in addition if you want a way to project new vectors into a reduced svd space:

l=2 # desired vector space dimension

U,d,Vt = linalg.svd(X,full_matrices=False)

P=(U*(1/d))[:,:l]

xnew = dot(x,P)