Scipy.linalg.solve最大数组大小

Question

Is there a maximum size that scipy.linalg.solve ( numpy.linalg.solve also fails) can handle?

I'm working on porting some Matlab code (some poisson image cloning, which I didn't write) into python, and in Python I have the code:

print M.shape
print b.shape

x = scipy.linalg.solve(M,b)

Which keeps returning the output (and error):

(2550, 2550)
(2550, 1)


Traceback (most recent call last):
  File "D:\Other\WB_Lab\PythonPort\FaceManip\Code\test_runner.py", line 19, in <module>
    ibgFaceManipulator.main(testOut,pathName,ptsFile,subjectId,'2','0')
  File "D:\Other\WB_Lab\PythonPort\FaceManip\Code\ibgFaceManipulator.py", line 70, in main
    ManipulateFaceWrapper.main(outDir,subjectId,retry,attempts,listchoice,NewImageData,AllBlend,Patchizer)
  File "D:\Other\WB_Lab\PythonPort\FaceManip\Code\ManipulateFaceWrapper.py", line 31, in main
    ManipulateFace.main(outdir,newImageData,3,filename,patchizer,blendParam,eyeDistortionParams,mouthDistortionParams,[])
  File "D:\Other\WB_Lab\PythonPort\FaceManip\Code\ManipulateFace.py", line 38, in main
    FaceTex = PatchOverFaceWithSelectionPre.main(TexLib,FaceTex,OperationMaskNew,M,0)
  File "D:\Other\WB_Lab\PythonPort\FaceManip\Code\PatchOverFaceWithSelectionPre.py", line 104, in main
    tex = poissonImageCloneColor.main(tex,source_tex,imgProc.imerode(imMask,imgProc.strel('disk',1)))
  File "D:\Other\WB_Lab\PythonPort\FaceManip\Code\poissonImageCloneColor.py", line 22, in main
    imr = poissonImageClone.main(imir, imr,imMask)
  File "D:\Other\WB_Lab\PythonPort\FaceManip\Code\poissonImageClone.py", line 81, in main
    x = scipy.linalg.solve(np.squeeze(M),b)
  File "C:\Python27\lib\site-packages\scipy\linalg\basic.py", line 78, in solve
    raise ValueError('expected square matrix')
ValueError: expected square matrix

But I know for certain that M is square (of size 2550,2550 ). Both M and b are of dtype float64, and M is a csc formatted matrix.

Has anyone seen similar errors before?

I have found the numpy.linalg.cond(M) also fails. This time saying that the Matrix M is 0-dimensional (which is the same error numpy.linalg.solve gives me).

Answer 1

I did some more research and found that sparse matrices (which I'm using) are actually a special case. Scipy has a different solve function for them scipy.sparse.linalg.spsolve . Using this I was able to get solve the equation and get results.

Answer 2

Hmmm... Interesting. Mine gave the correct result. Could you try a simple matrix, like:

import numpy as np
from scipy.linalg import solve
M = np.eye(5)
b = np.ones(5)
solve(M, b)

to see whether the problem still remains? And also, what version is your scipy?