File size: 6,056 Bytes
246c042 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 | """Preprocessing pipeline builder."""
import logging
from dataclasses import dataclass
import numpy as np
import pandas as pd
from sklearn.compose import ColumnTransformer
from sklearn.impute import SimpleImputer
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import OrdinalEncoder, RobustScaler, OneHotEncoder
logger = logging.getLogger(__name__)
# Cardinality threshold: columns at or below this use OHE, above use ordinal
_OHE_MAX_CARDINALITY = 15
@dataclass
class PreprocessorResult:
"""Holds the fitted ColumnTransformer and the resulting feature names."""
pipeline: ColumnTransformer
feature_names: list[str]
def _extract_datetime_features(df: pd.DataFrame, datetime_cols: list[str]) -> pd.DataFrame:
"""Replace each datetime column with numeric calendar features.
Extracts: year, month, day, dayofweek, hour, is_weekend.
The original datetime column is dropped.
Args:
df: Input DataFrame (will be modified in-place via copy).
datetime_cols: List of datetime column names.
Returns:
DataFrame with datetime columns replaced by extracted features.
"""
df = df.copy()
for col in datetime_cols:
try:
dt = pd.to_datetime(df[col])
df[f"{col}_year"] = dt.dt.year
df[f"{col}_month"] = dt.dt.month
df[f"{col}_day"] = dt.dt.day
df[f"{col}_dayofweek"] = dt.dt.dayofweek
df[f"{col}_hour"] = dt.dt.hour
df[f"{col}_is_weekend"] = dt.dt.dayofweek.isin([5, 6]).astype(int)
df.drop(columns=[col], inplace=True)
logger.info("Extracted datetime features from column '%s'.", col)
except Exception as exc:
logger.warning("Could not extract datetime features from '%s': %s", col, exc)
return df
def build_preprocessor(
df: pd.DataFrame,
numerical_cols: list[str],
categorical_cols: list[str],
datetime_cols: list[str],
cardinality: dict[str, int],
) -> tuple[ColumnTransformer, pd.DataFrame]:
"""Build and return an unfitted ColumnTransformer plus datetime-expanded DataFrame.
Args:
df: The raw DataFrame (target column should already be removed by the caller).
numerical_cols: List of numerical feature column names.
categorical_cols: List of categorical feature column names.
datetime_cols: List of datetime column names.
cardinality: Mapping of column name → number of unique values.
Returns:
A tuple of:
- Unfitted ColumnTransformer (call .fit_transform(df_expanded) externally).
- DataFrame with datetime columns already replaced by numeric features.
"""
# Step 1 — expand datetime columns into numeric calendar features
df_expanded = _extract_datetime_features(df, datetime_cols)
# Datetime-derived columns are numeric; add them to numerical list
datetime_derived: list[str] = []
for col in datetime_cols:
for suffix in ("year", "month", "day", "dayofweek", "hour", "is_weekend"):
derived = f"{col}_{suffix}"
if derived in df_expanded.columns:
datetime_derived.append(derived)
all_numerical = numerical_cols + datetime_derived
# Step 2 — split categoricals by cardinality
low_card_cats = [c for c in categorical_cols if cardinality.get(c, 0) <= _OHE_MAX_CARDINALITY]
high_card_cats = [c for c in categorical_cols if cardinality.get(c, 0) > _OHE_MAX_CARDINALITY]
transformers = []
if all_numerical:
num_pipeline = Pipeline([
("imputer", SimpleImputer(strategy="median")),
("scaler", RobustScaler()),
])
transformers.append(("numerical", num_pipeline, all_numerical))
if low_card_cats:
low_cat_pipeline = Pipeline([
("imputer", SimpleImputer(strategy="constant", fill_value="MISSING")),
("encoder", OneHotEncoder(handle_unknown="ignore", sparse_output=False)),
])
transformers.append(("low_card_cat", low_cat_pipeline, low_card_cats))
if high_card_cats:
high_cat_pipeline = Pipeline([
("imputer", SimpleImputer(strategy="constant", fill_value="MISSING")),
(
"encoder",
OrdinalEncoder(
handle_unknown="use_encoded_value",
unknown_value=-1,
),
),
])
transformers.append(("high_card_cat", high_cat_pipeline, high_card_cats))
preprocessor = ColumnTransformer(
transformers=transformers,
remainder="drop",
verbose_feature_names_out=True,
)
logger.info(
"Preprocessor built: %d numerical, %d low-card cat (OHE), %d high-card cat (ordinal), "
"%d datetime cols expanded.",
len(all_numerical),
len(low_card_cats),
len(high_card_cats),
len(datetime_cols),
)
return preprocessor, df_expanded
def fit_preprocessor(
df: pd.DataFrame,
numerical_cols: list[str],
categorical_cols: list[str],
datetime_cols: list[str],
cardinality: dict[str, int],
) -> PreprocessorResult:
"""Build, fit, and return the preprocessor plus final feature names.
Args:
df: Feature DataFrame (target column removed).
numerical_cols: Numerical column names.
categorical_cols: Categorical column names.
datetime_cols: Datetime column names.
cardinality: Column → unique count mapping.
Returns:
PreprocessorResult with fitted pipeline and feature name list.
"""
preprocessor, df_expanded = build_preprocessor(
df, numerical_cols, categorical_cols, datetime_cols, cardinality
)
preprocessor.fit(df_expanded)
try:
feature_names = list(preprocessor.get_feature_names_out())
except Exception:
feature_names = []
logger.info("Preprocessor fitted. Output features: %d", len(feature_names))
return PreprocessorResult(pipeline=preprocessor, feature_names=feature_names)
|