Spaces:
Sleeping
Sleeping
File size: 10,967 Bytes
575d22a |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 |
"""Centralized preprocessing pipeline for hotel cancellation prediction.
Provides a reusable class that encapsulates:
- Categorical handling strategy (currently: drop)
- Numeric scaling (StandardScaler)
- Feature ordering preservation
- Artifact persistence / loading
Future extension points:
- onehot / target / hybrid categorical strategies
- numeric imputation strategies
- feature selection masks
Usage:
pipeline = PreprocessingPipeline(categorical_strategy='drop', scale=True)
X_train_proc = pipeline.fit_transform(X_train)
X_test_proc = pipeline.transform(X_test)
pipeline.save('models/preprocessor.pkl')
# Later / inference
pipeline = PreprocessingPipeline.load('models/preprocessor.pkl')
X_new = pipeline.transform(X_incoming)
"""
from __future__ import annotations
from dataclasses import dataclass, asdict
from typing import List, Optional, Dict, Any, Tuple
import pandas as pd
import joblib
from sklearn.preprocessing import StandardScaler
import os
import numpy as np
@dataclass
class PreprocessingState:
categorical_strategy: str
scaled_numeric: List[str]
dropped_columns: List[str]
feature_order: List[str]
scale: bool
# One-hot specific
onehot_categories: Optional[Dict[str, List[str]]] = None
# Target encoding specific
target_mappings: Optional[Dict[str, Dict[str, float]]] = None
target_global_mean: Optional[float] = None
target_encoded_columns: Optional[List[str]] = None
class PreprocessingPipeline:
def __init__(self, categorical_strategy: str = 'drop', scale: bool = True, target_min_samples: int = 5, target_smoothing: float = 10.0):
self.categorical_strategy = categorical_strategy
self.scale = scale
self._scaler: Optional[StandardScaler] = None
self.state: Optional[PreprocessingState] = None
# target encoding hyperparams
self.target_min_samples = target_min_samples
self.target_smoothing = target_smoothing
def _apply_onehot_fit(self, X: pd.DataFrame) -> Tuple[pd.DataFrame, Dict[str, List[str]]]:
cat_cols = [c for c in X.columns if X[c].dtype == 'object' or pd.api.types.is_categorical_dtype(X[c])]
categories: Dict[str, List[str]] = {}
transformed_parts = [X[[c]] for c in X.columns if c not in cat_cols]
for c in cat_cols:
cats = sorted([str(v) for v in X[c].dropna().unique()])
categories[c] = cats
for val in cats:
col_name = f"{c}__{val}"
transformed_parts.append((X[c].astype(str) == val).astype(int).to_frame(col_name))
X_new = pd.concat(transformed_parts, axis=1)
return X_new, categories
def _apply_onehot_transform(self, X: pd.DataFrame) -> pd.DataFrame:
assert self.state and self.state.onehot_categories
cat_schema = self.state.onehot_categories
out_parts = []
# Numeric / other passthrough first (original columns that were not categorical at fit time)
for c in self.state.feature_order:
# original feature_order contains post-onehot columns already; skip here
pass
# Reconstruct expected columns deterministically
for base_col, cats in cat_schema.items():
series = X[base_col].astype(str) if base_col in X.columns else pd.Series([None]*len(X), index=X.index)
for val in cats:
col_name = f"{base_col}__{val}"
out_parts.append((series == val).astype(int).rename(col_name))
# Add any numeric columns (those not in cat_schema keys)
numeric_like = [c for c in X.columns if c not in cat_schema]
for c in numeric_like:
if c not in self.state.feature_order and any(c.startswith(f"{k}__") for k in cat_schema):
# skip inadvertent collision
continue
if c in cat_schema:
continue
if pd.api.types.is_numeric_dtype(X[c]):
out_parts.append(X[c])
X_new = pd.concat(out_parts, axis=1)
# Align to stored feature order
missing = [c for c in self.state.feature_order if c not in X_new.columns]
for m in missing:
X_new[m] = 0 # unseen category -> all zeros
X_new = X_new[self.state.feature_order]
return X_new
def _compute_target_encoding(self, X: pd.DataFrame, y: pd.Series) -> Tuple[pd.DataFrame, Dict[str, Dict[str,float]], float, List[str]]:
cat_cols = [c for c in X.columns if X[c].dtype == 'object' or pd.api.types.is_categorical_dtype(X[c])]
mappings: Dict[str, Dict[str, float]] = {}
global_mean = float(y.mean())
X_encoded = X.copy()
encoded_cols: List[str] = []
for c in cat_cols:
stats = y.groupby(X[c]).agg(['mean','count'])
# smoothing: (count*mean + smoothing*global) / (count + smoothing)
smooth = (stats['count'] * stats['mean'] + self.target_smoothing * global_mean) / (stats['count'] + self.target_smoothing)
mapping = smooth.to_dict()
mappings[c] = mapping
new_col = f"{c}__te"
encoded_cols.append(new_col)
X_encoded[new_col] = X[c].map(mapping).fillna(global_mean)
# Drop original categorical columns
X_encoded = X_encoded.drop(columns=cat_cols)
return X_encoded, mappings, global_mean, encoded_cols
def _apply_target_transform(self, X: pd.DataFrame) -> pd.DataFrame:
assert self.state and self.state.target_mappings is not None
global_mean = self.state.target_global_mean
X_new = X.copy()
# For each mapping, create encoded column
for col, mapping in self.state.target_mappings.items():
new_col = f"{col}__te"
series = X_new[col] if col in X_new.columns else pd.Series([None]*len(X_new), index=X_new.index)
X_new[new_col] = series.map(mapping).fillna(global_mean)
# Drop raw categorical cols
X_new = X_new.drop(columns=list(self.state.target_mappings.keys()))
# Align order / add any missing
missing = [c for c in self.state.feature_order if c not in X_new.columns]
for m in missing:
X_new[m] = 0.0
X_new = X_new[self.state.feature_order]
return X_new
def fit(self, X: pd.DataFrame, y: Optional[pd.Series] = None) -> 'PreprocessingPipeline':
X = X.copy()
dropped: List[str] = []
onehot_categories: Optional[Dict[str, List[str]]] = None
target_mappings: Optional[Dict[str, Dict[str, float]]] = None
target_global_mean: Optional[float] = None
target_encoded_cols: Optional[List[str]] = None
if self.categorical_strategy == 'drop':
non_numeric = [c for c in X.columns if not pd.api.types.is_numeric_dtype(X[c])]
if non_numeric:
X = X.drop(columns=non_numeric)
dropped = non_numeric
elif self.categorical_strategy == 'onehot':
X, onehot_categories = self._apply_onehot_fit(X)
elif self.categorical_strategy == 'target':
if y is None:
raise ValueError("Target series y must be provided for target encoding strategy.")
X, target_mappings, target_global_mean, target_encoded_cols = self._compute_target_encoding(X, y)
else:
raise NotImplementedError(f"Categorical strategy '{self.categorical_strategy}' not implemented.")
numeric_cols = [c for c in X.columns if pd.api.types.is_numeric_dtype(X[c])]
if self.scale and numeric_cols:
self._scaler = StandardScaler()
self._scaler.fit(X[numeric_cols])
self.state = PreprocessingState(
categorical_strategy=self.categorical_strategy,
scaled_numeric=numeric_cols if self.scale else [],
dropped_columns=dropped,
feature_order=list(X.columns),
scale=self.scale,
onehot_categories=onehot_categories,
target_mappings=target_mappings,
target_global_mean=target_global_mean,
target_encoded_columns=target_encoded_cols
)
return self
def transform(self, X: pd.DataFrame) -> pd.DataFrame:
if self.state is None:
raise RuntimeError("Pipeline not fitted.")
X = X.copy()
if self.state.categorical_strategy == 'drop':
for col in self.state.dropped_columns:
if col in X.columns:
X = X.drop(columns=col)
missing = [c for c in self.state.feature_order if c not in X.columns]
if missing:
raise ValueError(f"Incoming data missing columns required by preprocessor: {missing}")
X = X[self.state.feature_order]
elif self.state.categorical_strategy == 'onehot':
X = self._apply_onehot_transform(X)
elif self.state.categorical_strategy == 'target':
X = self._apply_target_transform(X)
else:
raise NotImplementedError(f"Unknown strategy {self.state.categorical_strategy}")
if self.scale and self._scaler is not None:
# Ensure float dtype prior to scaling assignment to avoid pandas FutureWarning
for col in self.state.scaled_numeric:
if not pd.api.types.is_float_dtype(X[col]):
X[col] = X[col].astype('float64')
X.loc[:, self.state.scaled_numeric] = self._scaler.transform(X[self.state.scaled_numeric])
return X
def fit_transform(self, X: pd.DataFrame, y: Optional[pd.Series] = None) -> pd.DataFrame:
return self.fit(X, y).transform(X)
def save(self, path: str):
os.makedirs(os.path.dirname(path), exist_ok=True)
payload: Dict[str, Any] = {
'state': asdict(self.state) if self.state else None,
'categorical_strategy': self.categorical_strategy,
'scale': self.scale,
'scaler': self._scaler,
'target_min_samples': self.target_min_samples,
'target_smoothing': self.target_smoothing
}
joblib.dump(payload, path)
@classmethod
def load(cls, path: str) -> 'PreprocessingPipeline':
payload = joblib.load(path)
pipe = cls(
categorical_strategy=payload.get('categorical_strategy', 'drop'),
scale=payload.get('scale', True),
target_min_samples=payload.get('target_min_samples', 5),
target_smoothing=payload.get('target_smoothing', 10.0)
)
state_dict = payload.get('state')
if state_dict:
pipe.state = PreprocessingState(**state_dict)
pipe._scaler = payload.get('scaler')
return pipe
def to_metadata(self) -> Dict[str, Any]:
return asdict(self.state) if self.state else {}
"""Helper for future extension: registration of new categorical strategies.
Currently omitted for brevity."""
|