
 
         elif metric == "rmse" or metric == "logloss":
             # For single metric case (rmse or logloss)
-            # Need to get all experiments first to evaluate instance properties
-            experiments = db.query(cls).filter(cls.name.ilike(f"%{name}%")).all()
-
-            if not experiments:
-                return None
 
             # Filter out experiments without scores and sort by the selected metric
             filtered_experiments = []

lecrapaud/feature_engineering.py

@@ -220,7 +220,7 @@ class FeatureEngineeringEngine:
         Returns:
             pd.DataFrame: Original dataframe with new encoded columns added
         """
-
+        # TODO: target encoding needs to be fit / transform based at inference time.
         df: pd.DataFrame = self.data
         columns_te_groupby: list[list[str]] = self.columns_te_groupby
         columns_te_target: list[str] = self.columns_te_target

lecrapaud/model_selection.py

@@ -12,6 +12,7 @@ import glob
 from pathlib import Path
 import pickle
 from pydantic import BaseModel
+import ast
 
 os.environ["COVERAGE_FILE"] = str(Path(".coverage").resolve())
 
@@ -122,6 +123,7 @@ class ModelEngine:
         plot: bool = False,
         log_dir: str = None,
     ):
+        self.threshold = None
         self.path = path
         if path:
             self.load()
@@ -152,8 +154,6 @@ class ModelEngine:
         else:
             self.scaler_y = None
 
-        self.threshold = None
-
     def fit(self, *args):
         if self.recurrent:
             fit = self.fit_recurrent
@@ -1070,7 +1070,7 @@ class ModelSelectionEngine:
         best_score_overall = scores_tracking.iloc[0, :]
         best_model_name = best_score_overall["MODEL_NAME"]
         if self.target_type == "classification":
-            best_thresholds = best_score_overall["THRESHOLDS"]
+            best_thresholds = ast.literal_eval(best_score_overall["THRESHOLDS"])
             joblib.dump(best_thresholds, f"{self.target_dir}/thresholds.pkl")
         else:
             best_thresholds = None
@@ -1852,7 +1852,9 @@ def apply_thresholds(
     """
 
     # Case 1: Per-class thresholds
-    if isinstance(threshold, dict):
+    if not isinstance(threshold, int):
+        if isinstance(threshold, str):
+            threshold = ast.literal_eval(threshold)
         class_predictions = []
         class_probabilities = []
 
@@ -1862,6 +1864,7 @@ def apply_thresholds(
                 metrics.get("threshold") if isinstance(metrics, dict) else metrics[0]
             )
             if _threshold is not None:
+                class_label = int(class_label)
                 if class_label not in pred_proba.columns:
                     continue  # skip missing class
                 col = pred_proba[class_label]

{lecrapaud-0.18.2.dist-info → lecrapaud-0.18.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: lecrapaud
-Version: 0.18.2
+Version: 0.18.4
 Summary: Framework for machine and deep learning, with regression, classification and time series analysis
 License: Apache License
 Author: Pierre H. Gallet

{lecrapaud-0.18.2.dist-info → lecrapaud-0.18.4.dist-info}/RECORD

@@ -13,7 +13,7 @@ lecrapaud/db/alembic/versions/2025_08_25_1434-7ed9963e732f_add_best_score_to_mod
 lecrapaud/db/alembic.ini,sha256=Zw2rdwsKV6c7J1SPtoFIPDX08_oTP3MuUKnNxBDiY8I,3796
 lecrapaud/db/models/__init__.py,sha256=Lhyw9fVLdom0Fc6yIP-ip8FjkU1EwVwjae5q2VM815Q,740
 lecrapaud/db/models/base.py,sha256=J9ew-0z_-tnWAwhVvOmVDys2R6jPF_oSca_ny6wpXQE,7606
-lecrapaud/db/models/experiment.py,sha256=rgNpCNXMei5VhJDNKxelpwqv7iTxoPJ2kkffGaua2sA,14710
+lecrapaud/db/models/experiment.py,sha256=LjsMTY-PA9HZ27D2sz2fWy7HvwFqiS0dXKaiKF-S3k4,14868
 lecrapaud/db/models/feature.py,sha256=5o77O2FyRObnLOCGNj8kaPSGM3pLv1Ov6mXXHYkmnYY,1136
 lecrapaud/db/models/feature_selection.py,sha256=mk42xuw1Sm_7Pznfg7TNc5_S4hscdw79QgIe3Bt9ZRI,3245
 lecrapaud/db/models/feature_selection_rank.py,sha256=Ydsb_rAT58FoSH13wkGjGPByzsjPx3DITXgJ2jgZmow,2198
@@ -26,7 +26,7 @@ lecrapaud/db/models/utils.py,sha256=-a-nWWmpJ2XzidIxo2COVUTrGZIPYCfBzjhcszJj_bM,
 lecrapaud/db/session.py,sha256=E93WXcFFILFAIeH61ft2Egs7D-6caqs0oi4zCkO5Lq4,2822
 lecrapaud/directories.py,sha256=0LrANuDgbuneSLker60c6q2hmGnQ3mKHIztTGzTx6Gw,826
 lecrapaud/experiment.py,sha256=1xLWjOrqAxJh9CdXOx9ppQuRFRRj0GH-xYZqg-ty9hI,2463
-lecrapaud/feature_engineering.py,sha256=fqvYuGDGb7oZJYPq1WgJlrXs5XLk8AwvqfCiyapK7Yc,39115
+lecrapaud/feature_engineering.py,sha256=ib1afBrwqePiXUaw0Cpe6hY3VNl5afg8YVntb88SCT4,39199
 lecrapaud/feature_selection.py,sha256=6ry-oVPQHbipm1XSE5YsH7AY0lQFt4CFbWiHiRs1nxg,43593
 lecrapaud/integrations/openai_integration.py,sha256=hHLF3fk5Bps8KNbNrEL3NUFa945jwClE6LrLpuMZOd4,7459
 lecrapaud/jobs/__init__.py,sha256=ZkrsyTOR21c_wN7RY8jPhm8jCrL1oCEtTsf3VFIlQiE,292
@@ -37,10 +37,10 @@ lecrapaud/misc/tabpfn_tests.ipynb,sha256=VkgsCUJ30d8jaL2VaWtQAgb8ngHPNtPgnXLs7QQ
 lecrapaud/misc/test-gpu-bilstm.ipynb,sha256=4nLuZRJVe2kn6kEmauhRiz5wkWT9AVrYhI9CEk_dYUY,9608
 lecrapaud/misc/test-gpu-resnet.ipynb,sha256=27Vu7nYwujYeh3fOxBNCnKJn3MXNPKZU-U8oDDUbymg,4944
 lecrapaud/misc/test-gpu-transformers.ipynb,sha256=k6MBSs_Um1h4PykvE-LTBcdpbWLbIFST_xl_AFW2jgI,8444
-lecrapaud/model_selection.py,sha256=QOwOsn1WEBzR-2ZpHvhzv9Qz47delkBdNziHy-auY3o,72302
+lecrapaud/model_selection.py,sha256=WbFn4wiykD8DOJ_7OsZLoocp-q4GDzW0dXCf-hHhl74,72471
 lecrapaud/search_space.py,sha256=-JkzuMhaomdwiWi4HvVQY5hiw3-oREemJA16tbwEIp4,34854
 lecrapaud/utils.py,sha256=eMnNVKWTqzXCLaaxSbKLBrThkOWoJrieifr9PNqFD5Y,8375
-lecrapaud-0.18.2.dist-info/LICENSE,sha256=MImCryu0AnqhJE_uAZD-PIDKXDKb8sT7v0i1NOYeHTM,11350
-lecrapaud-0.18.2.dist-info/METADATA,sha256=6cd1hmfZhbG1E0Tu23mexomMsMwiVbCeWrRYyN9mrpM,11081
-lecrapaud-0.18.2.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
-lecrapaud-0.18.2.dist-info/RECORD,,
+lecrapaud-0.18.4.dist-info/LICENSE,sha256=MImCryu0AnqhJE_uAZD-PIDKXDKb8sT7v0i1NOYeHTM,11350
+lecrapaud-0.18.4.dist-info/METADATA,sha256=R9NN1DuZMKM9r4Z51Wy2lpaRL6ZAJiRDQ7ksgYt4hi4,11081
+lecrapaud-0.18.4.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
+lecrapaud-0.18.4.dist-info/RECORD,,