统计学习方法 感知机python实现

感知机python实现

补习机器学习基础知识，把以前的代码复习一遍。
统计学习方法一书中感知机学习算法对偶形式的python实现，分为两部分代码，数据生成部分和感知机学习部分，可独立运行。

数据生成

先拟定一超平面，在超平面两侧生成正负样本

# -*- coding: utf-8 -*-
import numpy as np
from mpl_toolkits.mplot3d import axes3d
import matplotlib.pyplot as plt

class data_set:	#数据集类
	#__init__:初始化函数
	#	num：需要构造的数据量
	#	feature_size:特征数量
	#	w, b:超平面参数
	#	low：特征最小值
	#	high：特征最大值
	#build_set:构造数据集函数
	#	data：数据集
	#	label：标签
	#show_set:数据集可视化
	def __init__(self, num, feature_size, w, b, low, high):
		self.num = num
		self.feature_size = feature_size
		self.w = w
		self.b = b
		self.low = low
		self.high = high
	
	def build_set(self):
		row_data = np.random.randint(self.low, self.high, size = (self.num * 2, self.feature_size))
		row_label = np.zeros(self.num * 2, dtype = int)
		data = np.empty(shape = [0, self.feature_size], dtype = int)
		label = np.empty(shape = [0], dtype = int)
		data_num = 0
		for i in range(self.num * 2):
			x = row_data[i, :]
			if np.dot(self.w.transpose(), x) + self.b > 0.1:
				row_label[i] = 1
			elif np.dot(self.w.transpose(), x) + self.b < -0.1:
				row_label[i] = -1
			else: continue
			data = np.append(data, [row_data[i, :]], axis = 0)
			label = np.append(label, [row_label[i]], axis = 0)
			data_num += 1
			if data_num == self.num:
				break
		self.data, self.label = data, label
		return
	
	def show_set(self):
		for i in range(self.num):
			plt.scatter(self.data[i, 0] , self.data[i, 1], marker='x', color='red' if self.label[i] == -1 else 'green', s=5)
		x1 = np.linspace(self.low, self.high, self.high - self.low)
		x2 = (-self.b - self.w[0] * x1) / (self.w[1] + 0.001)
		plt.plot(x1, x2, color = 'blue')
		plt.show()	
	
	def show_set_3D(self):
		fig = plt.figure()
		ax = fig.gca(projection='3d')
		for i in range(self.label.shape[0]):
			ax.scatter(self.data[i, 0] , self.data[i, 1], self.data[i, 2], marker='x', color='red' if self.label[i] == -1 else 'green', s=5)
		x1 = np.linspace(np.min(self.data[:, 0]), np.max(self.data[:, 0]), np.max(self.data[:, 0]) - np.min(self.data[:, 0]))
		x2 = np.linspace(np.min(self.data[:, 0]), np.max(self.data[:, 0]), np.max(self.data[:, 0]) - np.min(self.data[:, 0]))
		x1, x2 = np.meshgrid(x1, x2)
		x3 = (-self.b - self.w[0] * x1 - self.w[1] * x2) / (self.w[2] + 0.001)
		surf = ax.plot_surface(x1, x2, x3, rstride=1, cstride=1, cmap=plt.get_cmap('rainbow'), linewidth=0, antialiased=True)
		plt.show()
			
if __name__ == '__main__':
	set_ = data_set(100, 3, np.array([1, 1, 1]), 3, -10, 10)
	set_.build_set()
	set_.show_set_3D()

感知机学习

初始化w和b均为0，根据统计学习方法书中的公式实现感知机的学习

# -*- coding: utf-8 -*-
import numpy as np
from data_building import data_set
from mpl_toolkits.mplot3d import axes3d
from matplotlib import cm
import matplotlib.pyplot as plt

class perceptron:
	#__init__:初始化函数
	#	data：数据集
	#	label：标签
	#	lr:学习率
	#isError:判断数据是否误分类函数
	#show_result:分类结果可视化
	def __init__(self, data, label, lr):
		self.data = data
		self.label = label
		self.lr = lr
		self.Alpha = np.zeros(label.shape[0], dtype = float)
		self.b = 0
		self.w = np.zeros(data.shape[1], dtype = float)
		self.Gram = np.dot(self.data, self.data.transpose())
	
	def isError(self, index):
		if self.label[index] *((self.Alpha * self.label * self.Gram[index, :]).sum() + self.b) <= 0:
			return True
		return False
	
	def train(self, max_step):
		isFinish = False
		while not isFinish:
			for i in range(self.label.shape[0]):
				if self.isError(i):
					self.Alpha[i] += self.lr
					self.b += (self.lr * self.label[i]).sum()
					isFinish = True
					for j in range(self.label.shape[0]):
						if self.isError(j):
							isFinish = False
				if isFinish:
					break
			self.w = np.dot(self.Alpha * self.label, self.data)
		return

	def show_result(self):
		for i in range(self.label.shape[0]):
			plt.scatter(self.data[i, 0] , self.data[i, 1], marker='x', color='red' if self.label[i] == -1 else 'green', s=5)
		x1 = np.linspace(np.min(self.data[:, 0]), np.max(self.data[:, 0]), np.max(self.data[:, 0]) - np.min(self.data[:, 0]))
		x2 = (-self.b - self.w[0] * x1) / (self.w[1] + 0.001)
		plt.plot(x1, x2, color = 'blue')
		plt.show()	
	
	def show_result_3D(self):		
		fig = plt.figure()
		ax = fig.gca(projection='3d')
		for i in range(self.label.shape[0]):
			ax.scatter(self.data[i, 0] , self.data[i, 1], self.data[i, 2], marker='x', color='red' if self.label[i] == -1 else 'green', s=5)
		x1 = np.linspace(np.min(self.data[:, 0]), np.max(self.data[:, 0]), np.max(self.data[:, 0]) - np.min(self.data[:, 0]))
		x2 = np.linspace(np.min(self.data[:, 0]), np.max(self.data[:, 0]), np.max(self.data[:, 0]) - np.min(self.data[:, 0]))
		x1, x2 = np.meshgrid(x1, x2)
		x3 = (-self.b - self.w[0] * x1 - self.w[1] * x2) / (self.w[2] + 0.001)
		surf = ax.plot_surface(x1, x2, x3, rstride=1, cstride=1, cmap=plt.get_cmap('rainbow'), linewidth=0, antialiased=True)
		plt.show()	
			

if __name__ == '__main__':
	set_ = data_set(100, 3, np.array([1, 1, 1]), 3, -10, 10)
	set_.build_set()
	model = perceptron(set_.data, set_.label, 0.1)
	model.train(100)
	print(model.w, model.b)
	model.show_result_3D()