import numpy as np import pandas as pd def mean_imputation(data, column): """平均値補填の実装""" # 欠損していない値の平均を計算 mean_value = data[column].mean() # 欠損値を平均値で補填 data[column].fillna(mean_value, inplace=True) return data # 使用例 df = pd.DataFrame({ 'age': [25, 30, np.nan, 35, np.nan, 42], 'income': [400, np.nan, 600, 550, np.nan, 750] }) # 年齢の欠損値を平均値で補填 df_imputed = mean_imputation(df.copy(), 'age') print(f"年齢の平均値: {df['age'].mean():.1f}")

from sklearn.linear_model import LinearRegression import numpy as np import pandas as pd def regression_imputation(data, target_col, predictor_cols): """回帰補填の実装""" # 完全ケースを抽出 complete_cases = data.dropna(subset=[target_col] + predictor_cols) # 説明変数と目的変数を分離 X_complete = complete_cases[predictor_cols] y_complete = complete_cases[target_col] # 回帰モデルの訓練 model = LinearRegression() model.fit(X_complete, y_complete) # 欠損値の予測 missing_mask = data[target_col].isna() if missing_mask.any(): X_missing = data.loc[missing_mask, predictor_cols] predicted_values = model.predict(X_missing) # 欠損値を予測値で補填 data.loc[missing_mask, target_col] = predicted_values return data, model # 使用例 df = pd.DataFrame({ 'age': [25, 30, 35, 28, 42, 38, 33], 'education': [12, 16, 18, 14, 20, 16, 14], 'income': [400, np.nan, 600, np.nan, 750, 650, 500] }) # 年収を年齢と教育年数で予測して補填 df_imputed, model = regression_imputation( df.copy(), 'income', ['age', 'education'] ) print(f"回帰係数: {model.coef_}") print(f"切片: {model.intercept_:.2f}")

from sklearn.experimental import enable_iterative_imputer from sklearn.impute import IterativeImputer import numpy as np import pandas as pd def mice_imputation(data, max_iter=10, random_state=42): """MICE補填の実装""" # IterativeImputerを使用（MICE の実装） imputer = IterativeImputer( max_iter=max_iter, random_state=random_state ) # 数値データのみを抽出 numeric_columns = data.select_dtypes(include=[np.number]).columns numeric_data = data[numeric_columns] # MICE補填を実行 imputed_array = imputer.fit_transform(numeric_data) # DataFrameに変換 imputed_df = pd.DataFrame(imputed_array, columns=numeric_columns) # 元のデータフレームにマージ result = data.copy() result[numeric_columns] = imputed_df return result, imputer # 使用例 df = pd.DataFrame({ 'age': [25, np.nan, 35, 28, np.nan, 38, 33], 'income': [400, 450, np.nan, 430, 750, 650, np.nan], 'education': [12, 16, 18, np.nan, 20, 16, 14], 'satisfaction': [4, 5, np.nan, 4, 2, 5, np.nan] }) # MICE補填を実行 df_mice, mice_model = mice_imputation(df) print("MICE補填完了") print(df_mice.head())

from sklearn.impute import KNNImputer from sklearn.preprocessing import StandardScaler import numpy as np import pandas as pd def knn_imputation(data, n_neighbors=5, weights='uniform'): """k-NN補填の実装""" # データの標準化（距離計算のため） scaler = StandardScaler() numeric_columns = data.select_dtypes(include=[np.number]).columns # k-NN Imputerの初期化 imputer = KNNImputer( n_neighbors=n_neighbors, weights=weights # 'uniform' or 'distance' ) # 数値データを抽出・標準化 numeric_data = data[numeric_columns] scaled_data = scaler.fit_transform(numeric_data) # k-NN補填を実行 imputed_scaled = imputer.fit_transform(scaled_data) # 元のスケールに戻す imputed_data = scaler.inverse_transform(imputed_scaled) # DataFrameに変換 result = data.copy() result[numeric_columns] = pd.DataFrame( imputed_data, columns=numeric_columns ) return result, imputer # 使用例 df = pd.DataFrame({ 'age': [25, 30, np.nan, 28, 42, 38, np.nan], 'income': [400, np.nan, 600, 430, 750, 650, 500], 'education': [12, 16, 18, 14, np.nan, 16, 14], 'satisfaction': [4, 5, 3, 4, 2, np.nan, 4] }) # k=3 でk-NN補填を実行 df_knn, knn_model = knn_imputation(df, n_neighbors=3) print("k-NN補填完了") print(df_knn.head())

import numpy as np import pandas as pd from sklearn.experimental import enable_iterative_imputer from sklearn.impute import IterativeImputer from scipy import stats def multiple_imputation(data, m=5, analysis_func=None): """多重代入法の実装""" imputed_datasets = [] analysis_results = [] # m個の補填されたデータセットを作成 for i in range(m): # 異なるランダムシードでMICE補填 imputer = IterativeImputer(random_state=i) numeric_cols = data.select_dtypes(include=[np.number]).columns imputed_array = imputer.fit_transform(data[numeric_cols]) imputed_df = data.copy() imputed_df[numeric_cols] = pd.DataFrame( imputed_array, columns=numeric_cols ) imputed_datasets.append(imputed_df) # 各データセットで分析を実行 if analysis_func: result = analysis_func(imputed_df) analysis_results.append(result) return imputed_datasets, analysis_results def pool_results(estimates, variances): """Rubinのルールで結果を統合""" m = len(estimates) # 統合推定値 pooled_estimate = np.mean(estimates) # 分散内変動 within_variance = np.mean(variances) # 分散間変動 between_variance = np.var(estimates, ddof=1) # 総分散 total_variance = within_variance + (1 + 1/m) * between_variance # 自由度 degrees_freedom = (m - 1) * (1 + within_variance / ((1 + 1/m) * between_variance))**2 return { 'estimate': pooled_estimate, 'variance': total_variance, 'std_error': np.sqrt(total_variance), 'degrees_freedom': degrees_freedom } # 使用例：回帰分析での多重代入 def regression_analysis(data): """回帰分析の例""" from sklearn.linear_model import LinearRegression from sklearn.metrics import r2_score X = data[['age', 'education']] y = data['income'] model = LinearRegression() model.fit(X, y) return { 'coefficients': model.coef_, 'intercept': model.intercept_, 'r2': r2_score(y, model.predict(X)) } # データ例 df = pd.DataFrame({ 'age': [25, 30, np.nan, 28, 42, 38, 33], 'education': [12, 16, 18, np.nan, 20, 16, 14], 'income': [400, np.nan, 600, 430, 750, 650, 500] }) # 多重代入を実行 datasets, results = multiple_imputation( df, m=5, analysis_func=regression_analysis ) print(f"作成されたデータセット数: {len(datasets)}") print(f"分析結果数: {len(results)}")