sklearn神经网络回归示例

以你之姓@ 2022-09-04 03:45 234阅读 0赞

# 1. 数据集 #

这里主要采用sklearn自带的波士顿房价预测的数据集

from sklearn.datasets import load_boston
    boston = load_boston()
    data = boston['data']  # shape=(506, 13) 
    target = boston['target']  # shape=(506,)
    # print(boston['DESCR'])

# 2. 划分数据集 #

划分数据集，将样本划分为训练集和验证集

from sklearn.model_selection import train_test_split
    train_data, test_data, train_target, test_target = train_test_split(data, target, test_size=0.2, random_state=42)
    # 训练集和测试集的shape：(404, 13)与 (102, 13)

对训练集进行标准化，同时将标准化的规则应用到验证集

from sklearn.preprocessing import StandardScaler
    std = StandardScaler()
    train_data_std = std.fit_transform(train_data)
    test_data_std = std.transform(test_data)

# 3. 构建模型 #

regr = MLPRegressor(solver='adam', hidden_layer_sizes=(50, 50), activation='tanh', max_iter=5000).fit(train_data_std, train_target)
    y_pred = regr.predict(test_data_std)

# 4. 评价回归模型 #

关于回归模型的评价指标如下：

<table> 
 <thead> 
 <tr> 
 <th>方法名称</th> 
 <th>最优值</th> 
 <th>sklearn函数</th> 
 </tr> 
 </thead> 
 <tbody> 
 <tr> 
 <td>平均绝对误差</td> 
 <td>0.0</td> 
 <td>sklearn.metrics.mean_absolute_error</td> 
 </tr> 
 <tr> 
 <td>均方误差</td> 
 <td>0.0</td> 
 <td>sklearn.metrics.mean_squared_error</td> 
 </tr> 
 <tr> 
 <td>中值绝对误差</td> 
 <td>0.0</td> 
 <td>sklearn.metrics.median_absolute_error</td> 
 </tr> 
 <tr> 
 <td>可解释方差值</td> 
 <td>1.0</td> 
 <td>sklearn.metrics.explained_variance_score</td> 
 </tr> 
 <tr> 
 <td> R 2 R^2 R2值</td> 
 <td>1.0</td> 
 <td>sklearn.metrics.r2_score</td> 
 </tr> 
 </tbody> 
</table>

其中，平均绝对误差、均方误差、中值绝对误差的值越接近0，模型性能越好。可解释方差值和 R 2 R^2 R2值越接近1，模型性能越好。

此处采用平均绝对误差作为评价指标

error = mean_absolute_error(test_target, y_pred)
    print(error)  # 2.3726167664920306

以折线图展示预测值与真实值

plt.figure()
    plt.plot(range(len(y_pred)), y_pred, color='blue')
    plt.plot(range(len(y_pred)), test_target, color='red')
    plt.show()

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM4MDQ4NzU2_size_16_color_FFFFFF_t_70]

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzM4MDQ4NzU2_size_16_color_FFFFFF_t_70]: /images/20220829/4b22a30df86941518b42a34eacef2fde.png