如何使用 Python （可能还有 Scipy）估计一个巨大的、稀疏的 csr_matrix 的等级？

Question

I have a huge, sparse matrix in the type of scipy.sparse.csr.csr_matrix that I need to estimate its rank. I find this on scipy.org that seems perfect for this job, but it doesn't support csr_matrix .

from scipy.sparse import load_npz
from scipy.linalg.interpolative import estimate_rank

X = load_npz("https://drive.google.com/uc?export=download&id=1SSR6JWEqG4DXRU9qo78682D9pGJF3Wr0")
print("Rank:", estimate_rank(X, eps=100))

TypeError: invalid input type (must be array or LinearOperator)

The sparse matrix has over 50K rows and nearly 40K columns. Converting it to a numpy array first seems pointless. What should I do to make it work?

---

The following doesn't work either.

from scipy.sparse import load_npz, linalg
from scipy.linalg.interpolative import estimate_rank

X = load_npz("https://drive.google.com/uc?export=download&id=1SSR6JWEqG4DXRU9qo78682D9pGJF3Wr0")
print("Rank:", estimate_rank(linag.aslinearoperator(X), eps=100))

ValueError Traceback (most recent call last) in () 3 4 print(type(X)) ----> 5 print("Rank of the Document-Term Matrix:", estimate_rank(aslinearoperator(X), eps=1))

1 frames /usr/local/lib/python3.6/dist-packages/scipy/linalg/_interpolative_backend.py in idd_findrank(eps, m, n, matvect) 659:rtype: int 660 """ --> 661 k, ra, ier = _id.idd_findrank(eps, m, n, matvect) 662 if ier: 663 raise _RETCODE_ERROR

ValueError: failed to create intent(cache|hide)|optional array-- must have defined dimensions but got (-1216667648,)

Answer 1

I have used sparse, but haven't used estimate_rank . But I can read errors and docs.

In [23]: from scipy import sparse                                                                      
In [24]: from scipy.sparse import linalg                                                               
In [25]: M = sparse.random(100,100,.2, 'csr')   

In [36]: inter.estimate_rank(M,.001)                                                                   
---------------------------------------------------------------------------
...
TypeError: invalid input type (must be array or LinearOperator)

testing the array option:

In [37]: inter.estimate_rank(M.A,.1)                                                                   
Out[37]: 100

testing the linearoperator option:

In [38]: from scipy.sparse import linalg                                                               
In [39]: L = linalg.aslinearoperator(M)                                                                
In [40]: L                                                                                             
Out[40]: <100x100 MatrixLinearOperator with dtype=float64>
In [41]: inter.estimate_rank(L,.001)                                                                   
Out[41]: 99