我在2012版的matlab里安装上了libsvm工具包，还要安装python吗？

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

    我安装了libsvm工具包，看到有人用python软件结合libsvm一起使用。我是想用matlab结合SVM来做数据预测，不知道还用不用安装python软件.python软件是怎么使用我也不懂,麻烦大家给个方法和建议。

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：MATLAB python Libsvm atlab matla 工具包 安装 软件 python matlab

你说的这个是Python来实现支持向量机问题，Python有机器学习的实现办法
比如——
  
from numpy import *  
import time  
import matplotlib.pyplot as plt   
  
# calulate kernel value  
def calcKernelValue(matrix_x, sample_x, kernelOption):  
    kernelType = kernelOption[0]  
    numSamples = matrix_x.shape[0]  
    kernelValue = mat(zeros((numSamples, 1)))        
    if kernelType == 'linear':  
        kernelValue = matrix_x * sample_x.T  
    elif kernelType == 'rbf':  
        sigma = kernelOption[1]  
        if sigma == 0:  
            sigma = 1.0  
        for i in xrange(numSamples):  
            diff = matrix_x[i, :] - sample_x  
            kernelValue = exp(diff * diff.T / (-2.0 * sigma**2))  
    else:  
        raise NameError('Not support kernel type! You can use linear or rbf!')  
    return kernelValue  
  
# calculate kernel matrix given train set and kernel type  
def calcKernelMatrix(train_x, kernelOption):  
    numSamples = train_x.shape[0]  
    kernelMatrix = mat(zeros((numSamples, numSamples)))  
    for i in xrange(numSamples):  
        kernelMatrix[:, i] = calcKernelValue(train_x, train_x[i, :], kernelOption)  
    return kernelMatrix  
  
# define a struct just for storing variables and data  
class SVMStruct:  
    def __init__(self, dataSet, labels, C, toler, kernelOption):  
        self.train_x = dataSet # each row stands for a sample  
        self.train_y = labels  # corresponding label  
        self.C = C             # slack variable  
        self.toler = toler     # termination condition for iteration  
        self.numSamples = dataSet.shape[0] # number of samples  
        self.alphas = mat(zeros((self.numSamples, 1))) # Lagrange factors for all samples  
        self.b = 0  
        self.errorCache = mat(zeros((self.numSamples, 2)))  
        self.kernelOpt = kernelOption  
        self.kernelMat = calcKernelMatrix(self.train_x, self.kernelOpt)  
         
# calculate the error for alpha k  
def calcError(svm, alpha_k):  
    output_k = float(multiply(svm.alphas, svm.train_y).T * svm.kernelMat[:, alpha_k] + svm.b)  
    error_k = output_k - float(svm.train_y[alpha_k])  
    return error_k  
  
# update the error cache for alpha k after optimize alpha k  
def updateError(svm, alpha_k):  
    error = calcError(svm, alpha_k)  
    svm.errorCache[alpha_k] = [1, error]  
  
  
# select alpha j which has the biggest step  
def selectAlpha_j(svm, alpha_i, error_i):  
    svm.errorCache[alpha_i] = [1, error_i] # mark as valid(has been optimized)  
    candidateAlphaList = nonzero(svm.errorCache[:, 0].A)[0] # mat.A return array  
    maxStep = 0; alpha_j = 0; error_j = 0  

    # find the alpha with max iterative step  
    if len(candidateAlphaList) > 1:  
        for alpha_k in candidateAlphaList:  
            if alpha_k == alpha_i:   
                continue  
            error_k = calcError(svm, alpha_k)  
            if abs(error_k - error_i) > maxStep:  
                maxStep = abs(error_k - error_i)  
                alpha_j = alpha_k  
                error_j = error_k  
    # if came in this loop first time, we select alpha j randomly  
    else:            
        alpha_j = alpha_i  
        while alpha_j == alpha_i:  
            alpha_j = int(random.uniform(0, svm.numSamples))  
        error_j = calcError(svm, alpha_j)       
    return alpha_j, error_j  
  
# the inner loop for optimizing alpha i and alpha j  
def innerLoop(svm, alpha_i):  
    error_i = calcError(svm, alpha_i)  
    ### check and pick up the alpha who violates the KKT condition  
    ## satisfy KKT condition  
    # 1) yi*f(i) >= 1 and alpha == 0 (outside the boundary)  
    # 2) yi*f(i) == 1 and 0<alpha< C (on the boundary)  
    # 3) yi*f(i) <= 1 and alpha == C (between the boundary)  
    ## violate KKT condition  
    # because y*E_i = y*f(i) - y^2 = y*f(i) - 1, so  
    # 1) if y*E_i < 0, so yi*f(i) < 1, if alpha < C, violate!(alpha = C will be correct)   
    # 2) if y*E_i > 0, so yi*f(i) > 1, if alpha > 0, violate!(alpha = 0 will be correct)  
    # 3) if y*E_i = 0, so yi*f(i) = 1, it is on the boundary, needless optimized  
    if (svm.train_y[alpha_i] * error_i < -svm.toler) and (svm.alphas[alpha_i] < svm.C) or\  
        (svm.train_y[alpha_i] * error_i > svm.toler) and (svm.alphas[alpha_i] > 0):  
        # step 1: select alpha j  
        alpha_j, error_j = selectAlpha_j(svm, alpha_i, error_i)  
        alpha_i_old = svm.alphas[alpha_i].copy()  
        alpha_j_old = svm.alphas[alpha_j].copy()  
        # step 2: calculate the boundary L and H for alpha j  
        if svm.train_y[alpha_i] != svm.train_y[alpha_j]:  
            L = max(0, svm.alphas[alpha_j] - svm.alphas[alpha_i])  
            H = min(svm.C, svm.C + svm.alphas[alpha_j] - svm.alphas[alpha_i])  
        else:  
            L = max(0, svm.alphas[alpha_j] + svm.alphas[alpha_i] - svm.C)  
            H = min(svm.C, svm.alphas[alpha_j] + svm.alphas[alpha_i])  
        if L == H:  
            return 0  
        # step 3: calculate eta (the similarity of sample i and j)  
        eta = 2.0 * svm.kernelMat[alpha_i, alpha_j] - svm.kernelMat[alpha_i, alpha_i] \  
                  - svm.kernelMat[alpha_j, alpha_j]  
        if eta >= 0:  
            return 0  
        # step 4: update alpha j  
        svm.alphas[alpha_j] -= svm.train_y[alpha_j] * (error_i - error_j) / eta  
        # step 5: clip alpha j  
        if svm.alphas[alpha_j] > H:  
            svm.alphas[alpha_j] = H  
        if svm.alphas[alpha_j] < L:  
            svm.alphas[alpha_j] = L  
        # step 6: if alpha j not moving enough, just return      
        if abs(alpha_j_old - svm.alphas[alpha_j]) < 0.00001:  
            updateError(svm, alpha_j)  
            return 0  
        # step 7: update alpha i after optimizing aipha j  
        svm.alphas[alpha_i] += svm.train_y[alpha_i] * svm.train_y[alpha_j] \  
                                * (alpha_j_old - svm.alphas[alpha_j])  
        # step 8: update threshold b  
        b1 = svm.b - error_i - svm.train_y[alpha_i] * (svm.alphas[alpha_i] - alpha_i_old) \  
                                                    * svm.kernelMat[alpha_i, alpha_i] \  
                             - svm.train_y[alpha_j] * (svm.alphas[alpha_j] - alpha_j_old) \  
                                                    * svm.kernelMat[alpha_i, alpha_j]  
        b2 = svm.b - error_j - svm.train_y[alpha_i] * (svm.alphas[alpha_i] - alpha_i_old) \  
                                                    * svm.kernelMat[alpha_i, alpha_j] \  
                             - svm.train_y[alpha_j] * (svm.alphas[alpha_j] - alpha_j_old) \  
                                                    * svm.kernelMat[alpha_j, alpha_j]  
        if (0 < svm.alphas[alpha_i]) and (svm.alphas[alpha_i] < svm.C):  
            svm.b = b1  
        elif (0 < svm.alphas[alpha_j]) and (svm.alphas[alpha_j] < svm.C):  
            svm.b = b2  
        else:  
            svm.b = (b1 + b2) / 2.0  
        # step 9: update error cache for alpha i, j after optimize alpha i, j and b  
        updateError(svm, alpha_j)  
        updateError(svm, alpha_i)  
        return 1  
    else:  
        return 0  
  

# the main training procedure  
def trainSVM(train_x, train_y, C, toler, maxIter, kernelOption = ('rbf', 1.0)):  
    # calculate training time  
    startTime = time.time()  
    # init data struct for svm  
    svm = SVMStruct(mat(train_x), mat(train_y), C, toler, kernelOption)     
    # start training  
    entireSet = True  
    alphaPairsChanged = 0  
    iterCount = 0  
    # Iteration termination condition:  
    #   Condition 1: reach max iteration  
    #   Condition 2: no alpha changed after going through all samples,  
    #                in other words, all alpha (samples) fit KKT condition  
    while (iterCount < maxIter) and ((alphaPairsChanged > 0) or entireSet):  
        alphaPairsChanged = 0  
        # update alphas over all training examples  
        if entireSet:  
            for i in xrange(svm.numSamples):  
                alphaPairsChanged += innerLoop(svm, i)  
            print '---iter:%d entire set, alpha pairs changed:%d' % (iterCount, alphaPairsChanged)  
            iterCount += 1  
        # update alphas over examples where alpha is not 0 & not C (not on boundary)  
        else:  
            nonBoundAlphasList = nonzero((svm.alphas.A > 0) * (svm.alphas.A < svm.C))[0]  
            for i in nonBoundAlphasList:  
                alphaPairsChanged += innerLoop(svm, i)  
            print '---iter:%d non boundary, alpha pairs changed:%d' % (iterCount, alphaPairsChanged)  
            iterCount += 1  
        # alternate loop over all examples and non-boundary examples  
        if entireSet:  
            entireSet = False  
        elif alphaPairsChanged == 0:  
            entireSet = True  
    print 'Congratulations, training complete! Took %fs!' % (time.time() - startTime)  
    return svm  
# testing your trained svm model given test set  
def testSVM(svm, test_x, test_y):  
    test_x = mat(test_x)  
    test_y = mat(test_y)  
    numTestSamples = test_x.shape[0]  
    supportVectorsIndex = nonzero(svm.alphas.A > 0)[0]  
    supportVectors      = svm.train_x[supportVectorsIndex]  
    supportVectorLabels = svm.train_y[supportVectorsIndex]  
    supportVectorAlphas = svm.alphas[supportVectorsIndex]  
    matchCount = 0  
    for i in xrange(numTestSamples):  
        kernelValue = calcKernelValue(supportVectors, test_x[i, :], svm.kernelOpt)  
        predict = kernelValue.T * multiply(supportVectorLabels, supportVectorAlphas) + svm.b  
        if sign(predict) == sign(test_y):  
            matchCount += 1  
    accuracy = float(matchCount) / numTestSamples  
    return accuracy  
  
# show your trained svm model only available with 2-D data  
def showSVM(svm):  
    if svm.train_x.shape[1] != 2:  
        print "Sorry! I can not draw because the dimension of your data is not 2!"  
        return 1  
    # draw all samples  
    for i in xrange(svm.numSamples):  
        if svm.train_y == -1:  
            plt.plot(svm.train_x[i, 0], svm.train_x[i, 1], 'or')  
        elif svm.train_y == 1:  
            plt.plot(svm.train_x[i, 0], svm.train_x[i, 1], 'ob')  
    # mark support vectors  
    supportVectorsIndex = nonzero(svm.alphas.A > 0)[0]  
    for i in supportVectorsIndex:  
        plt.plot(svm.train_x[i, 0], svm.train_x[i, 1], 'oy')  
    # draw the classify line  
    w = zeros((2, 1))  
    for i in supportVectorsIndex:  
        w += multiply(svm.alphas * svm.train_y, svm.train_x[i, :].T)   
    min_x = min(svm.train_x[:, 0])[0, 0]  
    max_x = max(svm.train_x[:, 0])[0, 0]  
    y_min_x = float(-svm.b - w[0] * min_x) / w[1]  
    y_max_x = float(-svm.b - w[0] * max_x) / w[1]  
    plt.plot([min_x, max_x], [y_min_x, y_max_x], '-g')  
    plt.show()  

可以贴进去直接跑下试试，我跑过了，木有问题哒～  

sklearn。。。找个文档读读就行，急用的话没必要造轮子