[SciPy-User] Eigenvectors of sparse symmetric matrix

Hao Xiong hao@biostat.ucsf....
Tue Oct 26 11:47:00 CDT 2010


 Thanks, Lutz, for checking this for me.
I am beginning to suspect my problem is a compilation issue.
On my Gentoo machine after installing arpack package separately,
scipy.sparse.linalg.test() gets

Running unit tests for scipy.sparse.linalg
NumPy version 1.5.0
NumPy is installed in /usr/lib64/python2.6/site-packages/numpy
SciPy version 0.8.0
SciPy is installed in /usr/lib64/python2.6/site-packages/scipy
Python version 2.6.5 (release26-maint, Oct 24 2010, 22:13:01) [GCC 4.4.4]
nose version 0.11.4
..../usr/lib64/python2.6/site-packages/scipy/sparse/linalg/dsolve/linsolve.py:259:
DeprecationWarning: scipy.sparse.linalg.dsolve.umfpack will be removed,
install scikits.umfpack instead
  ' install scikits.umfpack instead', DeprecationWarning )
../usr/lib64/python2.6/site-packages/scipy/sparse/linalg/dsolve/linsolve.py:75:
DeprecationWarning: scipy.sparse.linalg.dsolve.umfpack will be removed,
install scikits.umfpack instead
  ' install scikits.umfpack instead', DeprecationWarning )
....................K...Segmentation fault

So a few questions:
1. Should I install arpack package separately? I think SciPy source
comes with arpack, so it may not be necessary;
    but could it hurt?
2. What is the recommended way of supplying flags to SciPy's build
scripts? I have set my CFLAGS to
    "-march=native -mtune=native -pipe -O3 -msse." But if I compile
SciPy directly with this, build
    fails. Gentoo filters flags so its build system succeeds.

Thanks,
Hao


On 10/25/2010 10:48 PM, Lutz Maibaum wrote:
> On Mon, Oct 25, 2010 at 9:26 PM, Hao Xiong <hao@biostat.ucsf.edu> wrote:
>> Second, changing tolerance to 1e-4, 1e-5, 1e-15, all quash warning but
>> do not solve the problem: all zero eigenvalues and eigenvectors.
> Interesting. Did you get a warning about non-convergence before? I'm
> not sure what's going on with the 1e-15 tolerance, but the other ones
> are probably too large because the smallest eigenvalues seem to be
> very close to zero. For your matrix, I get
>
> In [28]: scipy.sparse.linalg.eigen_symmetric(a,3, which='SM',
> tol=1e-12, maxiter=10000000)
> Out[28]:
> (array([  2.13492457e-17,   9.54397558e-07,   1.77823892e-06]),
>  array([[-0.1       ,  0.05414805, -0.03697394],
>        [-0.1       ,  0.04864488,  0.12934357],
>        [-0.1       ,  0.09785515, -0.02710373],
>        [-0.1       ,  0.12079696, -0.095882  ],
>        [-0.1       , -0.0923547 ,  0.06131896],
>        [-0.1       ,  0.08566995,  0.02677637],
>        [-0.1       , -0.00999687, -0.03477074],
>        [-0.1       ,  0.10420394,  0.0904605 ],
>        [-0.1       , -0.01824658,  0.12656016],
>        [-0.1       ,  0.03850215, -0.16266434],
>        [-0.1       ,  0.10411247,  0.12564386],
>        [-0.1       ,  0.0847864 , -0.08006353],
>        [-0.1       ,  0.06881941, -0.03651171],
>        [-0.1       ,  0.02100945,  0.12596802],
>        [-0.1       ,  0.00698142, -0.14227155],
>        [-0.1       , -0.0994901 , -0.01865521],
>        [-0.1       ,  0.05150979, -0.13437869],
>        [-0.1       ,  0.12983085,  0.10247783],
>        [-0.1       ,  0.20811634,  0.04037155],
>        [-0.1       ,  0.14284242,  0.07440172],
>        [-0.1       ,  0.08759283,  0.00897286],
>        [-0.1       ,  0.11652933,  0.11583934],
>        [-0.1       ,  0.08273911, -0.12928089],
>        [-0.1       ,  0.15103551,  0.08544608],
>        [-0.1       , -0.10887856, -0.03683742],
>        [-0.1       ,  0.08946787,  0.01810116],
>        [-0.1       , -0.21466925,  0.08808048],
>        [-0.1       ,  0.01112506,  0.11875543],
>        [-0.1       ,  0.03862264, -0.03816272],
>        [-0.1       , -0.08819346,  0.0469191 ],
>        [-0.1       , -0.08715582, -0.10397484],
>        [-0.1       ,  0.09957673,  0.12540574],
>        [-0.1       , -0.10165562,  0.10154619],
>        [-0.1       , -0.02138075,  0.06997714],
>        [-0.1       , -0.02087899, -0.04523328],
>        [-0.1       ,  0.07205966,  0.00801408],
>        [-0.1       ,  0.06474043,  0.00830429],
>        [-0.1       ,  0.08648864, -0.00438077],
>        [-0.1       ,  0.09298343,  0.04886763],
>        [-0.1       ,  0.07158097,  0.0782138 ],
>        [-0.1       ,  0.01239778, -0.15765419],
>        [-0.1       , -0.05888361,  0.03320853],
>        [-0.1       ,  0.08010641,  0.08588525],
>        [-0.1       ,  0.03127534, -0.15888655],
>        [-0.1       ,  0.15375382, -0.00072328],
>        [-0.1       ,  0.1309185 ,  0.01948518],
>        [-0.1       , -0.21072633,  0.05625481],
>        [-0.1       ,  0.00123581, -0.19868411],
>        [-0.1       , -0.04948594, -0.1179604 ],
>        [-0.1       ,  0.03724257, -0.18880828],
>        [-0.1       , -0.05376647,  0.11361879],
>        [-0.1       ,  0.05143578, -0.11411724],
>        [-0.1       , -0.04570302,  0.13384669],
>        [-0.1       , -0.05617232, -0.09347502],
>        [-0.1       , -0.20512585,  0.0484587 ],
>        [-0.1       , -0.027912  , -0.1848302 ],
>        [-0.1       ,  0.14621125, -0.00988872],
>        [-0.1       , -0.10030626, -0.09077817],
>        [-0.1       , -0.0363287 ,  0.02784762],
>        [-0.1       , -0.21623947,  0.06780762],
>        [-0.1       , -0.06138235, -0.1349    ],
>        [-0.1       , -0.09814152, -0.04398249],
>        [-0.1       ,  0.12720599,  0.00705402],
>        [-0.1       , -0.01507454, -0.18508998],
>        [-0.1       , -0.00798772, -0.23027451],
>        [-0.1       ,  0.0084914 , -0.14105232],
>        [-0.1       , -0.00326878,  0.1905542 ],
>        [-0.1       , -0.11332749, -0.01003244],
>        [-0.1       ,  0.16373491,  0.07366324],
>        [-0.1       , -0.15722344,  0.05073253],
>        [-0.1       ,  0.04282908,  0.05747035],
>        [-0.1       , -0.11459224,  0.1258188 ],
>        [-0.1       , -0.03079556,  0.12889243],
>        [-0.1       , -0.06469642,  0.15025778],
>        [-0.1       ,  0.18106343,  0.04211254],
>        [-0.1       , -0.11284705,  0.05143415],
>        [-0.1       , -0.14384552, -0.01344659],
>        [-0.1       ,  0.00723068, -0.19844225],
>        [-0.1       , -0.05921825, -0.12152038],
>        [-0.1       , -0.20116698,  0.08917197],
>        [-0.1       , -0.17052863,  0.05183343],
>        [-0.1       , -0.01618908, -0.05137175],
>        [-0.1       , -0.04433511, -0.06585839],
>        [-0.1       ,  0.03211274,  0.1278789 ],
>        [-0.1       ,  0.12588347, -0.08004173],
>        [-0.1       , -0.08788273, -0.10250587],
>        [-0.1       ,  0.01156012, -0.00283793],
>        [-0.1       , -0.05788733, -0.08254978],
>        [-0.1       ,  0.08076025, -0.03895826],
>        [-0.1       ,  0.02232925,  0.1221555 ],
>        [-0.1       , -0.11745394, -0.02053012],
>        [-0.1       ,  0.01355673,  0.12304368],
>        [-0.1       , -0.17090312,  0.03684983],
>        [-0.1       ,  0.21020815,  0.00314479],
>        [-0.1       , -0.05367038, -0.13541344],
>        [-0.1       , -0.11600589, -0.07075897],
>        [-0.1       , -0.02728704,  0.09827715],
>        [-0.1       , -0.09539752,  0.04939529],
>        [-0.1       ,  0.0058363 ,  0.18280319],
>        [-0.1       , -0.04509233, -0.0022041 ]]))
>
> which seems in good agreement with the dense solution.
>
> Best,
>
>   Lutz



More information about the SciPy-User mailing list