奇异值分解(Singular Value Decomposition,以下简称SVD)是在机器学习领域广泛应用的矩阵分解算法,这里对SVD原理 应用和代码实现做一个总结。

1 实对称方阵的矩阵分解

对于一个(n imes n)实对称方阵(A),如果存在一个向量(v)是矩阵(A)的特征向量,可以表示成下面的形式:(Av=lambda v)
其中,(lambda)是特征向量(v)对应的特征值。如果矩阵(A)(n)个线性无关的特征向量,那么矩阵(A)可以分解为: (A=QSigma Q^{-1})
其中,(Q)是矩阵(A)的特征向量组成的(n imes n)的方阵,(Sigma)是对角矩阵,每一个对角线元素就是一个特征值。注意到,特征值分解是有局限的,这里的矩阵是方阵,实际应用中,常见矩阵并不全是方阵。下面介绍SVD,SVD可以对任意矩阵进行分解。

2 奇异值分解(SVD)

假设矩阵(A)(m imes n)的矩阵,则(AA^{T})(m imes m)的方阵,(A^{T}A)(n imes n)的方阵,对这两个方阵矩阵分解:(AA^{T}=ULambda _{1}U^{T}) (A^{T}A=VLambda _{2}V^{T})
其中,(Lambda _{1})(Lambda _{2})是对角矩阵,且对角线上非零元素均相同,也就是说两个方阵有相同的非零特征值,令非零特征值为(sigma _{1},sigma _{2},cdots ,sigma _{k}),注意,(kleq m)(kleq n)。根据(sigma _{1},sigma _{2},cdots ,sigma _{k})可以得到矩阵(A)的特征值为:(lambda _{1}=sqrt{sigma _{1}},lambda _{2}=sqrt{sigma _{2}},cdots ,lambda _{k}=sqrt{sigma _{k}})
下面就可以得到奇异值分解的表达式为: $$A=ULambda V^{T}$$
其中,(U)是一个(m)( imes m)的矩阵,(Sigma)(m imes n)的矩阵,除了主对角线上的元素外其余全为0,主对角线上的每个元素称为奇异值,(V)是一个(n imes n)的矩阵。(U)(V)都是酉矩阵,满足(U^{T}U=I)(V^{T}V=I)

3 SVD代码实现

SVD

>>> from numpy import *
>>> U,Sigma,VT=linalg.svd([[1,1],[7,7]])
>>> U
array([[-0.14142136, -0.98994949],
       [-0.98994949,  0.14142136]])
>>> Sigma
array([10.,  0.])
>>> VT
array([[-0.70710678, -0.70710678],
       [-0.70710678,  0.70710678]])

在一个更大的数据集上进行更多的分解

大数据集SVD

def loadExData():
    return [[0, 0, 0, 2, 2],
            [0, 0, 0, 3, 3],
            [0, 0, 0, 1, 1],
            [1, 1, 1, 0, 0],
            [2, 2, 2, 0, 0],
            [5, 5, 5, 0, 0],
            [1, 1, 1, 0, 0]]
控制台运行效果

>>> import svdRec
>>> import imp
>>> imp.reload(svdRec)

>>> Data=svdRec.loadExData()
>>> from numpy import *
>>> U,Sigma,VT=linalg.svd(Data)
>>> Sigma
array([9.64365076e+00, 5.29150262e+00, 7.40623935e-16, 4.05103551e-16,
       2.21838243e-32])

重构原始矩阵的近似矩阵

重构原始矩阵

>>> Sig3=mat([[Sigma[0],0,0],[0,Sigma[1],0],[0,0,Sigma[2]]])
>>> U[:,:3]*Sig3*VT[:3,:]
matrix([[ 5.03302006e-17,  1.95279569e-15,  1.70575023e-15,
          2.00000000e+00,  2.00000000e+00],
        [-7.69233911e-16,  3.14619452e-16,  4.54614459e-16,
          3.00000000e+00,  3.00000000e+00],
        [-2.02143152e-16,  6.40186235e-17,  1.38124528e-16,
          1.00000000e+00,  1.00000000e+00],
        [ 1.00000000e+00,  1.00000000e+00,  1.00000000e+00,
         -1.52065993e-33, -1.21652794e-33],
        [ 2.00000000e+00,  2.00000000e+00,  2.00000000e+00,
         -3.04131986e-33, -2.43305589e-33],
        [ 5.00000000e+00,  5.00000000e+00,  5.00000000e+00,
          1.82479192e-33,  1.45983353e-33],
        [ 1.00000000e+00,  1.00000000e+00,  1.00000000e+00,
         -1.52065993e-33, -1.21652794e-33]])

相似度计算

from numpy import *
from numpy import linalg as la

#欧式
def ecludSim(inA,inB):  #假定inA inB都是列向量
    return 1.0/(1.0+la.norm(inA - inB))

#皮尔逊相关系数
def pearsSim(inA, inB):
    if len(inA) < 3: return 1.0  #检查是否存在3个或更多的点 如果不存在返回1.0 此时两个向量完全相关
    return 0.5+0.5*corrcoef(inA,inB,rowvar=0)[0][1]

#余弦相似度
def cosSim(inA, inB):
    num = float(inA.T*inB)
    denom = la.norm(inA)*la.norm(inB)
    return 0.5+0.5*(num/denom)

测试相似度计算效果:

测试相似度

>>> imp.reload(svdRec)

>>> from numpy import *
>>> myMat = mat(svdRec.loadExData())
>>> svdRec.ecludSim(myMat[:,0],myMat[:,4])
0.12973190755680383
>>> svdRec.ecludSim(myMat[:,0],myMat[:,0])
1.0
>>> svdRec.cosSim(myMat[:,0],myMat[:,4])
0.5
>>> svdRec.cosSim(myMat[:,0],myMat[:,0])
1.0
>>> svdRec.pearsSim(myMat[:,0],myMat[:,4])
0.20596538173840329
>>> svdRec.pearsSim(myMat[:,0],myMat[:,0])
1.0

3.1 基于物品相似度的推荐引擎

推荐引擎

def standEst(dataMat, user, simMeas, item): #用来计算在给定相似度计算方法的条件下 用户对物品的估计评分值
    n = shape(dataMat)[1]
    simTotal = 0.0; ratSimTotal = 0.0
    for j in range(n):
        userRating = dataMat[user,j]
        if userRating == 0: continue
        overLap = nonzero(logical_and(dataMat[:, item].A>0, dataMat[:,j].A>0))[0]  #寻找两个用户都评级的物品
        if len(overLap) == 0: similarity = 0
        else: similarity = simMeas(dataMat[overLap,item], dataMat[overLap,j])
        simTotal += similarity
        ratSimTotal += similarity*userRating
    if simTotal == 0: return 0
    else: return ratSimTotal/simTotal

def recommend(dataMat, user, N=3, simMeas=cosSim, estMethod=standEst): #推荐引擎
unratedItems = nonzero(dataMat[user,:].A==0)[1] #寻找未评级的物品
if len(unratedItems) == 0: return 'you rated everything'
itemScores = []
for item in unratedItems:
estimatedScore = estMethod(dataMat, user, simMeas, item)
itemScores.append((item, estimatedScore))
return sorted(itemScores, key=lambda jj: jj[1], reverse=True)[:N] #寻找前N个未评级物品

控制台运行效果

>>> imp.reload(svdRec)

>>> myMat=mat(svdRec.loadExData())
>>> myMat[0,1]=myMat[0,0]=myMat[1,0]=myMat[2,0]=4
>>> myMat[3,3]=2
>>> myMat
matrix([[4, 4, 0, 2, 2],
        [4, 0, 0, 3, 3],
        [4, 0, 0, 1, 1],
        [1, 1, 1, 2, 0],
        [2, 2, 2, 0, 0],
        [5, 5, 5, 0, 0],
        [1, 1, 1, 0, 0]])
>>> svdRec.recommend(myMat,2)
[(2, 2.5), (1, 2.0243290220056256)]
>>> svdRec.recommend(myMat,2,simMeas=svdRec.ecludSim)
[(2, 3.0), (1, 2.8266504712098603)]
>>> svdRec.recommend(myMat,2,simMeas=svdRec.pearsSim)
[(2, 2.5), (1, 2.0)]

利用SVD提高推荐效果,实际的数据集会比用于展示recommend()函数功能的myMat矩阵稀疏得多

新数据集

def loadExData2():
    return[[0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 5],
           [0, 0, 0, 3, 0, 4, 0, 0, 0, 0, 3],
           [0, 0, 0, 0, 4, 0, 0, 1, 0, 4, 0],
           [3, 3, 4, 0, 0, 0, 0, 2, 2, 0, 0],
           [5, 4, 5, 0, 0, 0, 0, 5, 5, 0, 0],
           [0, 0, 0, 0, 5, 0, 1, 0, 0, 5, 0],
           [4, 3, 4, 0, 0, 0, 0, 5, 5, 0, 1],
           [0, 0, 0, 4, 0, 4, 0, 0, 0, 0, 4],
           [0, 0, 0, 2, 0, 2, 5, 0, 0, 1, 2],
           [0, 0, 0, 0, 5, 0, 0, 0, 0, 4, 0],
           [1, 0, 0, 0, 0, 0, 0, 1, 2, 0, 0]]
控制台运行效果

>>> from numpy import *
>>> from numpy import linalg as la
>>> imp.reload(svdRec)

>>> U,Sigma,VT=la.svd(mat(svdRec.loadExData2()))
>>> Sigma
array([15.77075346, 11.40670395, 11.03044558,  4.84639758,  3.09292055,
        2.58097379,  1.00413543,  0.72817072,  0.43800353,  0.22082113,
        0.07367823])
>>> Sig2=Sigma**2   #Sigma平方
>>> sum(Sig2)   #计算总能量
541.9999999999995
>>> sum(Sig2)*0.9  #计算总能量的90%
487.7999999999996
>>> sum(Sig2[:2])  #计算前两个元素所包含的能量
378.8295595113579
>>> sum(Sig2[:3]) #计算前三个元素所包含的能量
500.50028912757926

基于SVD评分估计

基于SVD评分估计

def svdEst(dataMat, user, simMeas, item):
    n = shape(dataMat)[1]
    simTotal = 0.0; ratSimTotal = 0.0
    U,Sigma,VT = la.svd(dataMat)
    Sig4 = mat(eye(4)*Sigma[:4])  #建立对角矩阵
    xformedItems = dataMat.T*U[:,:4]*Sig4  #构建转换后的物品
    for j in range(n):
        userRating = dataMat[user, j]
        if userRating == 0 or j == item: continue
        similarity = simMeas(xformedItems[item,:].T,xformedItems[j,:].T)
        print('the %d and %d similarity is: %f' % (item, j, similarity))
        simTotal += similarity
        ratSimTotal += similarity*userRating
    if simTotal == 0: return 0
    else: return ratSimTotal/simTotal
控制台运行效果

>>> imp.reload(svdRec)

>>> svdRec.recommend(myMat, 1, estMethod=svdRec.svdEst)
the 1 and 0 similarity is: 0.977045
the 1 and 3 similarity is: 0.881104
the 1 and 4 similarity is: 0.867220
the 2 and 0 similarity is: 0.943303
the 2 and 3 similarity is: 0.813362
the 2 and 4 similarity is: 0.795492
[(2, 3.369610179675583), (1, 3.3584999687218007)]
>>> svdRec.recommend(myMat, 1, estMethod=svdRec.svdEst, simMeas=svdRec.pearsSim)
the 1 and 0 similarity is: 0.985932
the 1 and 3 similarity is: 0.883061
the 1 and 4 similarity is: 0.857259
the 2 and 0 similarity is: 0.948792
the 2 and 3 similarity is: 0.795831
the 2 and 4 similarity is: 0.766698
[(2, 3.377805913868774), (1, 3.3616436918254204)]

基于SVD的图像压缩

基于SVD的图像压缩

def printMat(inMat, thresh=0.8):
    for i in range(32):
        for k in range(32):
            if float(inMat[i,k]) > thresh:
                print(1)
            else: print (0)
        print (' ')

def imgCompress(numSV=3, thresh=0.8):
myl = []
for line in open('0_5.txt').readlines():
newRow = []
for i in range(32):
newRow.append(int(line[i]))
myl.append(newRow)
myMat = mat(myl)
print("original matrix")
printMat(myMat, thresh)
U,Sigma,VT = la.svd(myMat)
SigRecon = mat(zeros((numSV, numSV)))
for k in range(numSV):#construct diagonal matrix from vector
SigRecon[k,k] = Sigma[k]
reconMat = U[:,:numSV]
SigRecon
VT[:numSV,:]
print("reconstructed matrix using %d singular values**" % numSV)
printMat(reconMat, thresh)

控制台运行效果

>>> imp.reload(svdRec)

>>> svdRec.imgCompress(2)

参考:机器学习实战