sklearn PCA產生numpy.linalg.linalg.LinAlgError

Question

我想在矩陣上運行pca，但只有numpy.linalg.linalg.LinAlgError。 我附上了矩陣和代碼。

在此處獲取矩陣： http : //workupload.com/file/YvSVhGJA

import numpy as np
from sklearn.decomposition import PCA
matrix = np.load("matrix.npy")
transformed = PCA(n_components=3).fit_transform(matrix)

這是完整的堆棧跟蹤，但是我認為您可以重現它。

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/user/anaconda/lib/python2.7/site-packages/sklearn/decomposition/pca.py", line 242, in fit_transform
    U, S, V = self._fit(X)
  File "/home/user/anaconda/lib/python2.7/site-packages/sklearn/decomposition/pca.py", line 275, in _fit
    U, S, V = linalg.svd(X, full_matrices=False)
  File "/home/user/anaconda/lib/python2.7/site-packages/scipy/linalg/decomp_svd.py", line 109, in svd
    raise LinAlgError("SVD did not converge")
numpy.linalg.linalg.LinAlgError: SVD did not converge

任何幫助表示贊賞。

PS：

np.__version__
'1.9.2'

sklearn.__version__
'0.15.2'

PPS：我正在運行Linux

Answer 1

它可以在Mac上運行，我想這對您沒有太大幫助。
嘗試X[:,:100]然后：1000？

有針對SVD的LAPACK測試 ； 他們看起來令人生畏。
“告訴我有關LAPACK安裝的所有信息”命令將很有用，但我不會立即看到它。

from __future__ import division
import platform
import sys
import numpy as np
from numpy.distutils.system_info import get_info


np.set_printoptions( threshold=100, edgeitems=10, linewidth=80,
        formatter = dict( float = lambda x: "%.2g" % x ))  # float arrays %.2g

def versions():
    print "versions: numpy %s  python %s " % (
        np.__version__, sys.version.split()[0] )
    if platform.system() == "Darwin":
        print "mac %s" % platform.mac_ver()[0]
    else:
        print platform.platform( terse=1 )  # ?
    for info in "blas_opt lapack_opt " .split():
        print "%s: %s" % (info, get_info( info, 0 ))
    print ""

versions()

#...............................................................................
X = np.load( "matrix.npy" )
print "X:", X.shape, np.percentile( X, q=[0,25,50,75,100] )

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

print "np.linalg.svd: X %s -> U %s  sing %s  Vt %s" % (
    X.shape, U.shape, sing.shape, Vt.shape )
print "svd sing:", sing

---

versions: numpy 1.9.2  python 2.7.6 
mac 10.8.3
blas_opt: {'extra_link_args': ['-Wl,-framework', '-Wl,Accelerate'], 'extra_compile_args': ['-msse3', '-DAPPLE_ACCELERATE_SGEMV_PATCH', '-I/System/Library/Frameworks/vecLib.framework/Headers'], 'define_macros': [('NO_ATLAS_INFO', 3)]}
lapack_opt: {'extra_link_args': ['-Wl,-framework', '-Wl,Accelerate'], 'extra_compile_args': ['-msse3', '-DAPPLE_ACCELERATE_SGEMV_PATCH'], 'define_macros': [('NO_ATLAS_INFO', 3)]}

X: (384, 5000) [-4.4e+02 -20 -0.27 17 4.5e+02]
np.linalg.svd: X (384, 5000) -> U (384, 384)  sing (384,)  Vt (384, 5000)
svd sing: [5e+04 2.3e+04 2.1e+04 1.3e+04 1.2e+04 1.1e+04 1.1e+04 4.3e+03 3.3e+03 1.8e+03
 ..., 0.00014 0.00014 0.00013 0.00013 0.00011 5.3e-12 5.3e-12 5.1e-16 1.3e-16
 3.3e-17]