推荐阅读:
本次实验主要是自定义一个内核 Kernel 函数,然后使用 sklearn.datasets 自带的鸢尾花 Iris 数据集样本进行 SVM 分类。
import numpy as np
import pandas as pd
import matplotlib as mpl
import matplotlib.pyplot as plt
from sklearn import svm # sklearn自带SVM分类器
from sklearn import datasets # 导入数据集
from sklearn.model_selection import train_test_split
h = .02 # 网格中的步长
# 设置颜色
cm_light = mpl.colors.ListedColormap(['#A0FFA0', '#FFA0A0', '#A0A0FF'])
cm_dark = mpl.colors.ListedColormap(['g', 'r', 'b'])
# 自定义 Kernel 函数
def my_kernel(X, Y):
"""
We create a custom kernel:
(2 0)
k(X, Y) = X ( ) Y.T
(0 1)
"""
M = np.array( [ [2, 0], [0, 1.0] ] )
# M = np.array( [ [2, 1], [1, 1.0] ] )
return np.dot(np.dot(X, M), Y.T)
# 导入鸢尾花(Iris)数据集
iris = datasets.load_iris()
X = iris.data # 取数据集的特征向量
Y = iris.target # 取数据集的标签(鸢尾花类型)
X = X[:, 0:2] # 取前两列特征向量用作分类
x_train, x_test, y_train, y_test = train_test_split(X, Y, train_size = 0.8, random_state = 6)
clf = svm.SVC(kernel = my_kernel) # 创建 SVM 分类器
clf.fit(x_train, y_train.ravel()) # 拟合数据
x1_min, x1_max = X[:, 0].min(), X[:, 0].max() # 第0列的范围
x2_min, x2_max = X[:, 1].min(), X[:, 1].max() # 第1列的范围
xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h)) # 生成网格采样点
Z = clf.predict(np.c_[xx.ravel(), yy.ravel()]) # 预测分类值
Z = Z.reshape(xx.shape) # 使之与输入的形状相同
plt.pcolormesh(xx, yy, Z, cmap = cm_light)
plt.scatter(X[:, 0], X[:, 1], c = Y, edgecolors = 'k', cmap = cm_dark) # 样本
plt.xlim(x1_min, x1_max)
plt.ylim(x2_min, x2_max)
plt.title('3-Class classification using Support Vector Machine with custom kernel', fontsize = 12)
plt.axis('tight')
plt.show()
# 计算准确率
print("The scores of train set is %f" %(clf.score(x_train, y_train))) # 训练集准确率
print("The scores of test set is %f" %(clf.score(x_test, y_test))) # 测试集准确率
基于 Anaconda + Jupyter Notebook 环境的运行结果为:
推荐阅读: