単変量分析の完全ガイド — 各カラムの特徴を徹底的に理解する

import pandas as pd
import numpy as np
from scipy import stats

df = pd.read_csv("dataset.csv")

# 基本的な記述統計量
print(df.describe())

# より詳細な統計量を追加
def detailed_stats(series: pd.Series) -> dict:
    """数値カラムの詳細な記述統計量を計算する。"""
    return {
        "件数": series.count(),
        "平均": series.mean(),
        "中央値": series.median(),
        "最頻値": series.mode().iloc[0] if not series.mode().empty else None,
        "標準偏差": series.std(),
        "分散": series.var(),
        "最小値": series.min(),
        "25%点": series.quantile(0.25),
        "75%点": series.quantile(0.75),
        "最大値": series.max(),
        "範囲": series.max() - series.min(),
        "IQR": series.quantile(0.75) - series.quantile(0.25),
        "歪度": series.skew(),
        "尖度": series.kurtosis(),
        "変動係数": series.std() / series.mean() if series.mean() != 0 else None,
    }

for col in df.select_dtypes(include=["number"]).columns:
    print(f"\n=== {col} ===")
    for k, v in detailed_stats(df[col]).items():
        if isinstance(v, float):
            print(f"  {k}: {v:.4f}")
        else:
            print(f"  {k}: {v}")

from scipy.stats import skew, kurtosis

# 数値カラムの歪度・尖度を一括計算
numeric_cols = df.select_dtypes(include=["number"]).columns

skew_kurt = pd.DataFrame({
    "歪度": df[numeric_cols].apply(skew, nan_policy="omit"),
    "尖度": df[numeric_cols].apply(kurtosis, nan_policy="omit"),
})

skew_kurt["歪みの方向"] = skew_kurt["歪度"].apply(
    lambda s: "右に歪み" if s > 0.5 else ("左に歪み" if s < -0.5 else "ほぼ対称")
)
skew_kurt["尖り具合"] = skew_kurt["尖度"].apply(
    lambda k: "尖った分布" if k > 1 else ("平たい分布" if k < -1 else "正規分布に近い")
)

print(skew_kurt.round(3))

# カテゴリ型カラムの分析
for col in df.select_dtypes(include=["object", "category"]).columns:
    print(f"\n=== {col} ===")
    print(f"  ユニーク値数: {df[col].nunique()}")
    print(f"  最頻値: {df[col].mode().iloc[0]}")
    print(f"  最頻値の割合: {df[col].value_counts(normalize=True).iloc[0]:.2%}")
    print(f"\n  上位10カテゴリ:")
    vc = df[col].value_counts()
    for val, count in vc.head(10).items():
        pct = count / len(df) * 100
        bar = "█" * int(pct / 2)
        print(f"    {val:30s} {count:>6,} ({pct:5.1f}%) {bar}")
    if len(vc) > 10:
        others = vc.iloc[10:].sum()
        print(f"    {'(その他)':30s} {others:>6,} ({others / len(df) * 100:5.1f}%)")

# パーセンタイル分析
percentiles = [0.01, 0.05, 0.1, 0.25, 0.5, 0.75, 0.9, 0.95, 0.99]
for col in df.select_dtypes(include=["number"]).columns:
    vals = df[col].quantile(percentiles)
    print(f"\n{col}:")
    for p, v in vals.items():
        print(f"  {p*100:5.1f}%: {v:>12,.2f}")

    # 外れ値の候補を検出
    p99 = df[col].quantile(0.99)
    max_val = df[col].max()
    if max_val > p99 * 3:  # 99%点の3倍以上
        n_extreme = (df[col] > p99 * 3).sum()
        print(f"  ⚠ 極端な外れ値の候補: {n_extreme}件（99%点の3倍超）")