Day 082 — 特徴量エンジニアリング応用 — Titanic の隠れた特徴量を掘り起こす

# Title 抽出
df['Title'] = df['Name'].str.extract(r', ([A-Za-z]+)\.', expand=False)
rare = [t for t in df['Title'].unique() if t not in ['Mr', 'Miss', 'Mrs', 'Master']]
df['Title'] = df['Title'].replace(rare, 'Rare')

# FamilySize / IsAlone
df['FamilySize'] = df['SibSp'] + df['Parch'] + 1
df['IsAlone']    = (df['FamilySize'] == 1).astype(int)

# Deck
df['Deck'] = df['Cabin'].str[0].fillna('U')

def add_features(df, deck_le=None):
    df = df.copy()
    df['Title'] = df['Name'].str.extract(r', ([A-Za-z]+)\.', expand=False)
    rare = [t for t in df['Title'].unique() if t not in ['Mr','Miss','Mrs','Master']]
    df['Title'] = df['Title'].replace(rare, 'Rare').map({'Mr':0,'Miss':1,'Mrs':2,'Master':3,'Rare':4}).fillna(4)
    df['FamilySize'] = df['SibSp'] + df['Parch'] + 1
    df['IsAlone']    = (df['FamilySize'] == 1).astype(int)
    df['Deck_raw'] = df['Cabin'].str[0].fillna('U')
    if deck_le is None:
        deck_le = LabelEncoder()
        df['Deck'] = deck_le.fit_transform(df['Deck_raw'])
    else:
        known = set(deck_le.classes_)
        df['Deck_raw'] = df['Deck_raw'].apply(lambda x: x if x in known else 'U')
        df['Deck'] = deck_le.transform(df['Deck_raw'])
    df.drop(columns=['Deck_raw'], inplace=True)
    return df, deck_le

train, le = add_features(train)
test,  _  = add_features(test, le)

import pandas as pd
import numpy as np
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import StratifiedKFold, cross_val_score
from sklearn.preprocessing import StandardScaler, LabelEncoder
from sklearn.pipeline import Pipeline

train = pd.read_csv('/kaggle/input/titanic/train.csv')
test  = pd.read_csv('/kaggle/input/titanic/test.csv')

# ── 基本前処理（Day081 と同じ） ──
def basic_preprocess(df, age_med=None, fare_med=None, emb_mode=None):
    df = df.copy()
    if age_med  is None: age_med  = df['Age'].median()
    if fare_med is None: fare_med = df['Fare'].median()
    if emb_mode is None: emb_mode = df['Embarked'].mode()[0]
    df['Age']      = df['Age'].fillna(age_med)
    df['Fare']     = df['Fare'].fillna(fare_med)
    df['Embarked'] = df['Embarked'].fillna(emb_mode)
    df['Sex']      = df['Sex'].map({'male': 0, 'female': 1})
    df['Embarked'] = df['Embarked'].map({'S': 0, 'C': 1, 'Q': 2})
    return df, age_med, fare_med, emb_mode

train, a, f, e = basic_preprocess(train)
test,  _, _, _ = basic_preprocess(test, a, f, e)

# ── 特徴量エンジニアリング ──
def add_features(df, deck_le=None):
    df = df.copy()
    # ① Title
    df['Title'] = df['Name'].str.extract(r', ([A-Za-z]+)\.', expand=False)
    rare = [t for t in df['Title'].unique() if t not in ['Mr','Miss','Mrs','Master']]
    df['Title'] = df['Title'].replace(rare, 'Rare')
    df['Title'] = df['Title'].map({'Mr':0,'Miss':1,'Mrs':2,'Master':3,'Rare':4}).fillna(4)
    # ② FamilySize / IsAlone
    df['FamilySize'] = df['SibSp'] + df['Parch'] + 1
    df['IsAlone']    = (df['FamilySize'] == 1).astype(int)
    # ③ Deck
    df['Deck_raw'] = df['Cabin'].str[0].fillna('U')
    if deck_le is None:
        deck_le = LabelEncoder()
        df['Deck'] = deck_le.fit_transform(df['Deck_raw'])
    else:
        known = set(deck_le.classes_)
        df['Deck_raw'] = df['Deck_raw'].apply(lambda x: x if x in known else 'U')
        df['Deck'] = deck_le.transform(df['Deck_raw'])
    df.drop(columns=['Deck_raw'], inplace=True)
    return df, deck_le

train, le = add_features(train)
test,  _  = add_features(test, le)

# ── CV 比較 ──
y   = train['Survived']
skf = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)

base_features = ['Pclass','Sex','Age','SibSp','Parch','Fare','Embarked']
new_features  = base_features + ['Title','FamilySize','IsAlone','Deck']

def cv_score(features, label):
    pipe   = Pipeline([('sc',StandardScaler()),('lr',LogisticRegression(max_iter=1000,random_state=42))])
    scores = cross_val_score(pipe, train[features], y, cv=skf, scoring='accuracy')
    print(f"{label}: {scores.mean():.4f} ± {scores.std():.4f}")
    return scores.mean()

b = cv_score(base_features, "ベースライン（Day081）")
n = cv_score(new_features,  "新特徴量追加後      ")
print(f"改善量: +{n-b:.4f}")

# ── 係数（特徴量重要度） ──
pipe_final = Pipeline([('sc',StandardScaler()),('lr',LogisticRegression(max_iter=1000,random_state=42))])
pipe_final.fit(train[new_features], y)
coefs = pd.Series(np.abs(pipe_final.named_steps['lr'].coef_[0]), index=new_features).sort_values(ascending=False)
print("\n特徴量重要度（|係数|）:\n", coefs.to_string())

# ── 提出ファイル ──
sub = pd.DataFrame({'PassengerId': test['PassengerId'], 'Survived': pipe_final.predict(test[new_features])})
sub.to_csv('submission_fe.csv', index=False)
print(f"\nsubmission_fe.csv: {len(sub)} 行")