解决TypeError: 'numpy.ndarray' object is not callable错误：房价预测线性回归问题

问题背景 #

正在开展房价预测的简单线性回归任务，采用RMSE作为评估指标，编写代码后触发TypeError: 'numpy.ndarray' object is not callable错误，无法定位问题来源。

原代码 #

# A function that take one input of the dataset and return the RMSE (of the test data), and the intercept and coefficient
def simple_linear_model(train, test, input_feature):
    regr = linear_model.LinearRegression() # Create a linear regression object
    regr.fit(train.values(columns = [input_feature]), train.as_matrix(columns = ['Price'])) # Train the model
    RMSE = mean_squared_error(test.values(columns = ['Price']), 
                              regr.predict(test.values(columns = [input_feature])))**0.5 # Calculate the RMSE on test data
    return RMSE, regr.intercept_[0], regr.coef_[0][0]

input_list = data.columns.values.tolist() # list of column name
input_list.remove('Price')
simple_linear_result = pd.DataFrame(columns = ['feature', 'RMSE', 'intercept', 'coefficient'])

# loop that calculate the RMSE of the test data for each input 

for p in input_list:
    RMSE, w1, w0 = simple_linear_model(train_data, test_data, p)
    simple_linear_result = simple_linear_result.append({'feature':p, 'RMSE':RMSE, 'intercept':w0, 'coefficient': w1}
                                                       ,ignore_index=True)
simple_linear_result.sort_values('RMSE').head(10) # display the 10 best estimators

报错信息 #

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Input In [12], in <cell line: 6>()
      5 # loop that calculate the RMSE of the test data for each input 
      6 for p in input_list:
----> 7     RMSE, w1, w0 = simple_linear_model(train_data, test_data, p)
      8     simple_linear_result = simple_linear_result.append({'feature':p, 'RMSE':RMSE, 'intercept':w0, 'coefficient': w1}
      9                                                        ,ignore_index=True)
     10 simple_linear_result.sort_values('RMSE').head(10)

Input In [11], in simple_linear_model(train, test, input_feature)
      2 def simple_linear_model(train, test, input_feature):
      3     regr = linear_model.LinearRegression() # Create a linear regression object
----> 4     regr.fit(train.values(columns = [input_feature]), train.as_matrix(columns = ['Price'])) # Train the model
      5     RMSE = mean_squared_error(test.values(columns = ['Price']), 
      6                               regr.predict(test.values(columns = [input_feature])))**0.5 # Calculate the RMSE on test data
      7     return RMSE, regr.intercept_[0], regr.coef_[0][0]

TypeError: 'numpy.ndarray' object is not callable

错误原因 #

错误根源在以下两处：

• train.values(columns = [input_feature])：values是pandas DataFrame的属性，不是方法，直接访问train.values会返回一个numpy数组。你试图用调用方法的方式（加括号传参数）去使用它，就会触发错误——numpy数组不能被当作函数调用。
• train.as_matrix(columns = ['Price'])：as_matrix是已被废弃的pandas方法，且同样错误地用了类似方法调用的传参方式，另外即使正确使用，它返回的也是numpy数组，后续操作也会有问题。

此外，线性回归模型的输入特征需要是二维数组，直接取单列的values会得到一维数组，也会导致模型拟合报错。

修正后的代码 #

from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error
import pandas as pd

# 定义线性回归模型函数，返回测试集RMSE、截距和系数
def simple_linear_model(train, test, input_feature):
    regr = LinearRegression() 
    # 提取特征：转为二维数组
    X_train = train[[input_feature]].values
    # 提取目标变量：转为二维数组
    y_train = train[['Price']].values
    regr.fit(X_train, y_train)
    
    # 计算测试集RMSE
    X_test = test[[input_feature]].values
    y_test = test[['Price']].values
    y_pred = regr.predict(X_test)
    RMSE = mean_squared_error(y_test, y_pred)**0.5
    
    return RMSE, regr.intercept_[0], regr.coef_[0][0]

input_list = data.columns.values.tolist()
input_list.remove('Price')
simple_linear_result = pd.DataFrame(columns=['feature', 'RMSE', 'intercept', 'coefficient'])

# 遍历所有特征计算结果
for p in input_list:
    RMSE, w1, w0 = simple_linear_model(train_data, test_data, p)
    simple_linear_result = simple_linear_result.append(
        {'feature':p, 'RMSE':RMSE, 'intercept':w0, 'coefficient': w1},
        ignore_index=True
    )

# 展示RMSE最小的10个特征模型
simple_linear_result.sort_values('RMSE').head(10)

关键修正点 #

• 用train[[input_feature]].values替代错误的train.values(columns=...)，确保特征是二维数组
• 用train[['Price']].values替代train.as_matrix(columns=...)，同时保证目标变量是二维数组
• 拆分预测和RMSE计算步骤，让代码更清晰易维护

内容的提问来源于stack exchange，提问作者user20273524