技術標籤：work

對資料建模之前，填補缺失值是必不可少的一步，這裡把用隨機森林模型快速預測缺失值的方法總結如下，以方便日後的工作。

# data_df: DataFrame型別的資料
# obj_column：待填補缺失值的列名
# missing_other_column：資料中含義空值的其他列
########## 缺失值處理

def fill_miss_byRandomForest(data_df , obj_column, missing_other_column ):
    ## 先把有缺失的其他列刪除掉missing_other_column
    data_df =
 
 data_df.drop(missing_other_column , axis = 1)
    # 分成已知該特徵和未知該特徵兩部分
    known = data_df[data_df[obj_column].notnull()]
    unknown = data_df[data_df[obj_column].isnull()]
    # y為結果標籤值
    y_know = known[obj_column]
    # X為特徵屬性值
    X_know= known.drop(obj_column , axis = 1)
    from sklearn.ensemble import
 
 RandomForestRegressor
    rfr = RandomForestRegressor(random_state=0, n_estimators=200,max_depth=3,n_jobs=-1)
    rfr.fit(X_know,y_know)
    # 用得到的模型進行未知特徵值預測
    # X為特徵屬性值
    X_unknow= unknown.drop(obj_column , axis = 1)
    predicted = rfr.predict(X_unknow).round(0)
    data_df.loc[(data_df[obj_column]
 
.isnull()), obj_column] = predicted
    return data_df

if __name__ == '__main__':
    import pandas as pd
    import numpy as np
    data = pd.read_csv('data/cs-training.csv')
    data.describe()
    data.columns.tolist()
    ### 呼叫fill_miss_byRandomForest函式，補充MonthlyIncome的缺失值
    data1_MonthlyIncome=fill_miss_byRandomForest(data , 'MonthlyIncome' , 'NumberOfDependents')#用隨機森林填補MonthlyIncome 的缺失值