决策树算法的sklearn实现及其调参

约定不等于承诺〃 2022-03-08 04:10 818阅读 0赞

# 回归决策树 #

import pandas as pd
    
    # 导入房价数据
    from sklearn.datasets.california_housing import fetch_california_housing
    housing = fetch_california_housing()
    
    # 查看房价数据的具体信息
    housing

housing的具体信息如下图所示：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70][]

# 将housing中的特征数据和标签数据构造成DataFrame对象
    df = pd.DataFrame(housing.data,columns=housing.feature_names)
    df.head()

![20190323210423455.png][]

# 查看房价数据的统计信息
    df.describe()

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 1][]

# 利用sklearn构造回归决策树模型
    from sklearn import tree
    dtr = tree.DecisionTreeRegressor(max_depth = 3)  # 设置树的深度为3
    dtr.fit(housing.data, housing.target)
    
    # 输出结果为：
    # DecisionTreeRegressor(criterion='mse', max_depth=3, max_features=None,
    # max_leaf_nodes=None, min_impurity_decrease=0.0,
    # min_impurity_split=None, min_samples_leaf=1,
    # min_samples_split=2, min_weight_fraction_leaf=0.0,
    # presort=False, random_state=None, splitter='best')

# 构造dot_data数据，用于可视化决策树结果
    dot_data = \
        tree.export_graphviz(
            dtr,                # 决策树模型名称
            out_file = None,   # 输出文件的句柄或名称，默认值是None
            feature_names = housing.feature_names,  # 训练所用特征名称
            filled = True,     # 指定是否给节点填充不同的颜色
            impurity = True,   # 指定在节点中显示不纯度，如MSE、MAE等
            rounded = True     # 设置节点在拐点处的形状为圆形
        )
    
    import pydotplus
    graph = pydotplus.graph_from_dot_data(dot_data)
    from IPython.display import Image
    Image(graph.create_png())

回归决策树模型的可视化结果：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 2][]

# 调用train_test_split将原始数据划分成训练集和测试集(验证集)
    from sklearn.model_selection import train_test_split
    data_train, data_test, target_train, target_test = \
        train_test_split(housing.data, housing.target, test_size = 0.1, random_state = 42)   # test_size指定测试集所占的比例
    dtr = tree.DecisionTreeRegressor(random_state = 42)
    dtr.fit(data_train, target_train)
    dtr.score(data_test, target_test)  # 输出结果为0.637355881715626

score的说明：

score(X, y, sample_weight=None) method of sklearn.tree.tree.DecisionTreeRegressor instance
        Returns the coefficient of determination R^2 of the prediction.
        
        The coefficient R^2 is defined as (1 - u/v), where u is the residual
        sum of squares ((y_true - y_pred) ** 2).sum() and v is the total
        sum of squares ((y_true - y_true.mean()) ** 2).sum().
        The best possible score is 1.0 and it can be negative (because the
        model can be arbitrarily worse). A constant model that always
        predicts the expected value of y, disregarding the input features,
        would get a R^2 score of 0.0.
        
        Parameters
        ----------
        X : array-like, shape = (n_samples, n_features)
            Test samples. For some estimators this may be a
            precomputed kernel matrix instead, shape = (n_samples,
            n_samples_fitted], where n_samples_fitted is the number of
            samples used in the fitting for the estimator.
        
        y : array-like, shape = (n_samples) or (n_samples, n_outputs)
            True values for X.
        
        sample_weight : array-like, shape = [n_samples], optional
            Sample weights.
        
        Returns
        -------
        score : float
            R^2 of self.predict(X) wrt. y.

# 导出各个特征的重要性值
    pd.Series(dtr.feature_importances_, index = housing.feature_names)

### ![20190323212300296.png][] ###

### 使用交叉验证调整回归决策树模型中的参数(网格搜索) ###

from sklearn.model_selection import GridSearchCV
    tree_param_grid = { 'max_depth':list((5,10,12)),'min_samples_split': list((2,4,6)),'min_samples_leaf':list((1,2,4))}
    grid = GridSearchCV(tree.DecisionTreeRegressor(),param_grid=tree_param_grid, cv=5)
    grid.fit(data_train, target_train)
    grid.cv_results_['mean_test_score'], grid.best_params_, grid.best_score_

交叉验证结果：

![20190323213013365.png][]

# 采用由交叉验证的到的“最佳”参数重新训练回顾决策树
    dtr2 = tree.DecisionTreeRegressor(max_depth=10,min_samples_leaf=4,min_samples_split=4)
    dtr2.fit(data_train, target_train)
    dtr2.score(data_test, target_test)   # 输出结果为0.6970711600432351

# 分类决策树 #

# 导入拍拍贷贷款数据集
    df = pd.read_excel('loans.xlsx','Sheet1')  
    df.head()

df左半部分数据：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 3][]

df右半部分数据：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 4][]

数据集中的特征说明：

*  grade：贷款级别
 *  `sub_grade`: 贷款细分级别
 *  `short_emp`：一年以内短期雇佣
 *  `emp_length_num`：受雇年限
 *  `home_ownership`：居住状态（自有，按揭，租住）
 *  dti：贷款占收入比例
 *  purpose：贷款用途
 *  term：贷款周期
 *  `last_delinq_none`：贷款申请人是否有不良记录
 *  `last_major_derog_none`：贷款申请人是否有还款逾期90天以上记录
 *  `revol_util`：透支额度占信用比例
 *  `total_rec_late_fee`：逾期罚款总额
 *  `safe_loans`：贷款是否安全

df.shape   # 输出结果为(46508, 13)

# 查看原始数据中中是否含有缺失值
    df.isnull().any()    # 从输出结果可以看到原始数据中不含缺失值
    
    # 输出结果为：
    # grade                    False
    # sub_grade                False
    # short_emp                False
    # emp_length_num           False
    # home_ownership           False
    # dti                      False
    # purpose                  False
    # term                     False
    # last_delinq_none         False
    # last_major_derog_none    False
    # revol_util               False
    # total_rec_late_fee       False
    # safe_loans               False
    # dtype: bool

# 从原始数据中提取出特征集和标签集
    X = df.drop('safe_loans', axis=1)
    y = df.safe_loans
    
    # 查看标签集中的数据分布是否均匀
    y.value_counts()
    
    # 输出结果为：
    #  1    23358
    # -1    23150

# 将特征集中的类别属性映射成对应的数值
    from sklearn.preprocessing import LabelEncoder
    from collections import defaultdict
    d = defaultdict(LabelEncoder)
    X_trans = X.apply(lambda x: d[x.name].fit_transform(x),axis=0)
    X_trans.head()

转换后的特征数据：

![2019032413530919.png][]

![20190324135321836.png][]

# 将数据划分成训练集和测试集(验证集)
    from sklearn.model_selection import train_test_split
    X_train, X_test, y_train, y_test = train_test_split(X_trans, y, random_state=1)
    
    # 调用sklearn包建立决策树模型
    from sklearn import tree
    clf = tree.DecisionTreeClassifier(max_depth=3,random_state=10)
    clf = clf.fit(X_train, y_train)
    
    column_names = X_train.columns.tolist()   # 提取特征名称
    
    #构造用于可视化决策树的dot数据
    dot_data = \
        tree.export_graphviz(
            clf,                # 决策树模型名称
            out_file = None,   # 输出文件的句柄或名称，默认值是None
            feature_names = column_names,  # 训练所用特征名称
            filled = True,     # 指定是否给节点填充不同的颜色
            impurity = True,   # 指定在节点中显示不纯度，如MSE、MAE等
            rounded = True     # 设置节点在拐点处的形状为圆形
        )
    
    # 可视化决策树
    import pydotplus
    graph = pydotplus.graph_from_dot_data(dot_data)
    from IPython.display import Image
    Image(graph.create_png())

分类决策树模型的可视化结果：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 5][]

# 评估模型性能
    from sklearn.metrics import accuracy_score
    accuracy_score(y_test, clf.predict(X_test))  # accuracy_score默认情况下计算的是分类准确率
    
    # 输出结果为0.6161520598606691

### 使用交叉验证调整分类决策树模型中的参数(网格搜索) ###

from sklearn.model_selection import GridSearchCV
    tree_param_grid = { 'max_depth':list((3,6,9)),'min_samples_split': list((2,4,6)),'min_samples_leaf':list((1,2,4))}
    grid = GridSearchCV(tree.DecisionTreeClassifier(),param_grid=tree_param_grid, cv=5)
    grid.fit(X_train, y_train)
    grid.cv_results_['mean_test_score'], grid.best_params_, grid.best_score_

交叉验证结果：

![20190324140134370.png][]

# 采用由交叉验证的到的“最佳”参数重新训练分类决策树
    clf2 = tree.DecisionTreeClassifier(max_depth=9,min_samples_leaf=2,min_samples_split=4)
    clf2 = clf2.fit(X_test, y_test)
    accuracy_score(y_test, clf2.predict(X_test)) 
       
    # 输出结果为0.6852154468048508

# 利用AUC度量分类器的性能
    from sklearn.metrics import roc_auc_score
    auc_score = roc_auc_score(y_test,clf2.predict(X_test))
    auc_score    # AUC的值与accuracy_score的值十分接近
    
    # auc_score的值为0.6852070693397126

决策树算法原理参考：

《机器学习》——周志华

《Python数据分析与机器学习实战》——唐宇迪

[http://www.mamicode.com/info-detail-2412736.html][http_www.mamicode.com_info-detail-2412736.html]

[http://www.cnblogs.com/pinard/p/6050306.html][http_www.cnblogs.com_pinard_p_6050306.html]

[http://www.cnblogs.com/pinard/p/6053344.html][http_www.cnblogs.com_pinard_p_6053344.html]

决策树算法调参参考：

[https://www.cnblogs.com/pinard/p/6056319.html][https_www.cnblogs.com_pinard_p_6056319.html]

[https://www.cnblogs.com/wanglei5205/p/8581354.html][https_www.cnblogs.com_wanglei5205_p_8581354.html]

[https://www.cnblogs.com/jiaxin359/p/8641976.html][https_www.cnblogs.com_jiaxin359_p_8641976.html]

其他参考：

[https://www.jianshu.com/p/67a71e366516][https_www.jianshu.com_p_67a71e366516]

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70]: /images/20220308/77b50bc5f6d24b12aa7359f73f82df80.png
[20190323210423455.png]: /images/20220308/e407243f2b4e4caf8f4dc805d10c3955.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 1]: /images/20220308/6c971aa3c17d48ca95f53ccce5d2f6a7.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 2]: /images/20220308/2ecae912ff974ee1a64c5b4498fcf691.png
[20190323212300296.png]: /images/20220308/137efe3dc6ea4808a76ce93ff6781997.png
[20190323213013365.png]: /images/20220308/516eed43a0f743d4894aca4f153baf84.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 3]: /images/20220308/6dd55a18810b418dbb34a92ba4d26377.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 4]: /images/20220308/a591f567ce574b828031320bfe52cbab.png
[2019032413530919.png]: /images/20220308/5ae73b3cb3934e23a2151834a2231368.png
[20190324135321836.png]: /images/20220308/29a321c8e6804ef6a4300254b312f545.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMDgwODUw_size_16_color_FFFFFF_t_70 5]: /images/20220308/5a8514b55b3f4295bb3608efd65a9a61.png
[20190324140134370.png]: /images/20220308/28c180554d404ed4a1e5d722a7cd0f1c.png
[http_www.mamicode.com_info-detail-2412736.html]: http://www.mamicode.com/info-detail-2412736.html
[http_www.cnblogs.com_pinard_p_6050306.html]: http://www.cnblogs.com/pinard/p/6050306.html
[http_www.cnblogs.com_pinard_p_6053344.html]: http://www.cnblogs.com/pinard/p/6053344.html
[https_www.cnblogs.com_pinard_p_6056319.html]: https://www.cnblogs.com/pinard/p/6056319.html
[https_www.cnblogs.com_wanglei5205_p_8581354.html]: https://www.cnblogs.com/wanglei5205/p/8581354.html
[https_www.cnblogs.com_jiaxin359_p_8641976.html]: https://www.cnblogs.com/jiaxin359/p/8641976.html
[https_www.jianshu.com_p_67a71e366516]: https://www.jianshu.com/p/67a71e366516