超参数调节

雨点打透心脏的1/2处 2022-05-19 05:26 164阅读 0赞

## 1.网格搜索参数GridSearchCV 类 ##

class sklearn.model_selection.GridSearchCV(estimator, param_grid, scoring=None, fit_params=None, n_jobs=1, 
    iid=True, refit=True, cv=None, verbose=0, pre_dispatch=‘2*n_jobs’, error_score=’raise’, return_train_score=’warn’)

> 1.  estimator ： 分类器
> 2.  param\_grid : 需要调参的参数。格式为：param\_grid = \{'criterion': \['gini', 'entropy'\],  
>     'max\_depth': \[2,3,4,5,6\],  
>     'min\_samples\_split':\[2,3,4,5,6\],  
>     'min\_samples\_leaf':\[2,3,4,5,6\]  
>     \}
> 3.  scoring: 模型评分标准：如下
> 4.  refit : 默认为True ，在搜索参数结束后，用最佳参数结果再次fit一遍全部数据集
> 5.  cv： 交叉验证参数，默认为3折
> 6.  **verbose**：日志冗长度，int：冗长度，0：不输出训练过程，1：偶尔输出，>1：对每个子模型都输出
> 7.  ***pre\_dispatch=‘2\*n\_jobs’：*** 指定总共分发的并行任务数。

scoring参数：分类、聚类和回归的评估方法都有。

['accuracy', 'adjusted_mutual_info_score', 'adjusted_rand_score', 'average_precision', 'completeness_score',
     'explained_variance', 'f1', 'f1_macro', 'f1_micro', 'f1_samples', 'f1_weighted', 'fowlkes_mallows_score',
     'homogeneity_score', 'mutual_info_score', 'neg_log_loss', 'neg_mean_absolute_error', 'neg_mean_squared_error',
     'neg_mean_squared_log_error', 'neg_median_absolute_error', 'normalized_mutual_info_score', 'precision', 'precision_macro', 
    'precision_micro', 'precision_samples', 'precision_weighted', 'r2', 'recall', 'recall_macro', 'recall_micro', 
    'recall_samples', 'recall_weighted', 'roc_auc', 'v_measure_score']

具体含义可以参考：[http://sklearn.apachecn.org/cn/0.19.0/modules/model\_evaluation.html\#scoring-parameter][http_sklearn.apachecn.org_cn_0.19.0_modules_model_evaluation.html_scoring-parameter]

## 2.使用网格搜索： ##

需要我们人工手动输入的参数称为超参数， 进行超参数的选择的过程叫做调参。

#coding=gbk
    #调整估计器的超参数,进行超参数的选择的过程叫做调参
    
    #1，网格搜索方法 GridSearchCV
    from sklearn import datasets
    from sklearn.model_selection import train_test_split
    from sklearn.svm import SVC
    from sklearn.metrics import classification_report 
    
    from sklearn.model_selection import GridSearchCV    #导入网格搜索方法的包
    # Each datapoint is a 8x8 image of a digit. 每一个数据点都是 8x8 的像素点
    digits = datasets.load_digits()
    print(digits.data.shape)    # (1797, 64)
    print(digits.data[:5,:])
    
    # import matplotlib.pyplot as plt 
    # plt.gray() 
    # plt.matshow(digits.images[3]) 
    # plt.show() 
    n_samples = len(digits.images)
    print(n_samples)    # 1797
    X = digits.data
    y = digits.target
    print(y[:5])    #[0 1 2 3 4] 对应的数字
    
    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.5, random_state=0)
    
    param_grid = [{
        'kernel':['rbf'],
        'gamma':[1e-3, 1e-4],
        'C':[1, 10,100,1000]
        },{
            'kernel':['linear'],
            'C':[1, 10,100,1000]
            }]
    scores =['precision', 'recall']
    for score in scores :
        print('score %s' %score)
        print('------')
        clf = GridSearchCV(SVC(), param_grid, cv=5, scoring='%s_macro'%score)   
        clf.fit(X_train, y_train)
        print('best params is :')
        print(clf.best_params_)
        print('grid score')
        print()
        means = clf.cv_results_['mean_test_score']
        stds = clf.cv_results_['std_test_score']
        
        for mean, std, params in zip(means, stds, clf.cv_results_['params']):
            print('%.3f (+-/%0.03f) for %r'%(mean, std*2, params))
        
        print('-----')
        print('classification report')
        y_true, y_pred = y_test, clf.predict(X_test)
        report = classification_report(y_true, y_pred)
        print(report)

输出结果：

score precision
    ------
    best params is :
    {'kernel': 'rbf', 'gamma': 0.001, 'C': 10}
    grid score
    
    0.986 (+-/0.016) for {'kernel': 'rbf', 'gamma': 0.001, 'C': 1}
    0.959 (+-/0.029) for {'kernel': 'rbf', 'gamma': 0.0001, 'C': 1}
    0.988 (+-/0.017) for {'kernel': 'rbf', 'gamma': 0.001, 'C': 10}
    0.982 (+-/0.026) for {'kernel': 'rbf', 'gamma': 0.0001, 'C': 10}
    0.988 (+-/0.017) for {'kernel': 'rbf', 'gamma': 0.001, 'C': 100}
    0.982 (+-/0.025) for {'kernel': 'rbf', 'gamma': 0.0001, 'C': 100}
    0.988 (+-/0.017) for {'kernel': 'rbf', 'gamma': 0.001, 'C': 1000}
    0.982 (+-/0.025) for {'kernel': 'rbf', 'gamma': 0.0001, 'C': 1000}
    0.975 (+-/0.014) for {'kernel': 'linear', 'C': 1}
    0.975 (+-/0.014) for {'kernel': 'linear', 'C': 10}
    0.975 (+-/0.014) for {'kernel': 'linear', 'C': 100}
    0.975 (+-/0.014) for {'kernel': 'linear', 'C': 1000}
    -----
    classification report
                 precision    recall  f1-score   support
    
              0       1.00      1.00      1.00        89
              1       0.97      1.00      0.98        90
              2       0.99      0.98      0.98        92
              3       1.00      0.99      0.99        93
              4       1.00      1.00      1.00        76
              5       0.99      0.98      0.99       108
              6       0.99      1.00      0.99        89
              7       0.99      1.00      0.99        78
              8       1.00      0.98      0.99        92
              9       0.99      0.99      0.99        92
    
    avg / total       0.99      0.99      0.99       899
    
    score recall
    ------
    best params is :
    {'kernel': 'rbf', 'gamma': 0.001, 'C': 10}
    grid score
    
    0.986 (+-/0.019) for {'kernel': 'rbf', 'gamma': 0.001, 'C': 1}
    0.957 (+-/0.029) for {'kernel': 'rbf', 'gamma': 0.0001, 'C': 1}
    0.987 (+-/0.019) for {'kernel': 'rbf', 'gamma': 0.001, 'C': 10}
    0.981 (+-/0.028) for {'kernel': 'rbf', 'gamma': 0.0001, 'C': 10}
    0.987 (+-/0.019) for {'kernel': 'rbf', 'gamma': 0.001, 'C': 100}
    0.981 (+-/0.026) for {'kernel': 'rbf', 'gamma': 0.0001, 'C': 100}
    0.987 (+-/0.019) for {'kernel': 'rbf', 'gamma': 0.001, 'C': 1000}
    0.981 (+-/0.026) for {'kernel': 'rbf', 'gamma': 0.0001, 'C': 1000}
    0.972 (+-/0.012) for {'kernel': 'linear', 'C': 1}
    0.972 (+-/0.012) for {'kernel': 'linear', 'C': 10}
    0.972 (+-/0.012) for {'kernel': 'linear', 'C': 100}
    0.972 (+-/0.012) for {'kernel': 'linear', 'C': 1000}
    -----
    classification report
                 precision    recall  f1-score   support
    
              0       1.00      1.00      1.00        89
              1       0.97      1.00      0.98        90
              2       0.99      0.98      0.98        92
              3       1.00      0.99      0.99        93
              4       1.00      1.00      1.00        76
              5       0.99      0.98      0.99       108
              6       0.99      1.00      0.99        89
              7       0.99      1.00      0.99        78
              8       1.00      0.98      0.99        92
              9       0.99      0.99      0.99        92
    
    avg / total       0.99      0.99      0.99       899

实际当中有用的参数，以clf表示我们的GridSearchCV对象

**clf.best\_params\_** 返回最好的参数

**clf.best\_score\_** 返回最好的测试分数，它的值和 clf.cv\_results\_\['mean\_test\_score'\]\[dt\_grid.best\_index\_\] 是相同的。

**clf.best\_index\_** 返回列表中分数最好的下表

**clf.best\_estimator\_** 返回最好的模型

**grid\_scores\_** **在sklearn 0.18中已经不赞成使用了，用下面的cv\_results\_来代替**

**clf.cv\_results\_** 返回使用交叉验证进行搜索的结果，它本身又是一个字典，里面又有很多内容，我们来看一下上面的clf.cv\_results\_.keys()里面有什么：

dict_keys(
    ['mean_fit_time', 'std_fit_time', 'mean_score_time', 'std_score_time', 
    'param_C', 'param_gamma', 'param_kernel', 'params', 
    'split0_test_score', 'split1_test_score', 'split2_test_score', 'split3_test_score', 'split4_test_score',
    'mean_test_score', 'std_test_score', 'rank_test_score', 
    'split0_train_score', 'split1_train_score', 'split2_train_score', 'split3_train_score', 'split4_train_score', 
    'mean_train_score', 'std_train_score'] )

可以分为上面几类： 第一类是时间， 第二类是参数， 第三类是测试分数，其中又分为每次交叉验证的参数和统计的参数，第四类是训练分数，其中也分为每次交叉验证的参数和统计的参数。

参考：[https://www.cnblogs.com/jiaxin359/p/8641976.html\#\_labelTop][https_www.cnblogs.com_jiaxin359_p_8641976.html_labelTop]

[http://sklearn.apachecn.org/cn/0.19.0/modules/grid\_search.html\#grid-search-tips][http_sklearn.apachecn.org_cn_0.19.0_modules_grid_search.html_grid-search-tips]

[http_sklearn.apachecn.org_cn_0.19.0_modules_model_evaluation.html_scoring-parameter]: http://sklearn.apachecn.org/cn/0.19.0/modules/model_evaluation.html#scoring-parameter
[https_www.cnblogs.com_jiaxin359_p_8641976.html_labelTop]: https://www.cnblogs.com/jiaxin359/p/8641976.html#_labelTop
[http_sklearn.apachecn.org_cn_0.19.0_modules_grid_search.html_grid-search-tips]: http://sklearn.apachecn.org/cn/0.19.0/modules/grid_search.html#grid-search-tips