lecrapaud 0.20.2__py3-none-any.whl → 0.21.1__py3-none-any.whl

lecrapaud/__init__.py

@@ -1 +1,5 @@
 from lecrapaud.api import *
+
+# Export default parameters for easy access
+from lecrapaud.api import ExperimentEngine
+DEFAULT_EXPERIMENT_PARAMS = ExperimentEngine.DEFAULT_PARAMS

lecrapaud/api.py

@@ -163,28 +163,96 @@ class ExperimentEngine:
         **kwargs: Additional configuration parameters
     """
 
+    # Default values for all experiment parameters
+    DEFAULT_PARAMS = {
+        # Feature Engineering
+        "columns_drop": [],
+        "columns_boolean": [],
+        "columns_date": [],
+        "columns_te_groupby": [],
+        "columns_te_target": [],
+        "for_training": True,
+        # Preprocessing
+        "time_series": False,
+        "val_size": 0.2,
+        "test_size": 0.2,
+        "columns_pca": [],
+        "pca_temporal": [],
+        "pca_cross_sectional": [],
+        "columns_onehot": [],
+        "columns_binary": [],
+        "columns_ordinal": [],
+        "columns_frequency": [],
+        # Feature Selection
+        "percentile": 20,
+        "corr_threshold": 80,
+        "max_features": 50,
+        "max_p_value_categorical": 0.05,
+        # Model Selection
+        "target_numbers": [],
+        "target_clf": [],
+        "models_idx": [],
+        "max_timesteps": 120,
+        "perform_hyperopt": True,
+        "number_of_trials": 20,
+        "perform_crossval": False,
+        "plot": True,
+        "preserve_model": True,
+        "target_clf_thresholds": {},
+        # Data structure
+        "date_column": None,
+        "group_column": None,
+    }
+
     def __init__(self, id: int = None, data: pd.DataFrame = None, **kwargs):
         """Initialize the experiment engine with either new or existing experiment."""
-        # Set all kwargs as instance attributes
-        if "models_idx" in kwargs:
-            kwargs["models_idx"] = normalize_models_idx(kwargs["models_idx"])
-        for key, value in kwargs.items():
-            setattr(self, key, value)
 
         if id:
+            # Load existing experiment
             self.experiment = Experiment.get(id)
-            kwargs.update(self.experiment.context)
-            experiment_dir = f"{tmp_dir}/{self.experiment.name}"
-            preprocessing_dir = f"{experiment_dir}/preprocessing"
-            data_dir = f"{experiment_dir}/data"
-            os.makedirs(preprocessing_dir, exist_ok=True)
-            os.makedirs(data_dir, exist_ok=True)
+            # Context from DB takes precedence over kwargs
+            effective_kwargs = {
+                **self.DEFAULT_PARAMS,
+                **kwargs,
+                **self.experiment.context,
+            }
         else:
             if data is None:
                 raise ValueError(
                     "Either id or data must be provided. Data can be a path to a folder containing trained models"
                 )
-            self.experiment = create_experiment(data=data, **kwargs)
+            # New experiment: merge defaults with provided kwargs
+            effective_kwargs = {**self.DEFAULT_PARAMS, **kwargs}
+
+        # Normalize models_idx if present
+        if "models_idx" in effective_kwargs:
+            effective_kwargs["models_idx"] = normalize_models_idx(
+                effective_kwargs["models_idx"]
+            )
+
+        # Set all parameters as instance attributes
+        for key, value in effective_kwargs.items():
+            setattr(self, key, value)
+
+        # Create experiment if new
+        if not id:
+            self.experiment = create_experiment(data=data, **effective_kwargs)
+
+        # Create directories
+        experiment_dir = f"{tmp_dir}/{self.experiment.name}"
+        preprocessing_dir = f"{experiment_dir}/preprocessing"
+        data_dir = f"{experiment_dir}/data"
+        os.makedirs(preprocessing_dir, exist_ok=True)
+        os.makedirs(data_dir, exist_ok=True)
+
+    @classmethod
+    def get_default_params(cls):
+        """Get the default parameters for experiments."""
+        return cls.DEFAULT_PARAMS.copy()
+
+    def get_effective_context(self):
+        """Get the effective context (merged defaults + experiment context)."""
+        return {k: getattr(self, k, v) for k, v in self.DEFAULT_PARAMS.items()}
 
     def train(self, data, best_params=None):
         logger.info("Running training...")

lecrapaud/db/alembic/versions/2025_10_28_2006-0a8fb7826e9b_add_number_of_targets_and_remove_other_.py

@@ -5,38 +5,71 @@ Revises: 033e0f7eca4f
 Create Date: 2025-10-28 20:06:54.792631
 
 """
+
 from typing import Sequence, Union
 
 from alembic import op
 import sqlalchemy as sa
 from sqlalchemy.dialects import mysql
+from lecrapaud.config import LECRAPAUD_TABLE_PREFIX
 
 # revision identifiers, used by Alembic.
-revision: str = '0a8fb7826e9b'
-down_revision: Union[str, None] = '033e0f7eca4f'
+revision: str = "0a8fb7826e9b"
+down_revision: Union[str, None] = "033e0f7eca4f"
 branch_labels: Union[str, Sequence[str], None] = None
 depends_on: Union[str, Sequence[str], None] = None
 
 
 def upgrade() -> None:
     # ### commands auto generated by Alembic - please adjust! ###
-    op.add_column('lecrapaud_experiments', sa.Column('number_of_targets', sa.Integer(), nullable=True))
-    op.drop_column('lecrapaud_experiments', 'corr_threshold')
-    op.drop_column('lecrapaud_experiments', 'max_features')
-    op.drop_column('lecrapaud_experiments', 'percentile')
-    op.drop_column('lecrapaud_experiments', 'type')
-    op.drop_index(op.f('ix_model_selection_scores_id'), table_name='lecrapaud_model_selection_scores')
-    op.create_index(op.f('ix_lecrapaud_model_selection_scores_id'), 'lecrapaud_model_selection_scores', ['id'], unique=False)
+    op.add_column(
+        f"{LECRAPAUD_TABLE_PREFIX}_experiments",
+        sa.Column("number_of_targets", sa.Integer(), nullable=True),
+    )
+    op.drop_column(f"{LECRAPAUD_TABLE_PREFIX}_experiments", "corr_threshold")
+    op.drop_column(f"{LECRAPAUD_TABLE_PREFIX}_experiments", "max_features")
+    op.drop_column(f"{LECRAPAUD_TABLE_PREFIX}_experiments", "percentile")
+    op.drop_column(f"{LECRAPAUD_TABLE_PREFIX}_experiments", "type")
+    op.drop_index(
+        op.f("ix_model_selection_scores_id"),
+        table_name=f"{LECRAPAUD_TABLE_PREFIX}_model_selection_scores",
+    )
+    op.create_index(
+        op.f("ix_model_selection_scores_id"),
+        f"{LECRAPAUD_TABLE_PREFIX}_model_selection_scores",
+        ["id"],
+        unique=False,
+    )
     # ### end Alembic commands ###
 
 
 def downgrade() -> None:
     # ### commands auto generated by Alembic - please adjust! ###
-    op.drop_index(op.f('ix_lecrapaud_model_selection_scores_id'), table_name='lecrapaud_model_selection_scores')
-    op.create_index(op.f('ix_model_selection_scores_id'), 'lecrapaud_model_selection_scores', ['id'], unique=False)
-    op.add_column('lecrapaud_experiments', sa.Column('type', mysql.VARCHAR(length=50), nullable=False))
-    op.add_column('lecrapaud_experiments', sa.Column('percentile', mysql.FLOAT(), nullable=False))
-    op.add_column('lecrapaud_experiments', sa.Column('max_features', mysql.INTEGER(), autoincrement=False, nullable=False))
-    op.add_column('lecrapaud_experiments', sa.Column('corr_threshold', mysql.FLOAT(), nullable=False))
-    op.drop_column('lecrapaud_experiments', 'number_of_targets')
+    op.drop_index(
+        op.f("ix_lecrapaud_model_selection_scores_id"),
+        table_name=f"{LECRAPAUD_TABLE_PREFIX}_model_selection_scores",
+    )
+    op.create_index(
+        op.f("ix_model_selection_scores_id"),
+        f"{LECRAPAUD_TABLE_PREFIX}_model_selection_scores",
+        ["id"],
+        unique=False,
+    )
+    op.add_column(
+        f"{LECRAPAUD_TABLE_PREFIX}_experiments",
+        sa.Column("type", mysql.VARCHAR(length=50), nullable=False),
+    )
+    op.add_column(
+        f"{LECRAPAUD_TABLE_PREFIX}_experiments",
+        sa.Column("percentile", mysql.FLOAT(), nullable=False),
+    )
+    op.add_column(
+        f"{LECRAPAUD_TABLE_PREFIX}_experiments",
+        sa.Column("max_features", mysql.INTEGER(), autoincrement=False, nullable=False),
+    )
+    op.add_column(
+        f"{LECRAPAUD_TABLE_PREFIX}_experiments",
+        sa.Column("corr_threshold", mysql.FLOAT(), nullable=False),
+    )
+    op.drop_column(f"{LECRAPAUD_TABLE_PREFIX}_experiments", "number_of_targets")
     # ### end Alembic commands ###

lecrapaud/experiment.py

@@ -28,6 +28,9 @@ def create_experiment(
     if kwargs.get("time_series") and not date_column:
         raise ValueError("date_column must be provided for time series experiments")
 
+    if experiment_name is None:
+        raise ValueError("experiment_name must be provided")
+
     dates = {}
     if date_column:
         dates["start_date"] = pd.to_datetime(data[date_column].iat[0])

lecrapaud/feature_engineering.py

@@ -94,7 +94,7 @@ class FeatureEngineeringEngine:
         self.data = data
         self.experiment = experiment
         self.for_training = for_training
-        
+
         # Get all parameters from experiment context
         self.columns_drop = self.experiment.context.get("columns_drop", [])
         self.columns_boolean = self.experiment.context.get("columns_boolean", [])
@@ -330,15 +330,19 @@ class PreprocessFeature:
         self.test_size = context.get("test_size", 0.2)
         self.target_numbers = context.get("target_numbers", [])
         self.target_clf = context.get("target_clf", [])
-        
+
         # Handle list parameters with uppercase conversion
         self.columns_pca = [col.upper() for col in context.get("columns_pca", [])]
         self.pca_temporal = context.get("pca_temporal", [])
         self.pca_cross_sectional = context.get("pca_cross_sectional", [])
         self.columns_onehot = [col.upper() for col in context.get("columns_onehot", [])]
         self.columns_binary = [col.upper() for col in context.get("columns_binary", [])]
-        self.columns_ordinal = [col.upper() for col in context.get("columns_ordinal", [])]
-        self.columns_frequency = [col.upper() for col in context.get("columns_frequency", [])]
+        self.columns_ordinal = [
+            col.upper() for col in context.get("columns_ordinal", [])
+        ]
+        self.columns_frequency = [
+            col.upper() for col in context.get("columns_frequency", [])
+        ]
 
         self.experiment_dir = self.experiment.path
         self.experiment_id = self.experiment.id
@@ -601,7 +605,7 @@ class PreprocessFeature:
 
         return df, pcas_dict
 
-    def add_pca_feature_cross_sectional(
+    def add_pca_feature_cross_sectional_old(
         self,
         df: pd.DataFrame,
         *,
@@ -653,8 +657,228 @@ class PreprocessFeature:
 
         return df, pcas_dict
 
+    def add_pca_feature_cross_sectional(
+        self,
+        df: pd.DataFrame,
+        *,
+        n_components: int = 5,
+        pcas: dict[str, Pipeline] | None = None,  # si fourni: transform only
+        impute_strategy: str = "median",
+        standardize: bool = True,
+        lookback_days: int = 365,  # nombre de jours à regarder en arrière pour le fit
+        refresh_frequency: int = 90,  # refresh la PCA tous les X jours
+    ) -> tuple[pd.DataFrame, dict[str, Pipeline]]:
+        """
+        Construit un pivot (index=index_col, columns=columns_col, values=value_col),
+        fit (ou réutilise) un Pipeline Imputer(+Scaler)+PCA, puis merge les scores
+        (par index_col) dans df. Renvoie (df_avec_features, pipe).
+
+        Pour les séries temporelles : fit la PCA uniquement sur les données passées
+        pour éviter le leakage, avec refresh périodique.
+
+        Gère le cas des données panel où on a plusieurs séries temporelles
+        (ex: plusieurs stocks avec les mêmes dates).
+        """
+
+        pcas_dict = {}
+        index_saved = df.index
+
+        for pca_cross_sectional in self.pca_cross_sectional:
+            name, index_col, columns_col, value_col = (
+                pca_cross_sectional[k] for k in ("name", "index", "columns", "value")
+            )
+            prefix = f"CS_PC_{name}"
+
+            # Vérifier si c'est une série temporelle avec index = date
+            # Les dates sont déjà en ordinal après cyclic_encode_date
+            is_time_series = self.time_series and index_col == self.date_column
+
+            if is_time_series:
+                # Cas spécial : PCA cross-sectional sur des données de panel time series
+                # Par exemple : PCA sur les returns de tous les stocks à chaque date
+                # pour capturer le régime de marché
+
+                all_scores = []
+
+                # Les dates sont déjà en ordinal
+                unique_dates = sorted(df[index_col].unique())
+
+                # Pour l'inference, utiliser la PCA fournie
+                if pcas is not None:
+                    pipe = pcas[name]
+                    pivot = df.pivot_table(
+                        index=index_col, columns=columns_col, values=value_col
+                    ).sort_index()
+                    scores = pipe.transform(pivot)
+                    cols = [f"{prefix}_{i}" for i in range(n_components)]
+                    scores_df = pd.DataFrame(scores, index=pivot.index, columns=cols)
+                else:
+                    # Training : fit PCA de manière expanding avec refresh périodique
+                    pipe = None
+                    last_fit_date = None
+
+                    for i, current_date_ordinal in enumerate(unique_dates):
+                        # Convertir l'ordinal en date pour les calculs de temps
+                        current_date = pd.Timestamp.fromordinal(
+                            int(current_date_ordinal)
+                        )
+
+                        # Déterminer si on doit refitter la PCA
+                        should_refit = pipe is None or (  # Première fois
+                            last_fit_date is not None
+                            and (current_date - last_fit_date).days >= refresh_frequency
+                        )
+
+                        if (
+                            should_refit and i > 30
+                        ):  # Attendre au moins 30 jours de données
+                            # Prendre les données des 'lookback_days' derniers jours
+                            lookback_start_date = current_date - pd.Timedelta(
+                                days=lookback_days
+                            )
+                            lookback_start_ordinal = pd.Timestamp.toordinal(
+                                lookback_start_date
+                            )
+
+                            # Masque pour les dates passées uniquement (éviter le leakage)
+                            mask_fit = (df[index_col] >= lookback_start_ordinal) & (
+                                df[index_col] < current_date_ordinal
+                            )
+                            df_fit = df[mask_fit]
+
+                            if len(df_fit) > 0:
+                                # Créer le pivot pour la période de lookback
+                                pivot_fit = df_fit.pivot_table(
+                                    index=index_col,
+                                    columns=columns_col,
+                                    values=value_col,
+                                ).sort_index()
+
+                                # Vérifier qu'on a assez de dates et de colonnes
+                                if (
+                                    len(pivot_fit) >= n_components
+                                    and pivot_fit.shape[1] >= n_components
+                                ):
+                                    # Créer nouveau pipeline
+                                    steps = [
+                                        (
+                                            "imputer",
+                                            SimpleImputer(strategy=impute_strategy),
+                                        )
+                                    ]
+                                    if standardize:
+                                        steps.append(
+                                            (
+                                                "scaler",
+                                                StandardScaler(
+                                                    with_mean=True, with_std=True
+                                                ),
+                                            )
+                                        )
+                                    pca = PCA(n_components=n_components, random_state=0)
+                                    steps.append(("pca", pca))
+                                    pipe = Pipeline(steps)
+                                    pipe.fit(pivot_fit)
+                                    last_fit_date = current_date
+
+                                    logger.debug(
+                                        f"PCA {name} refitted at date {current_date.strftime('%Y-%m-%d')} "
+                                        f"using {len(pivot_fit)} dates and {pivot_fit.shape[1]} columns"
+                                    )
+
+                        # Transform pour la date courante uniquement
+                        if pipe is not None:
+                            df_current = df[df[index_col] == current_date_ordinal]
+                            if len(df_current) > 0:
+                                pivot_current = df_current.pivot_table(
+                                    index=index_col,
+                                    columns=columns_col,
+                                    values=value_col,
+                                )
+                                try:
+                                    scores_current = pipe.transform(pivot_current)
+                                    scores_dict = {
+                                        index_col: [current_date_ordinal],
+                                        **{
+                                            f"{prefix}_{j}": [scores_current[0, j]]
+                                            for j in range(n_components)
+                                        },
+                                    }
+                                    all_scores.append(pd.DataFrame(scores_dict))
+                                except Exception as e:
+                                    # En cas d'erreur (ex: nouvelles colonnes), créer des valeurs manquantes
+                                    logger.debug(
+                                        f"PCA transform error at date {current_date}: {str(e)}"
+                                    )
+                                    scores_dict = {
+                                        index_col: [current_date_ordinal],
+                                        **{
+                                            f"{prefix}_{j}": [np.nan]
+                                            for j in range(n_components)
+                                        },
+                                    }
+                                    all_scores.append(pd.DataFrame(scores_dict))
+                        else:
+                            # Pas encore de PCA fittée, créer des NaN
+                            scores_dict = {
+                                index_col: [current_date_ordinal],
+                                **{
+                                    f"{prefix}_{j}": [np.nan]
+                                    for j in range(n_components)
+                                },
+                            }
+                            all_scores.append(pd.DataFrame(scores_dict))
+
+                    # Combiner tous les scores
+                    if all_scores:
+                        scores_df = pd.concat(all_scores, ignore_index=True)
+                    else:
+                        # Créer un DataFrame vide avec les bonnes colonnes
+                        cols = [f"{prefix}_{i}" for i in range(n_components)]
+                        scores_df = pd.DataFrame(columns=[index_col] + cols)
+
+                # Merger les scores
+                df = df.merge(scores_df, on=index_col, how="left")
+                df.index = index_saved
+                
+                # Forward fill puis 0 pour éviter les NaN
+                pca_cols = [col for col in df.columns if col.startswith(prefix)]
+                df[pca_cols] = df[pca_cols].fillna(method='ffill').fillna(0)
+                
+                pcas_dict.update({name: pipe})
+
+            else:
+                # Approche classique (non time series ou index != date)
+                pivot = df.pivot_table(
+                    index=index_col, columns=columns_col, values=value_col
+                ).sort_index()
+
+                # Pipeline à réutiliser entre train et test
+                if pcas is None:
+                    steps = [("imputer", SimpleImputer(strategy=impute_strategy))]
+                    if standardize:
+                        steps.append(
+                            ("scaler", StandardScaler(with_mean=True, with_std=True))
+                        )
+                    pca = PCA(n_components=n_components, random_state=0)
+                    steps.append(("pca", pca))
+                    pipe = Pipeline(steps)
+                    pipe.fit(pivot)  # <- fit sur TRAIN uniquement
+                else:
+                    pipe = pcas[name]  # <- TEST : on réutilise le pipe existant
+
+                scores = pipe.transform(pivot)  # shape: (n_index, n_components)
+                cols = [f"{prefix}_{i}" for i in range(n_components)]
+                scores_df = pd.DataFrame(scores, index=pivot.index, columns=cols)
+
+                df = df.merge(scores_df.reset_index(), on=index_col, how="left")
+                df.index = index_saved
+                pcas_dict.update({name: pipe})
+
+        return df, pcas_dict
+
     # ----------------- 2) PCA TEMPORELLE (liste de colonnes lags) ----------------
-    def add_pca_feature_temporal(
+    def add_pca_feature_temporal_old(
         self,
         df: pd.DataFrame,
         *,
@@ -717,6 +941,187 @@ class PreprocessFeature:
 
         return df, pcas_dict
 
+    def add_pca_feature_temporal(
+        self,
+        df: pd.DataFrame,
+        *,
+        n_components: int = 5,
+        pcas: dict[str, Pipeline] | None = None,
+        impute_strategy: str = "median",
+        standardize: bool = True,
+        lookback_days: int = 365,
+        refresh_frequency: int = 90,
+    ) -> tuple[pd.DataFrame, dict[str, Pipeline]]:
+        """
+        PCA temporelle pour time series avec support panel data.
+        Crée automatiquement les colonnes de lags et évite le look-ahead bias.
+        
+        Format pca_temporal simplifié:
+        [{"name": "LAST_20_RET", "column": "RET", "lags": 20}]
+        """
+        pcas_dict = {}
+        
+        for pca_config in self.pca_temporal:
+            # Support both old and new format
+            if "columns" in pca_config:
+                # Old format: use existing columns
+                name = pca_config["name"]
+                lag_columns = pca_config["columns"]
+                base_column = None
+                num_lags = len(lag_columns)
+            else:
+                # New format: create lag columns
+                name = pca_config["name"]
+                base_column = pca_config["column"].upper()
+                num_lags = pca_config.get("lags", 20)
+                
+                # Create lag columns if they don't exist
+                if self.group_column:
+                    # Panel data: create lags by group
+                    for lag in range(1, num_lags + 1):
+                        lag_col = f"{base_column}_-{lag}"
+                        if lag_col not in df.columns:
+                            df[lag_col] = df.groupby(self.group_column)[base_column].shift(lag)
+                else:
+                    # Simple time series
+                    for lag in range(1, num_lags + 1):
+                        lag_col = f"{base_column}_-{lag}"
+                        if lag_col not in df.columns:
+                            df[lag_col] = df[base_column].shift(lag)
+                
+                lag_columns = [f"{base_column}_-{i}" for i in range(1, num_lags + 1)]
+            
+            prefix = f"TMP_PC_{name}"
+            
+            # For time series: avoid look-ahead bias
+            if self.time_series and self.date_column:
+                all_scores = []
+                unique_dates = sorted(df[self.date_column].unique())
+                
+                if pcas is not None:
+                    # Inference: use provided PCA
+                    pipe = pcas[name]
+                    
+                    # Apply to all data at once
+                    mask = df[lag_columns].notna().all(axis=1)
+                    if mask.any():
+                        X_transform = df.loc[mask, lag_columns]
+                        scores = pipe.transform(X_transform)
+                        
+                        for i in range(n_components):
+                            df.loc[mask, f"{prefix}_{i}"] = scores[:, i]
+                    
+                    # Fill NaN with forward fill then 0
+                    pca_cols = [f"{prefix}_{i}" for i in range(n_components)]
+                    df[pca_cols] = df[pca_cols].fillna(method='ffill').fillna(0)
+                    
+                else:
+                    # Training: expanding window with periodic refresh
+                    pipe = None
+                    last_fit_date = None
+                    
+                    for current_date_ordinal in unique_dates:
+                        current_date = pd.Timestamp.fromordinal(int(current_date_ordinal))
+                        
+                        # Determine if we should refit
+                        should_refit = pipe is None or (
+                            last_fit_date is not None 
+                            and (current_date - last_fit_date).days >= refresh_frequency
+                        )
+                        
+                        if should_refit and len(df[df[self.date_column] < current_date_ordinal]) > num_lags * 2:
+                            # Get historical data for fitting
+                            lookback_start = current_date - pd.Timedelta(days=lookback_days)
+                            lookback_start_ordinal = pd.Timestamp.toordinal(lookback_start)
+                            
+                            mask_fit = (
+                                (df[self.date_column] >= lookback_start_ordinal) & 
+                                (df[self.date_column] < current_date_ordinal) &
+                                df[lag_columns].notna().all(axis=1)
+                            )
+                            
+                            if mask_fit.sum() >= n_components:
+                                X_fit = df.loc[mask_fit, lag_columns]
+                                
+                                # Create pipeline
+                                steps = []
+                                if impute_strategy is not None:
+                                    steps.append(("imputer", SimpleImputer(strategy=impute_strategy)))
+                                if standardize:
+                                    steps.append(("scaler", StandardScaler()))
+                                steps.append(("pca", PCA(n_components=n_components, random_state=0)))
+                                
+                                pipe = Pipeline(steps)
+                                pipe.fit(X_fit)
+                                last_fit_date = current_date
+                                
+                                logger.debug(
+                                    f"Temporal PCA {name} refitted at {current_date.strftime('%Y-%m-%d')} "
+                                    f"using {len(X_fit)} samples"
+                                )
+                        
+                        # Transform current date data
+                        if pipe is not None:
+                            mask_current = (
+                                (df[self.date_column] == current_date_ordinal) &
+                                df[lag_columns].notna().all(axis=1)
+                            )
+                            
+                            if mask_current.any():
+                                X_current = df.loc[mask_current, lag_columns]
+                                scores = pipe.transform(X_current)
+                                
+                                for i in range(n_components):
+                                    df.loc[mask_current, f"{prefix}_{i}"] = scores[:, i]
+                    
+                    # Fill NaN with forward fill then 0
+                    pca_cols = [f"{prefix}_{i}" for i in range(n_components)]
+                    for col in pca_cols:
+                        if col not in df.columns:
+                            df[col] = 0
+                    df[pca_cols] = df[pca_cols].fillna(method='ffill').fillna(0)
+                
+                pcas_dict[name] = pipe
+                
+            else:
+                # Non time-series: use original approach
+                mask = df[lag_columns].notna().all(axis=1)
+                
+                if pcas is None and mask.any():
+                    X_fit = df.loc[mask, lag_columns]
+                    
+                    steps = []
+                    if impute_strategy is not None:
+                        steps.append(("imputer", SimpleImputer(strategy=impute_strategy)))
+                    if standardize:
+                        steps.append(("scaler", StandardScaler()))
+                    steps.append(("pca", PCA(n_components=n_components, random_state=0)))
+                    
+                    pipe = Pipeline(steps)
+                    pipe.fit(X_fit)
+                    pcas_dict[name] = pipe
+                elif pcas is not None:
+                    pipe = pcas[name]
+                    pcas_dict[name] = pipe
+                else:
+                    continue
+                
+                if mask.any():
+                    X_transform = df.loc[mask, lag_columns]
+                    scores = pipe.transform(X_transform)
+                    
+                    for i in range(n_components):
+                        df.loc[mask, f"{prefix}_{i}"] = scores[:, i]
+                
+                # Fill missing values
+                pca_cols = [f"{prefix}_{i}" for i in range(n_components)]
+                for col in pca_cols:
+                    if col not in df.columns:
+                        df[col] = 0
+                df[pca_cols] = df[pca_cols].fillna(0)
+        
+        return df, pcas_dict
+
     # encoding categorical features
     def encode_categorical_features(
         self,

{lecrapaud-0.20.2.dist-info → lecrapaud-0.21.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lecrapaud
-Version: 0.20.2
+Version: 0.21.1
 Summary: Framework for machine and deep learning, with regression, classification and time series analysis
 License: Apache License
 License-File: LICENSE
@@ -12,9 +12,12 @@ Classifier: Programming Language :: Python :: 3.12
 Requires-Dist: catboost (>=1.2.8)
 Requires-Dist: category-encoders (>=2.8.1)
 Requires-Dist: celery (>=5.5.3)
+Requires-Dist: celery-redbeat (>=2.3.2)
 Requires-Dist: ftfy (>=6.3.1)
+Requires-Dist: hyperopt (>=0.2.7)
 Requires-Dist: joblib (>=1.5.1)
 Requires-Dist: keras (>=3.10.0)
+Requires-Dist: keras-tcn (>=3.5.6)
 Requires-Dist: lightgbm (>=4.6.0)
 Requires-Dist: matplotlib (>=3.10.3)
 Requires-Dist: mlxtend (>=0.23.4)

{lecrapaud-0.20.2.dist-info → lecrapaud-0.21.1.dist-info}/RECORD

@@ -1,5 +1,5 @@
-lecrapaud/__init__.py,sha256=oCxbtw_nk8rlOXbXbWo0RRMlsh6w-hTiZ6e5PRG_wp0,28
-lecrapaud/api.py,sha256=IQlH3wcSzxYgvlamfICNMwNsQGoaNxBJUPTlC9M0kBk,20321
+lecrapaud/__init__.py,sha256=7Wp_VF08UZP8o-GkpB4_yRjP4twQmpcTc3202OkPmHs,176
+lecrapaud/api.py,sha256=7OL_wbg9hCmlZ0WI6eCDkublntES3f320OZlpuKu8f4,22376
 lecrapaud/config.py,sha256=0NEg61QdLxQ97bVFDDXa6OwlWFEo_z8VIhX5KrD1ik0,1170
 lecrapaud/db/__init__.py,sha256=82o9fMfaqKXPh2_rt44EzNRVZV1R4LScEnQYvj_TjK0,34
 lecrapaud/db/alembic/README,sha256=MVlc9TYmr57RbhXET6QxgyCcwWP7w-vLkEsirENqiIQ,38
@@ -14,7 +14,7 @@ lecrapaud/db/alembic/versions/2025_08_28_1516-c36e9fee22b9_add_avg_precision_to_
 lecrapaud/db/alembic/versions/2025_08_28_1622-8b11c1ba982e_change_name_column.py,sha256=g6H2Z9MwB6UEiqdGlBoHBXpO9DTaWkwHt8FS6joVOm0,1191
 lecrapaud/db/alembic/versions/2025_10_25_0635-07e303521594_add_unique_constraint_to_score.py,sha256=FshOF1t-NWXrBtXT3wMNGFslJ4sWUxzvBEXSymu05cI,1043
 lecrapaud/db/alembic/versions/2025_10_26_1727-033e0f7eca4f_merge_score_and_model_trainings_into_.py,sha256=htHUD4zPJr-0z_DQfTi8r9RsFVe9m7SL0f7oRIvLIcQ,10999
-lecrapaud/db/alembic/versions/2025_10_28_2006-0a8fb7826e9b_add_number_of_targets_and_remove_other_.py,sha256=o3TNHq1GTFjxfk2zHWaUbq91khMJi6Xy6HToO9i54AU,2051
+lecrapaud/db/alembic/versions/2025_10_28_2006-0a8fb7826e9b_add_number_of_targets_and_remove_other_.py,sha256=0NBvOwPqMXpWnDEGiEBk_IeLKmXQ5ZcU-dqHeSEgsRQ,2557
 lecrapaud/db/alembic.ini,sha256=Zw2rdwsKV6c7J1SPtoFIPDX08_oTP3MuUKnNxBDiY8I,3796
 lecrapaud/db/models/__init__.py,sha256=-XoCN1eeLihnNxBMl90lXrgrTSDkMbeqgienMqFi5f4,702
 lecrapaud/db/models/base.py,sha256=0548x4ftd6Oim9BJmtD7Er4izM6u0QCrlTG5560384w,9458
@@ -29,8 +29,8 @@ lecrapaud/db/models/target.py,sha256=DKnfeaLU8eT8J_oh_vuFo5-o1CaoXR13xBbswme6Bgk
 lecrapaud/db/models/utils.py,sha256=-a-nWWmpJ2XzidIxo2COVUTrGZIPYCfBzjhcszJj_bM,1109
 lecrapaud/db/session.py,sha256=u9NCwUoV5VbtScRb6HOSQr4oTEjIwj0waP5mGlc1qJg,3735
 lecrapaud/directories.py,sha256=0LrANuDgbuneSLker60c6q2hmGnQ3mKHIztTGzTx6Gw,826
-lecrapaud/experiment.py,sha256=hhi6NdVKtxoyx_AGBB4iNEZZpd9b3rKs23qiLPf-mUk,2384
-lecrapaud/feature_engineering.py,sha256=UM-EIOsgYWedqsR9uA-09eaWSb9FofVxoE0rRcDelQ8,39173
+lecrapaud/experiment.py,sha256=LiecZS3P4igO_3nJ4IB-2b25CttQS2RePDnhBNucvdE,2478
+lecrapaud/feature_engineering.py,sha256=SvGrJXv24rVgH0QE5mRwJITcCLfUqgbV2Ep68bBVnJs,58794
 lecrapaud/feature_selection.py,sha256=Q9xWVmZsvRjX9mJHB_PY_KLXsEAYNLX7txSe0cniY4A,47529
 lecrapaud/integrations/openai_integration.py,sha256=hHLF3fk5Bps8KNbNrEL3NUFa945jwClE6LrLpuMZOd4,7459
 lecrapaud/jobs/__init__.py,sha256=ZkrsyTOR21c_wN7RY8jPhm8jCrL1oCEtTsf3VFIlQiE,292
@@ -44,7 +44,7 @@ lecrapaud/misc/test-gpu-transformers.ipynb,sha256=k6MBSs_Um1h4PykvE-LTBcdpbWLbIF
 lecrapaud/model_selection.py,sha256=o4_hOEp91_33HtMatVHU7YPc71KZ2hK7wucN63xqWkA,88017
 lecrapaud/search_space.py,sha256=caCehJklD3-sgmlisJj_GmuB7LJiVvTF71gEjPGDvV4,36336
 lecrapaud/utils.py,sha256=0k76HFETO0_NgCYUv8b3RTBLgry6MsDBaHJfpAplxCY,8855
-lecrapaud-0.20.2.dist-info/METADATA,sha256=FUXEVYVCJAoat8HUtsupISlRbK56YVxezYwCH6j4kBE,14239
-lecrapaud-0.20.2.dist-info/WHEEL,sha256=zp0Cn7JsFoX2ATtOhtaFYIiE2rmFAD4OcMhtUki8W3U,88
-lecrapaud-0.20.2.dist-info/licenses/LICENSE,sha256=MImCryu0AnqhJE_uAZD-PIDKXDKb8sT7v0i1NOYeHTM,11350
-lecrapaud-0.20.2.dist-info/RECORD,,
+lecrapaud-0.21.1.dist-info/METADATA,sha256=rKls8xvjhu9f72jTw2sjBYCmQPw-N02RSScSOjJ1E2g,14348
+lecrapaud-0.21.1.dist-info/WHEEL,sha256=zp0Cn7JsFoX2ATtOhtaFYIiE2rmFAD4OcMhtUki8W3U,88
+lecrapaud-0.21.1.dist-info/licenses/LICENSE,sha256=MImCryu0AnqhJE_uAZD-PIDKXDKb8sT7v0i1NOYeHTM,11350
+lecrapaud-0.21.1.dist-info/RECORD,,