mailRe: [bug #20821] Optimisation failure / round error on windows 64 bit


Others Months | Index by Date | Thread Index
>>   [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Header


Content

Posted by Edward d'Auvergne on June 20, 2013 - 11:02:
Hi,

For a quick test, I was wondering if you were able to try the official 
packages:

Python 2.7.5 (http://www.python.org/ftp/python/2.7.5/python-2.7.5.msi),
Numpy 
(http://sourceforge.net/projects/numpy/files/NumPy/1.7.1/numpy-1.7.1-win32-superpack-python2.7.exe/download),
Scipy 
(http://sourceforge.net/projects/scipy/files/scipy/0.12.0/scipy-0.12.0-win32-superpack-python2.7.exe/download),
wxPython 
(http://sourceforge.net/projects/wxpython/files/wxPython/2.8.12.1/wxPython2.8-win64-unicode-2.8.12.1-py27.exe/download?use_mirror=dfn).

These are the links to the current 32-bit MS Windows packages (if you
are currently reading this in the archives, you should find the more
up-to-date packages).  The reason is because there are currently no
official 64-bit packages for Numpy, Scipy, and wxPython.  Hmmmm, I'm
wondering if that is because the 64-bit code is buggy and has
precision issues!

It will take longer to download than to install all four pre-compiled
packages on a Windows system.  I don't think you need to be an admin
to do this.  It would be good to see if this combination of official
releases has the same terrible precision issue on your system.

Cheers,

Edward



On 20 June 2013 10:19, Edward d'Auvergne <edward@xxxxxxxxxxxxx> wrote:
Does this work with places=2 or 3 for the te value?  That would be
better.  The precision is shocking, I have to say.  I would suggest
that in the comments, that something about this being the WinPython
distribution would also be useful for these precision records.  I
would not think that it is the Microsoft C compiler that is a problem,
but rather the numpy packages bundled with this Python derivative.
The Microsoft compiler has never been an issue before, so I would
suggest that the WinPython people have a precision bug in their code.
Or maybe they have deliberately turned down their precision so that it
appears as if their distribution is much faster than standard
Python+numpy?  Is it also WinPython that has problems with icons in
the menus?

Regards,

Edward


On 20 June 2013 09:18, Troels E. Linnet
<NO-REPLY.INVALID-ADDRESS@xxxxxxx> wrote:
Follow-up Comment #5, bug #20821 (project relax):

Reply to bug #20821: (https://gna.org/bugs/?20821) Optimisation failure /
round error on windows 64 bit

I experienced more precision problems.

I had a terrible need, to turn down precision, quite a lot.

So much, that I think something is wrong with the MS
Visual Studio Express """ Professional """ 64 bit compilation.

-            self.assertAlmostEqual(spin.s2, s2, msg=mesg)
+            self.assertAlmostEqual(spin.s2, s2, places=6, msg=mesg)

-            self.assertAlmostEqual(spin.te / 1e-12, te, 5, msg=mesg)
+            self.assertAlmostEqual(spin.te / 1e-12, te, places=1, 
msg=mesg)

-            self.assertAlmostEqual(spin.rex * (2.0 * pi *
cdp.spectrometer_frq[cdp.ri_ids[0]])**2, rex * (2.0 * pi *
cdp.spectrometer_frq[cdp.ri_ids[0]])**2, msg=mesg)
+            self.assertAlmostEqual(spin.rex * (2.0 * pi *
cdp.spectrometer_frq[cdp.ri_ids[0]])**2, rex * (2.0 * pi *
cdp.spectrometer_frq[cdp.ri_ids[0]])**2, places=4, msg=mesg)

        # Optimisation values.
        s2 = 0.9699999999999995
        te = 2048.000000000022283
        rex = 0.14900000000000566 / (2.0 * pi *
cdp.spectrometer_frq[cdp.ri_ids[0]])**2

My value
# s2: 0.9700002183674102 : Diff -2.18367410709e-07
# te: 2048.015293187 Diff: -0.0152931869779
# rex: 0.14899473115727899 Diff: 5.26884272667e-06

# 64-bit Windows 7 (https://gna.org/bugs/?20821).
# System: Windows
# Release: 7
# Version: 6.1.7601
# Architecture: 64bit WindowsPE
# Machine: AMD64
# Processor: Intel64 Family 6 Model 37 Stepping 2, GenuineIntel
# Python version: 2.7.5
# numpy version: 1.7.1
# Compilation: Visual Studio Express 2012, Professional, for 64 bit
compilation
# s2: 0.9700002183674102
# te: 2048.015293187
# rex: 0.14899473115727899
# chi2: 2.3195994119090742e-10
# iter: 116
# f_count: 411
# g_count: 411
# h_count: 0
# warning: None

(file #18123)
    _______________________________________________________

Additional Item Attachment:

File name: patch3                         Size:2 KB


    _______________________________________________________

Reply to this item at:

  <http://gna.org/bugs/?20821>

_______________________________________________
  Message sent via/by Gna!
  http://gna.org/


_______________________________________________
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



Related Messages


Powered by MHonArc, Updated Thu Jun 20 13:20:08 2013