Re: Fatal problem with numpy 1.6.x and the disp_spin_speed branch changes. -- July 25, 2014 - 14:35

Hi Edward.

I would suggest an implementation of striding through the data.

The necessary functions for this implementation is already present in
lib/dispersion matrix_exponential.py and matrix_power.py.

I will look into this later.

I will try to see, if I can do a local downgrade of numpy to 1.6, to
check it is working.

Best
Troels


2014-07-25 14:09 GMT+02:00 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:
> Hi Troels,
>
> We have a major problem in trunk caused by the recent merger of the
> disp_spin_speed branch.  When I run the Relax_disp system tests, the
> unit tests, or the Relax_disp GUI tests, I see many, many failures due
> to problems with the numpy.linalg.eig() function on numpy 1.6.1,
> 1.6.2, and 1.7.0.  The error for one of the unit tests is:
>
> ======================================================================
> ERROR: Test the r2eff_ns_cpmg_2site_3D() function for no exchange when dw = 
> 0.0.
> ----------------------------------------------------------------------
> Traceback (most recent call last):
>   File 
> "/data/relax/relax-trunk/test_suite/unit_tests/_lib/_dispersion/test_ns_cpmg_2site_3d.py",
> line 130, in test_ns_cpmg_2site_3D_no_rex1
>     self.calc_r2eff()
>   File 
> "/data/relax/relax-trunk/test_suite/unit_tests/_lib/_dispersion/test_ns_cpmg_2site_3d.py",
> line 79, in calc_r2eff
>     r2eff_ns_cpmg_2site_3D(r180x=self.r180x, M0=self.M0,
> M0_T=self.M0_T, r20a=self.r20a*a, r20b=self.r20b*a, pA=self.pA,
> dw=dw_frq*a, dw_orig=dw_frq*a, kex=self.kex,
> inv_tcpmg=self.inv_relax_times*a, tcp=self.tau_cpmg*a,
> back_calc=self.R2eff, num_points=self.num_points*b, power=self.ncyc*a)
>   File "/data/relax/relax-trunk/lib/dispersion/ns_cpmg_2site_3d.py",
> line 299, in r2eff_ns_cpmg_2site_3D
>     Rexpo_mat = matrix_exponential_rank_NE_NS_NM_NO_ND_x_x(R_mat)
>   File "/data/relax/relax-trunk/lib/dispersion/matrix_exponential.py",
> line 78, in matrix_exponential_rank_NE_NS_NM_NO_ND_x_x
>     W, V = eig(A)
>   File "/data/python/lib/python2.5/site-packages/numpy/linalg/linalg.py",
> line 1015, in eig
>     _assertRank2(a)
>   File "/data/python/lib/python2.5/site-packages/numpy/linalg/linalg.py",
> line 155, in _assertRank2
>     two-dimensional' % len(a.shape)
> LinAlgError: 7-dimensional array given. Array must be
> two-dimensional
>
> ----------------------------------------------------------------------
>
> Here are the API differences between numpy versions:
>
> 1.8.1:  
> http://docs.scipy.org/doc/numpy-1.8.1/reference/generated/numpy.linalg.eig.html#numpy.linalg.eig
> 1.8.0:  
> http://docs.scipy.org/doc/numpy-1.8.0/reference/generated/numpy.linalg.eig.html#numpy.linalg.eig
> 1.7.0:  
> http://docs.scipy.org/doc/numpy-1.7.0/reference/generated/numpy.linalg.eig.html#numpy.linalg.eig
> 1.6.0:  
> http://docs.scipy.org/doc/numpy-1.6.0/reference/generated/numpy.linalg.eig.html#numpy.linalg.eig
>
> You can see that the input changes from (M, M) to (..., M, M) between
> 1.7.0 and 1.8.0.  We cannot require numpy >= 1.8 as many current
> distributions have not shifted to this version yet.  Most sys admins
> would kill me for even suggesting that ;)  Therefore we need to come
> up with a solution and quickly.
>
> The best would be to add an eig() function to the lib.compat module.
> For numpy >= 1.8.0 it will use the numpy.linalg.eig() function.  For
> the other versions, we need an alternative solution.  Maybe using slow
> Python looping over the higher dimensions would be ok.
>
> I will not report this as a bug, as no released relax versions are affected.
>
> Cheers,
>
> Edward
>
> _______________________________________________
> relax (http://www.nmr-relax.com)
>
> This is the relax-devel mailing list
> relax-devel@xxxxxxx
>
> To unsubscribe from this list, get a password
> reminder, or change your subscription options,
> visit the list information page at
> https://mail.gna.org/listinfo/relax-devel