SVM利用网格搜索和交叉验证进行超参选择

最新推荐文章于 2025-06-23 17:00:07 发布

原创最新推荐文章于 2025-06-23 17:00:07 发布 · 1.2w 阅读

40 ·

CC 4.0 BY-SA版权

机器学习专栏收录该内容

42 篇文章

订阅专栏

本文详细介绍使用Python和scikit-learn库进行SVM（支持向量机）模型的参数调优过程。通过生成两组正态分布数据，分别作为正类和负类，构建数据集并使用GridSearchCV进行交叉验证，寻找最佳参数组合。最终，模型在测试集上展现出优秀的分类性能。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

import numpy as np 
#产生正态分布的数据100组，中心点（0，0），其标准差σ为1
p=np.random.randn(100,2)
#将中心点移动到（3.5,3.5），作为正类
for i in range(100):
    p[i][0]+=3.5
    p[i][1]+=3.5

#产生正态分布的数据100组，中心点（0，0），其标准差σ为1，作为负类
f=np.random.randn(100,2)

import pandas as pd 

#将np数组转换成dataframe
df_p=pd.DataFrame(p,columns=['x','y'])
#加上标签z,正类标签1
df_p['z']=1

#将np数组转换成dataframe
df_f=pd.DataFrame(f,columns=['x','y'])
#加上标签z,负类标签0
df_f['z']=0

#将正负类合并成一个dataframe
res = pd.concat([df_p, df_f], axis=0)
res.head(10)

.dataframe tbody tr th {
    vertical-align: top;
}

.dataframe thead th {
    text-align: right;
}

	x	y	z
0	4.198390	2.360253	1
1	4.802608	4.237265	1
2	3.571716	4.246163	1
3	3.389474	2.777169	1
4	3.407080	2.969104	1
5	3.962996	4.118016	1
6	2.679486	1.483121	1
7	3.547707	3.448452	1
8	3.181810	4.744277	1
9	3.412752	3.591759	1

import matplotlib.pyplot as plt

#绘制出数据集的散点图
plt.scatter(res['x'], res['y'], c=res['z'],cmap=plt.cm.Paired)
plt.xlabel('x')
plt.ylabel('y')
plt.title('random data')
plt.show()

在这里插入图片描述

from sklearn.model_selection import train_test_split

#划分测试集
train_x,test_x,train_y,test_y=train_test_split(res[['x','y']],res[['z']],test_size=0.3,random_state=0)
len(train_x)

from sklearn.model_selection import GridSearchCV
from sklearn import svm
#构建参数网格
parameters =[{'kernel': ['linear'], 'C': [1, 10, 100, 1000]}]

clf=GridSearchCV(estimator=svm.SVC(),param_grid=parameters,cv=5,n_jobs=-1,scoring='precision')
clf.fit(train_x,train_y)

e:\Anaconda3\lib\site-packages\sklearn\utils\validation.py:578: DataConversionWarning: A column-vector y was passed when a 1d array was expected. Please change the shape of y to (n_samples, ), for example using ravel().
  y = column_or_1d(y, warn=True)

GridSearchCV(cv=5, error_score='raise',
       estimator=SVC(C=1.0, cache_size=200, class_weight=None, coef0=0.0,
  decision_function_shape='ovr', degree=3, gamma='auto', kernel='rbf',
  max_iter=-1, probability=False, random_state=None, shrinking=True,
  tol=0.001, verbose=False),
       fit_params=None, iid=True, n_jobs=-1,
       param_grid=[{'kernel': ['linear'], 'C': [1, 10, 100, 1000]}],
       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',
       scoring='precision', verbose=0)

clf.best_params_

{'C': 1, 'kernel': 'linear'}

print("Best parameters set found on development set:")
print()
print(clf.best_params_)
print()
print("Grid scores on development set:")
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("%0.3f (+/-%0.03f) for %r"
          % (mean, std * 2, params))
print()

Best parameters set found on development set:

{'C': 1, 'kernel': 'linear'}

Grid scores on development set:

0.987 (+/-0.053) for {'C': 1, 'kernel': 'linear'}
0.987 (+/-0.053) for {'C': 10, 'kernel': 'linear'}
0.987 (+/-0.053) for {'C': 100, 'kernel': 'linear'}
0.987 (+/-0.053) for {'C': 1000, 'kernel': 'linear'}

from sklearn.metrics import classification_report
print("Detailed classification report:")
print()
print("The model is trained on the full development set.")
print("The scores are computed on the full evaluation set.")
print()
print(classification_report(test_y, clf.predict(test_x)))
print()

Detailed classification report:

The model is trained on the full development set.
The scores are computed on the full evaluation set.

             precision    recall  f1-score   support

          0       1.00      1.00      1.00        31
          1       1.00      1.00      1.00        29

avg / total       1.00      1.00      1.00        60