# -*- coding: utf-8 -*-
"""
@File : SVM.py
@Software: PyCharm
"""
import numpy as np
# import pandas as pd
from scipy.optimize import fmin_l_bfgs_b
from scipy.optimize import fmin_tnc
# from scipy.optimize import fmin_bfgs
import matplotlib.pyplot as plt
import matplotlib
from matplotlib.patches import Circle
matplotlib.rcParams['font.family'] = 'SimHei' # 用来正常显示中文
plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号
# 定义目标函数，w是未知数据，args是已知数，求w的最优解
def func(w, *args):
X, Y, c = args
yp = np.dot(X, w) # w*x，注意已经将b作为x的第一列进行了计算，w*x就是我们现在预测的y
idx = np.where(yp * Y < 1)[0] # 找到分错的数据索引位置
e = yp[idx] - Y[idx] # y预测值-y真实值，误差
cost = np.dot(e, e) + c * np.dot(w, w) # 平方和损失，c：学习率，加w的二范式惩罚
grand = 2 * (np.dot(X[idx].T, e) + c * w) # 梯度下降？？
return cost, grand
def plotResult(w):
margin = 2 / np.sqrt(np.dot(w[1:3], w[1:3]))
plot_x = np.append(np.min(x, 0)[0] - 0.2, np.max(x, 0)[0] + 0.2)
plot_y = -(plot_x * w[1] + w[0]) / w[2]
plt.figure()
pos = (Y == 1) # 正类
neg = (y == -1) # 负类
plt.plot(x[pos][:, 0], x[pos][:, 1], "r+", label="正类")
plt.plot(x[neg][:, 0], x[neg][:, 1], "bo", label="负类")
plt.plot(plot_x, plot_y, "r-", label="分割超平面")
plt.plot(plot_x, plot_y + margin / 2, "g-.", label="")
plt.plot(plot_x, plot_y - margin / 2, "g-.", label="")
plt.xlabel('x1')
plt.ylabel('x2')
plt.title('SVM Demo')
plt.legend()
plt.show()
# 简易smo算法开始####################################################################
# 随机选择第2个alpha
def selectJrand(i, m):
j = i # we want to select any J not equal to i
while (j == i):
j = int(np.random.uniform(0, m))
return j
# 调整大于H或小于L的alpha值
def clipAlpha(aj, H, L):
if aj > H:
aj = H
if L > aj:
aj = L
return aj
# 公共函数，根据公式求w，简易smo算法及完整smo算法通用
def calcWs(alphas, dataArr, labelArr):
w = sum(np.array(alphas) * np.array(labelArr.reshape((-1, 1))) * np.array(np.array(dataArr)))
return w
# 输入变量：x、y、c：常数c、toler：容错率、maxIter：最大循环次数
def smoSimple(dataMatIn, classLabels, C, toler, maxIter):
# dataMatIn, classLabels, C, toler, maxIter=dataArr,lableArr,0.6,0.001,40
dataMatrix = np.mat(dataMatIn) # 数据x转换为matrix类型
labelMat = np.mat(classLabels).transpose() # 标签y转换为matrix类型，转换为一列
b = 0 # 截距b
m, n = np.shape(dataMatrix) # 数据x行数、列数
alphas = np.mat(np.zeros((m, 1))) # 初始化alpha，有多少行数据就产生多少个alpha
iter = 0 # 遍历计数器
while (iter < maxIter):
# print( "iteration number: %d" % iter)
alphaPairsChanged = 0 # 记录alpha是否已被优化，每次循环都重置
for i in range(m): # 按行遍历数据，类似随机梯度下降
# i=0
fXi = float(np.multiply(alphas, labelMat).T * (
dataMatrix * dataMatrix[i, :].T)) + b # 预测值y，g(x)函数，《统计学习方法》李航P127，7.104
Ei = fXi - float(labelMat[i]) # if checks if an example violates KKT conditions # 误差，Ei函数，P127，7.105
if ((labelMat[i] * Ei < -toler) and (alphas[i] < C)) or ((labelMat[i] * Ei > toler) and (alphas[i] > 0)):
# 找第一个alphas[i]，找到第一个满足判断条件的，判断负间隔or正间隔，并且保证0<alphas<C
j = selectJrand(i, m) # 随机找到第二个alphas[j]
fXj = float(np.multiply(alphas, labelMat).T * (dataMatrix * dataMatrix[j, :].T)) + b # 计算预测值
Ej = fXj - float(labelMat[j]) # 计算alphas[j]误差
alphaIold = alphas[i].copy() # 记录上一次alphas[i]值
alphaJold = alphas[j].copy() # 记录上一次alphas[j]值
if (labelMat[i] != labelMat[j]): # 计算H及L值，《统计学习方法》李航，P126
L = max(0, alphas[j] - alphas[i])
H = min(C, C + alphas[j] - alphas[i])
else:
L = max(0, alphas[j] + alphas[i] - C)
H = min(C, alphas[j] + alphas[i])
if L == H:
# print( "L==H")
continue
eta = 2.0 * dataMatrix[i, :] * dataMatrix[j, :].T - dataMatrix[i, :] * dataMatrix[i, :].T - dataMatrix[
j,
:] * dataMatrix[
j, :].T
# 《统计学习方法》李航P127，7.107，这里的eta与李航的一致，这里乘了负号
if eta >= 0:
# print("eta>=0")
continue
alphas[j] -= labelMat[j] * (Ei - Ej) / eta # 《统计学习方法》李航P127，7.107，更新alphas[j]
alphas[j] = clipAlpha(alphas[j], H, L) # alphas[j]调整大于H或小于L的alpha值
if (abs(alphas[j] - alphaJold) < 0.00001): # 调整后过小，则不更新alphas[i]
# print( "j not moving enough")
continue
alphas[i] += labelMat[j] * labelMat[i] * (alphaJold - alphas[j]) # 更新alphas[i]，《统计学习方法》李航P127，7.109
# 更新b值，《统计学习方法》李航P130，7.115，7.116
b1 = b - Ei - labelMat[i] * (alphas[i] - alphaIold) * dataMatrix[i, :] * dataMatrix[i, :].T - labelMat[
j] * (alphas[j] - alphaJold) * dataMatrix[i, :] * dataMatrix[j, :].T
b2 = b - Ej - labelMat[i] * (alphas[i] - alphaIold) * dataMatrix[i, :] * dataMatrix[j, :].T - labelMat[
j] * (alphas[j] - alphaJold) * dataMatrix[j, :] * dataMatrix[j, :].T
if (0 < alphas[i]) and (C > alphas[i]): # 判断符合条件的b
b = b1
elif (0 < alphas[j]) and (C > alphas[j]):
b = b2
else:
b = (b1 + b2) / 2.0
alphaPairsChanged += 1
# print( "iter: %d i:%d, pairs changed %d" % (iter,i,alphaPairsChanged))
if (alphaPairsChanged == 0):
iter += 1
else:
iter = 0
return b, alphas
# 画图
def plot_smoSimple(dataArrWithAlpha, b, w):
type1_x1 = []
type1_x2 = []
type2_x1 = []
type2_x2 = []
dataSet = dataArrWithAlpha
# 取两类x1及x2值画图
type1_x1 = dataSet[dataSet[:, -2] == -1][:, :-2][:, 0].tolist()
type1_x2 = dataSet[dataSet[:, -2] == -1][:, :-2][:, 1].tolist()
type2_x1 = dataSet[dataSet[:, -2] == 1][:, :-2][:, 0].tolist()
type2_x2 = dataSet[dataSet[:, -2] == 1][:, :-2][:, 1].tolist()
# 画点
fig = plt.figure()
ax = fig.add_subplot(111)
ax.scatter(type1_x1, type1_x2, marker='s', s=90)
ax.scatter(type2_x1, type2_x2, marker='o', s=50, c='red')
plt.title('Support Vectors Circled')
# 获取支持向量值，画椭圆
dataVectors = dataArrWithAlpha[dataArrWithAlpha[:, -1] > 0]
for d in dataVectors:
circle = Circle(d[0:2], 0.5, facecolor='none', edgecolor=(0, 0.8, 0.8), linewidth=3, alpha=0.5)
ax.add_patch(circle)
# 画分割超平面
b = b1.getA()[0][0] # 获得传入的b
w0 = w[0] # 0.8065
w1 = w[1] # -0.2761
x = np.arange(-2.0, 12.0, 0.1)
y = (-w0 * x - b) / w1
ax.plot(x, y)
ax.axis([-2, 12, -8, 6])
plt.show()
# 简易smo算法结束####################################################################
# 完整smo算法开始####################################################################
class optStruct:
def __init__(self, dataMatIn, classLabels, C, toler, kTup): # Initialize the structure with the parameters
self.X = dataMatIn
self.labelMat = classLabels
self.C = C
self.tol = toler
self.m = np.shape(dataMatIn)[0]
self.alphas = np.mat(np.zeros((self.m, 1)))
self.b = 0
self.eCache = np.mat(np.zeros((self.m, 2))) # first column is valid flag
self.K = np.mat(np.zeros((self.m, self.m)))
for i in range(self.m):
self.K[:, i] = kernelTrans(self.X, self.X[i, :], kTup)
# 核函数
def kernelTrans(X, A, kTup): # calc the kernel or transform data to a higher dimensional space
m, n = np.shape(X)
K = np.mat(np.zeros((m, 1)))
if kTup[0] == 'lin':
K = X * A.T # linear kernel
elif kTup[0] == 'rbf':
for j in range(m):
deltaRow = X[j, :] - A
K[j] = deltaRow * deltaRow.T
K = np.exp(K / (-1 * kTup[1] ** 2)) # divide in NumPy is element-wise not matrix like Matlab
else:
raise NameError('Houston We Have a Problem -- \
That Kernel is not recognized')
return K
# 计算误差
def calcEk(oS, k):
fXk = float(np.multiply(oS.alphas, oS.labelMat).T * oS.K[:, k] + oS.b)
Ek = fXk - float(oS.labelMat[k])
return Ek
# 寻找第2个步长最大的alphas[j]
def selectJ(i, oS, Ei): # this is the second choice -heurstic, and calcs Ej
maxK = -1;
maxDeltaE = 0;
Ej = 0
oS.eCache[i] = [1, Ei] # set valid #choose the alpha that gives the maximum delta E
validEcacheList = np.nonzero(oS.eCache[:, 0].A)[0]
if (len(validEcacheList)) > 1:
for k in validEcacheList: # loop through valid Ecache values and find the one that maximizes delta E
if k == i: continue # don't calc for i, waste of time
Ek = calcEk(oS, k)
deltaE = abs(Ei - Ek)
if (deltaE > maxDeltaE):
maxK = k;
maxDeltaE = deltaE;
Ej = Ek
return maxK, Ej
else: # in this case (first time around) we don't have any valid eCache values
j = selectJrand(i, oS.m)
Ej = calcEk(oS, j)
return j, Ej
# 计算误差存入缓存中
def updateEk(oS, k): # after any alpha has changed update the new value in the cache
Ek = calcEk(oS, k)
oS.eCache[k] = [1, Ek]
# 内循环，寻找第2个步长最大的alphas[j]
def innerL(i, oS):
Ei = calcEk(oS, i)
if ((oS.labelMat[i] * Ei < -oS.tol) and (oS.alphas[i] < oS.C)) or (
(oS.labelMat[i] * Ei > oS.tol) and (oS.alphas[i] > 0)):
j, Ej = selectJ(i, oS, Ei) # this has been changed from selectJrand
alphaIold = oS.alphas[i].copy();
alphaJold = oS.alphas[j].copy();
if (oS.labelMat[i] != oS.labelMat[j]):
L = max(0, oS.alphas[j] - oS.alphas[i])
H = min(oS.C, oS.C + oS.alphas[j] - oS.alphas[i])
else:
L = max(0, oS.alphas[j] + oS.alphas[i] - oS.C)
H = min(oS.C, oS.alphas[j] + oS.alphas[i])
if L == H:
# print ("L==H")
return 0
eta = 2.0 * oS.K[i, j] - oS.K[i, i] - oS.K[j, j] # changed for kernel
if eta >= 0:
# print( "eta>=0")
return 0
oS.alphas[j] -= oS.labelMat[j] * (Ei - Ej) / eta
oS.alphas[j] = clipAlpha(oS.alphas[j], H, L)
updateEk(oS, j) # added this for the Ecache
if (abs(oS.alphas[j] - alphaJold) < 0.00001):
# print( "j not moving enough")
return 0
oS.alphas[i] += oS.labelMat[j] * oS.labelMat[i] * (alphaJold - oS.alphas[j]) # update i by the same amount as j
updateEk(oS, i) # added this for the Ecache #the update is in the oppostie direction
b1 = oS.b - Ei - oS.labelMat[i] * (oS.alphas[i] - alphaIold) * oS.K[i, i] - oS.labelMat[j] * (
oS.alphas[j] - alphaJold) * oS.K[i, j]
b2 = oS.b - Ej - oS.labelMat[i] * (oS.alphas[i] - alphaIold) * oS.K[i, j] - oS.labelMat[j] * (
oS.alphas[j] - alphaJold) * oS.K[j, j]
if (0 < oS.alphas[i]) and (oS.C > oS.alphas[i]):
oS.b = b1
elif (0 < oS.alphas[j]) and (oS.C > oS.alphas[j]):
oS.b = b2
else:
oS.b = (b1 + b2) / 2.0
return 1
else:
return 0
# SMO主函数
def smoP(dataMatIn, classLabels, C, toler, maxIter, kTup=('lin', 0)): # full Platt SMO
oS = optStruct(np.mat(dataMatIn), np.mat(classLabels).transpose(), C, toler, kTup)
iter = 0
entireSet = True;
alphaPairsChanged = 0
while (iter < maxIter) and ((alphaPairsChanged > 0) or (entireSet)):
alphaPairsChanged = 0
if entireSet: # go over all
for i in range(oS.m):
alphaPairsChanged += innerL(i, oS)
# print( "fullSet, iter: %d i:%d, pairs changed %d" % (iter,i,alphaPairsChanged))
iter += 1
else: # go over non-bound (railed) alphas
nonBoundIs = np.nonzero((oS.alphas.A > 0) * (oS.alphas.A < C))[0]
for i in nonBoundIs:
alphaPairsChanged += innerL(i, oS)
# print( "non-bound, iter: %d i:%d, pairs changed %d" % (iter,i,alphaPairsChanged))
iter += 1
if entireSet:
entireSet = False # toggle entire set loop
elif (alphaPairsChanged == 0):
entireSet = True
# print( "iteration number: %d" % iter)
return oS.b, oS.alphas
# 测试Rbf数据
def testRbf(dataArrTrain, labelArrTrain, dataArrTest, labelArrTest, k1=1.3):
b, alphas = smoP(dataArrTrain, labelArrTrain, 200, 0.0001, 10000, ('rbf', k1)) # C=200 important
datMat = np.mat(dataArrTrain);
labelArrTrain = np.mat(labelArrTrain).transpose()
svInd = np.nonzero(alphas.A > 0)[0]
sVs = datMat[svInd] # get matrix of only support vectors
labelSV = labelArrTrain[svInd];
# print( "there are %d Support Vectors" % np.shape(sVs)[0])
m, n = np.shape(datMat)
errorCount = 0
for i in range(m):
kernelEval = kernelTrans(sVs, datMat[i, :], ('rbf', k1))
predict = kernelEval.T * np.multiply(labelSV, alphas[svInd]) + b
if np.sign(predict) != np.sign(labelArrTrain[i]): errorCount += 1
print("the training error rate is: %f" % (float(errorCount) / m))
errorCount = 0
datMat = np.mat(dataArrTest)
# labelMat = np.mat(labelArrTest).transpose()
m, n = np.shape(datMat)
for i in range(m):
kernelEval = kernelTrans(sVs, datMat[i, :], ('rbf', k1))
predict = kernelEval.T * np.multiply(labelSV, alphas[svInd]) + b
if np.sign(predict) != np.sign(labelArrTest[i]): errorCount += 1
print("the test error rate is: %f" % (float(errorCount) / m))
return b, alphas
# 画图，rbf核函数数据
def plot_smoCompletion():
xcord0 = [];
ycord0 = [];
xcord1 = [];
ycord1 = []
fw = open('D:/python_data/testSetRBF2.txt', 'w') # generate data
fig = plt.figure()
ax = fig.add_subplot(111)
xcord0 = [];
ycord0 = [];
xcord1 = [];
ycord1 = []
for i in range(100):
[x, y] = np.random.uniform(0, 1, 2)
xpt = x * np.cos(2.0 * np.pi * y);
ypt = x * np.sin(2.0 * np.pi * y)
if (x > 0.5):
xcord0.append(xpt);
ycord0.append(ypt)
label = -1.0
else:
xcord1.append(xpt);
ycord1.append(ypt)
label = 1.0
fw.write('%f\t%f\t%f\n' % (xpt, ypt, label))
ax.scatter(xcord0, ycord0, marker='s', s=90)
ax.scatter(xcord1, ycord1, marker='o', s=50, c='red')
plt.title('Non-linearly Separable Data for Kernel Method')
plt.show()
fw.close()
# 完整smo算法结束####################################################################
if __name__ == '__main__':
##1、SVM直接求参数值#############################################################
print('\n1、SVM直接求参数值，开始')
# 生成数据，《统计学习方法》李航，P103，例7.1
dataSet = np.array([[3, 3, 1], [4, 3, 1], [1, 1, -1]]) # ,[0,0,-1],[0,1,-1]
m, n = dataSet.shape
x = dataSet[:, :-1]
y = dataSet[:, -1] # .reshape((-1,1))
# 数据定义
X = np.append(np.ones([x.shape[0], 1]), x, 1) # x新增一列全1值，作为截距b
Y = y
c = 0.001 # 学习率
w = np.zeros(X.shape[1]) # 初始化一组w系数，全0，也可随机产生：np.random.rand(X.shape[1])
# bfgs_b方法求最优化问题
REF = fmin_l_bfgs_b(func, x0=w, args=(X, Y, c), approx_grad=False) # x0=np.random.rand(X.shape[1]) [0,0,0]
# 采用scipy.optimize其他包夜可以求得
REF2 = fmin_tnc(func, x0=w, args=(X, Y, c), approx_grad=False)
# 求得最优化计算后的w
w = REF[0].round(2) # 取得w值
print('w:', w[1:], 'b:', w[0]) # 与《统计学习方法》李航，P103，例7.1计算结果一致
# 画图
plotResult(w)
print('\n1、SVM直接求参数值，结束')
##2、SVM简易SMO算法#############################################################
print('\n2、SVM简易SMO算法，开始')
fileIn = 'D:/python_data/testSet.txt'
# dataSet=pd.read_table(fileIn,names=['x1','x2','y']).values
dataSet = np.loadtxt(fileIn)
dataArr = dataSet[:, :-1] # x
labelArr = dataSet[:, -1] # y
b1, alphas1 = smoSimple(dataArr, labelArr, 0.6, 0.001, 50) # 输入变量：x、y、c：常数c、toler：容错率、maxIter：最大循环次数
dataArrWithAlpha1 = np.array(np.concatenate((dataSet, alphas1), axis=1)) # 把alphas1与原始数据合并
w1 = calcWs(alphas1, dataArr, labelArr) # 根据alpha求w
print('b:', b1, '\nw:', w1, '\ndata，alphas，支撑向量:\n', dataArrWithAlpha1[
dataArrWithAlpha1[:, -1] > 0]) # 注意这里的筛选方式与pd.DataFrame筛选方式一致，array类型的才可以这样写，np.ndarray及np.matrix类型不可以使用
plot_smoSimple(dataArrWithAlpha1, b1, w1) # 画图
print('2、SVM简易SMO算法，结束')
##3、SVM完整SMO算法#############################################################
print('\n3、SVM完整SMO算法，开始')
dataSetTrain = np.loadtxt('D:/python_data/testSetRBF.txt')
dataSetTest = np.loadtxt('D:/python_data/testSetRBF2.txt')
# 训练集
dataArrTrain = dataSetTrain[:, :-1] # 训练集x
labelArrTrain = dataSetTrain[:, -1] # 训练集y
# 测试集
dataArrTest = dataSetTest[:, :-1] # 测试集x
labelArrTest = dataSetTest[:, -1] # 测试集y
# 调用主函数
b2, alphas2 = testRbf(dataArrTrain, labelArrTrain, dataArrTest, labelArrTest, k1=1.3)
w2 = calcWs(alphas2, dataArrTrain, labelArrTrain) # 根据alpha求w
dataArrWithAlpha2 = np.array(np.concatenate((dataSetTrain, alphas2), axis=1)) # 把alphas1与原始数据合并
print('b:', b1, '\nw:', w1, '\ndata，alphas，支撑向量:\n', dataArrWithAlpha2[
dataArrWithAlpha2[:, -1] > 0]) # 注意这里的筛选方式与pd.DataFrame筛选方式一致，array类型的才可以这样写，np.ndarray及np.matrix类型不可以使用
plot_smoCompletion() # 画图，训练集
扫一扫
751
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
