项目实训写实记录No.9

1.随机森林插补

随机森林填补：以待填补的缺失字段为目标，基于其它字段训练随机森林模型，用模型预测缺失字段的值

2.代码实现

def MonthlyIncome_pre(df):
    # 将MonthlyIncome中10以及一下的数值设置为NAN与其他NAN一起进行之后的预测
    df.loc[df.MonthlyIncome == 0, 'MonthlyIncome'] = np.nan
    tt = df.loc[df.MonthlyIncome <= 10, 'DebtRatio']
    tt1 = df.loc[df.MonthlyIncome <= 10, 'MonthlyIncome']
    for i in tt.index:
        tt[i] = tt[i] * tt1[i]
    df.loc[df.MonthlyIncome <= 10, 'DebtRatio'] = tt
    df.loc[df.MonthlyIncome <= 10, 'MonthlyIncome'] = np.nan

    # 使用随机森林预测
    random_state = 100
    rf_data = df[
        ['MonthlyIncome', 'NumberOfTime30-59DaysPastDueNotWorse', 'age', 'RevolvingUtilizationOfUnsecuredLines',
         'NumberOfOpenCreditLinesAndLoans', 'NumberRealEstateLoansOrLines']]

    # 以MonthlyIncome作为标签结果，以其它字段作为特征
    # 以缺失MonthlyIncome字段的样本作为测试数据，其它样本作为训练数据
    rf_train = rf_data[rf_data.MonthlyIncome.notnull()].values
    rf_test = rf_data[rf_data.MonthlyIncome.isnull()].values
    rf_model = RandomForestRegressor(random_state=random_state)
    # 根据已有数据去拟合随机森林模型
    rf_model.fit(rf_train[:, 1:], rf_train[:, 0])
    # 预测缺失值
    predicts = rf_model.predict(rf_test[:, 1:])
    print(predicts)

    # 将预测值作为插补值填充给之前的缺失样本
    df['MonthlyIncome_rf'] = df['MonthlyIncome']
    tt = df.loc[df.MonthlyIncome.isnull(), 'DebtRatio']
    tt1 = 0
    for i in tt.index:
        tt[i] = tt[i] / predicts[tt1]
        tt1 += 1
    df.loc[df.MonthlyIncome.isnull(), 'DebtRatio'] = tt
    df.loc[df.MonthlyIncome.isnull(), 'MonthlyIncome_rf'] = predicts
    df['MonthlyIncome_rf'] = df['MonthlyIncome_rf'].astype(int)
    return df