我正在尝试使用以下代码在

Python中遵循

Wikipedia Article on latent semantic indexing：

documentTermMatrix = array([[ 0.,1.,0.,1.],[ 0.,0.],[ 1.,0.]])

u,s,vt = linalg.svd(documentTermMatrix,full_matrices=False)

sigma = diag(s)

## remove extra dimensions...

numberOfDimensions = 4

for i in range(4,len(sigma) -1):

sigma[i][i] = 0

queryVector = array([[ 0.],# same as first column in documentTermMatrix

[ 0.],[ 0.],[ 1.],[ 1.]])

数学怎么说应该有效：

dtMatrixToQueryAgainst = dot(u,dot(s,vt))

queryVector = dot(inv(s),dot(transpose(u),queryVector))

similarityToFirst = cosineDistance(queryVector,dtMatrixToQueryAgainst[:,0]

# gives 'matrices are not aligned' error. should be 1 because they're the same

什么工作,数学看起来不正确:(从here)

dtMatrixToQueryAgainst = dot(s,vt)

queryVector = dot(transpose(u),queryVector)

similarityToFirst = cosineDistance(queryVector,dtMatrixToQueryAgainsst[:,0])

# gives 1,which is correct

为什么路由工作,而第一个没有,当我能找到关于LSA数学的所有东西显示第一个是正确的？我觉得我错过了一些明显的东西……