SVR的代码（python）
项目中一个早期版本的代码，PCA-SVR，参数寻优采用传统的GridsearchCV。
1 from sklearn.decomposition import PCA
2 from sklearn.svm import SVR
3 from sklearn.model_selection import train_test_split
4 from sklearn.model_selection import GridSearchCV
5 from sklearn.metrics import r2_score, mean_squared_error, mean_absolute_error
6 from sklearn.preprocessing import StandardScaler, MinMaxScaler
7 from numpy import *
8 import numpy as np
9 import matplotlib.pyplot as plt
10 import xlrd
11 from svmutil import *
12 import pandas as pd
13
14 '''前言'''
15 # pca - svr
16 # CG测试
17
18 '''预设参数'''
19 fname = "all01.xlsx" # 训练数据文件读取 26hao
20 random_1 = 34 # 样本集选取随机种子
21 random_2 = 4 # 训练集选取随机种子
22 newpca = 6 # 降维
23 yuzhi = 50 # 异常点阈值
24 rate_1 = 0.8 # 样本集验证集
25 rate_2 = 0.8 # 训练集测试集
26 bestc = 384 # c
27 bestg = 9 # gamma
28
29 '''数据读取'''
30 # xlrd生成对excel表进行操作的对象
31 ...
32
33 # 输入输出分割
34 data_x = data[:, 1:11]
35 data_y = data[:, 0:1]
36
37 '''PCA'''
38 pca = PCA(n_components=newpca) # 加载PCA算法，设置降维后主成分数目为
39 data_x = pca.fit_transform(data_x) # 对样本进行降维
40 print(pca.components_) # 输出主成分，即行数为降维后的维数，列数为原始特征向量转换为新特征的系数
41 print(pca.explained_variance_ratio_) # 新特征 每维所能解释的方差大小在全方差中所占比例
42
43 '''数据划分'''
44 # 样本数据分割
45 train_data_x, predict_data_x, train_data_y, predict_data_y = train_test_split(data_x, data_y, test_size=rate_1,
46 random_state=random_1)
47
48 # 训练数据分割
49 train_x, test_x, train_y, test_y = train_test_split(train_data_x, train_data_y, test_size=rate_2, random_state=random_2)
50 predict_x = predict_data_x
51 predict_y = predict_data_y
52
53 # reshape y
54 test_y = np.reshape(test_y, -1)
55 train_y = np.reshape(train_y, -1)
56 predict_y = np.reshape(predict_y, (-1, 1))
57
58 # StandardScaler x
59 ss_X = StandardScaler()
60 ss_X.fit(train_data_x) # 20%
61 train_x = ss_X.transform(train_x)
62 test_x = ss_X.transform(test_x)
63 predict_x = ss_X.transform(predict_x)
64
65 '''参数优化与SVR'''
66 # 网格搜索交叉验证（GridSearchCV）：以穷举的方式遍历所有可能的参数组合
67 # 测试用
68 # param_grid = {'gamma': [bestg], 'C': [bestc]}
69 # rbf_svr_cg = GridSearchCV(SVR(kernel='rbf'), param_grid, cv=5)
70 # rbf_svr_cg.fit(train_x,train_y)
71 # bestc = rbf_svr_cg.best_params_.get('C')
72 # bestg = rbf_svr_cg.best_params_.get('gamma')
73
74 # 最优参数
75 print(bestc, bestg)
76 param_grid = {'gamma': [bestg], 'C': [bestc]}
77 rbf_svr = SVR(kernel='rbf',param_grid) # 需要修改
78
79 # 训练
80 rbf_svr.fit(train_x, train_y)
81
82 # 预测
83 test_y_predict = rbf_svr.predict(test_x)
84 test_y_predict = np.reshape(test_y_predict, (-1, 1))
85 predict_y_predict = rbf_svr.predict(predict_x)
86 predict_y_predict = np.reshape(predict_y_predict, (-1, 1))
87
88 '''去异常点'''
89 print('样本集：', len(train_data_y))
90 print('验证集：', len(predict_data_y))
91 size = len(test_y_predict)
92 count = 0
93 for i in range(size):
94 if abs(test_y_predict[size - i - 1] - test_y[size - i - 1]) > yuzhi:
95 test_y_predict = np.delete(test_y_predict, size - i - 1)
96 test_y = np.delete(test_y, size - i - 1)
97 count = count + 1
98 print('测试集异常点', count)
99 size = len(predict_y_predict)
100 count = 0
101 for i in range(size):
102 if abs(predict_y_predict[size - i - 1] - predict_y[size - i - 1]) > yuzhi:
103 predict_y_predict = np.delete(predict_y_predict, size - i - 1)
104 predict_y = np.delete(predict_y, size - i - 1)
105 count = count + 1
106 print('验证集异常点', count)
107
108 '''评估'''
109 # # 使用r2__score模块，并输出评估结果，拟合程度，R2决定系数，衡量模型预测能力好坏（真实与预测的相关程度百分比）
110 # print('The value of R-squared of kernal=rbf is',r2_score(test_y,test_y_predict))
111 # # 使用mean_squared_error模块，输出评估结果，均方误差
112 # print('The mean squared error of kernal=rbf is',mean_squared_error(test_y,test_y_predict))
113 # # 使用mean_absolute_error模块，输出评估结果，平均绝对误差
114 # print('The mean absolute error of kernal=rbf is',mean_absolute_error(test_y,test_y_predict))
115
116 # 使用r2__score模块，并输出评估结果，拟合程度，R2决定系数，衡量模型预测能力好坏（真实与预测的相关程度百分比）
117 print('The value of R-squared of kernal=rbf is', r2_score(predict_y, predict_y_predict))
118 # 使用mean_squared_error模块，输出评估结果，均方误差
119 print('The mean squared error of kernal=rbf is', mean_squared_error(predict_y, predict_y_predict))
120 # 使用mean_absolute_error模块，输出评估结果，平均绝对误差
121 print('The mean absolute error of kernal=rbf is', mean_absolute_error(predict_y, predict_y_predict))
122 # r
123 X1 = pd.Series(np.reshape(predict_y,-1))
124 Y1 = pd.Series(np.reshape(predict_y_predict,-1))
125 print('The r is', X1.corr(Y1, method="pearson"))
126 print('The r is', sqrt(r2_score(predict_y, predict_y_predict)))
127
128 '''作图'''
129 # PRN
130 print('PRN:', fname)
131
132 # PCA
133 print()
134
135 # 残差
136 diff_predict = predict_y_predict - predict_y
137 plt.plot(diff_predict, color='black', label='error')
138 plt.xlabel("no.")
139 plt.ylabel("error(m)")
140 plt.title('xxx')
141 plt.grid()
142 plt.legend()
143 plt.show()
144
145 # 真实/模型_1
146 plt.plot(predict_y, color='g', label='dtu15mss')
147 plt.plot(predict_y_predict, color='b', label='pre')
148 plt.xlabel("xxx")
149 plt.ylabel("error(m)")
150 plt.title('xxx')
151 plt.grid()
152 plt.legend()
153 plt.show()
154
155 # 真实/模型_2
156 fig = plt.figure(3)
157 ax1 = fig.add_subplot(2, 1, 1)
158 ax1.plot(predict_y, color='g', label='dtu15mss')
159 ax2 = fig.add_subplot(2, 1, 2)
160 ax2.plot(predict_y_predict, color='b', label='pre')
161 plt.show()
162
163 # 真实/模型_3
164 p_x = [x for x in range(int(min(predict_y)) - 5, int(max(predict_y)) + 5)]
165 p_y = p_x
166 plt.plot(p_x, p_y, color='black', label='1')
167 plt.scatter(predict_y_predict, predict_y, s=10, color='b', marker='x',
168 label='0') # https://www.cnblogs.com/shanlizi/p/6850318.html
169 plt.xlabel('PRE')
170 plt.ylabel('DTU')
171 plt.show()