Re: [sympy] Test failure in 0.7.0 pysics.quantum.matrixcache

Wed, 22 Jun 2011 05:57:45 -0700

Seem like it would be more consistent to raise an ImportError rather than return None. That would also mean less change to existing code. Thinking a bit further, could this be done in an import hook (I haven't done one myself, so not sure)? 

Ted

On 2011-06-22, at 02:09, Aaron Meurer wrote:


So far, I've coded up an import_module function, that wraps the try,
except ImportError logic into a single unified function.  It has
support for checking the module version and the Python version and
returning None if they are too old, even if the module is installed,
and it also supports raising warnings when the module cannot be used.
You can see the code at
https://github.com/asmeurer/sympy/tree/external.

I've yet to apply this everywhere in SymPy (I will do it tomorrow).  I
still need to decide if it will be good enough to use
import_module('numpy', min_python_version=(2, 6)) everywhere in the
quantum module, or if I should centralize numpy importing in that
module even further.

But it should a *lot* less code, i.e., instead of

import sys
import warnings

try:
   if sys.version_info < (2, 6):
       warnings.warn("Python version too old to use numpy")
       raise ImportError
   import numpy
except ImportError:
   <do numpy not supported stuff>

in every single quantum file that uses numpy (including the tests), we
would just have

from sympy.external import import_module
numpy = import_module('numpy', min_python_version=(2, 6))
if not numpy:
   <do numpy not supported stuff>

Aaron Meurer
On Mon, Jun 20, 2011 at 6:11 PM, Aaron Meurer <asmeu...@gmail.com> wrote: 
It should at least be in a separate file for the quantum code.  I
didn't consider to make one file for it for all of SymPy. I'm waiting
to see what Brian Granger thinks, since he knows the most about the
quantum module.

Aaron Meurer
On Mon, Jun 20, 2011 at 6:06 PM, Ted Horst <ted.ho...@earthlink.net> wrote:
Thanks for investigating, Aaron. Maybe numpy importing should be pulled out into its own module. This would probably be necessary anyway if someone implements the "numpy array for sympy" idea that was talked about a while 
back.

Ted

On 2011-06-20, at 00:42, Aaron Meurer wrote:


The way the quantum code is written, it's not very easy to disable
numpy globally. I could do it, but it would require copying the same four lines checking the system version and emitting a warning in the 
half dozen files that import numpy.

So, to fix this right, I need advice from Brian and others who know
the quantum code. Where would be the best place to put some kind of global parameter defining if numpy is supported for the quantum code 
(similar to what the polys use for gmpy)?

Aaron Meurer
On Sun, Jun 19, 2011 at 8:38 PM, Aaron Meurer <asmeu...@gmail.com> wrote:
OK, this is definitely a bug in numpy (or Python). I created a file 
numpyfloattest.py with the following contents:

import numpy

class A(object):
  def __float__(self):
      return 2.0
  def __complex__(self):
      return 5.0j

a = A()

print numpy.matrix([[a]], dtype='complex')

And ran:
Aaron-Meurer:sympy aaronmeurer(0.7.0$%)$python2.5 numpyfloattest.py 
[[ 2.+0.j]]
Aaron-Meurer:sympy aaronmeurer(0.7.0$%)$python2.7 numpyfloattest.py 
[[ 0.+5.j]]

So for some reason, in Python 2.5, it runs complex(float()) on the
data.  Of course, this doesn't work very well if the numbers are
complex, like sympy.I.
I have created http://projects.scipy.org/numpy/ticket/1872 for this. 
In the meanwhile, I will just disable numpy in the quantum code in
Python 2.5.

Aaron Meurer
On Sun, Jun 19, 2011 at 8:04 PM, Aaron Meurer <asmeu...@gmail.com> wrote: 

I think I have a lead.  In numpy 1.6, this instead gives


numpy.matrix([[0, I]], dtype='complex')


Traceback (most recent call last):
 File "<stdin>", line 1, in <module>
 File
"/Library/Frameworks/Python.framework/Versions/2.5/lib/ python2.5/site-packages/numpy/matrixlib/defmatrix.py", 
line 257, in __new__
  arr = N.array(data, dtype=dtype, copy=copy)
 File "sympy/core/expr.py", line 129, in __float__
  raise ValueError("Symbolic value, can't compute")
ValueError: Symbolic value, can't compute
in Python 2.5 (and it works in Python 2.6 and 2.7). So I guess the bug is in SymPy somewhere, and numpy 1.5 was catching that ValueError 
and re-throwing it as a TypeError.

Aaron Meurer
On Sun, Jun 19, 2011 at 7:39 PM, Aaron Meurer <asmeu...@gmail.com> 
wrote:
On Sun, Jun 19, 2011 at 7:32 PM, Aaron Meurer <asmeu...@gmail.com > 
wrote:
I don't know much about numpy, so this one has me stumped. For some reason numpy.matrix([[0, I]], dtype='complex') fails in Python 2.5 (with the above error), but not in Python 2.6 or 2.7 (I is sympy.I).
Can someone with numpy knowledge give any insight here? Otherwise, I think the solution might have to be to disable numpy support in the 
quantum code in Python 2.5.
I haven't looked at why it kills the test runner yet. I'll look at 
that now.
Oh, this is rather obvious. It's because the exception is raised at import time. If it were raised when running an actual test, it would be caught by the test runner. This is probably OK behavior (anyway, 
it's not top priority to change it).

Aaron Meurer
By the way, a workaround might be to do np.matrix([map(complex, i) for i in m.tolist()], dtype=complex), i.e., manually convert the elements of the Matrix to complex, but this is rather hackish and I would 
prefer to aviod it if possible.

Aaron Meurer
On Sun, Jun 19, 2011 at 5:36 PM, Aaron Meurer <asmeu...@gmail.com > 
wrote:
Thanks for reporting this. It seems you've uncovered two bugs (the 
second one is that the error killed the test runner).
I'll see if I can figure out how to fix at least the numpy one for 
the
release. Do you know how to fix it? If so, please send a pull 
request.

Aaron Meurer
On Sat, Jun 18, 2011 at 1:03 PM, Ted Horst <ted.ho...@earthlink.net > 
wrote:
I'm a little late to the party, but I just tested 0.7.0 tip and got 
this
error (stopped testing):
This is with python 2.5.1, numpy 1.5.0, no scipy, no gmpy on OSX 
10.5.8 PPC
(big endian).

I can get the error from just importing
sympy.physics.quantum.matrixcache.
It makes sense because we are passing sympy.I with dtype 'complex' 
to
numpy.

Ted

Traceback (most recent call last):
 File "setup.py", line 269, in <module>
  'audit' : audit,
 File
"/System/Library/Frameworks/Python.framework/Versions/2.5/ lib/python2.5/distutils/core.py", 
line 151, in setup
  dist.run_commands()
 File
"/System/Library/Frameworks/Python.framework/Versions/2.5/ lib/python2.5/distutils/dist.py", 
line 974, in run_commands
  self.run_command(cmd)
 File
"/System/Library/Frameworks/Python.framework/Versions/2.5/ lib/python2.5/distutils/dist.py", 
line 994, in run_command
  cmd_obj.run()
 File "setup.py", line 161, in run
  if sympy.test():
 File "sympy/utilities/runtests.py", line 181, in test
  return t.test(sort=sort)
 File "sympy/utilities/runtests.py", line 487, in test
  self.test_file(f)
 File "sympy/utilities/runtests.py", line 499, in test_file
  execfile(filename, gl)
 File
"/users/ted/developer/opensource/sympy/sympy_github2/sympy/ physics/quantum/tests/test_gate.py", 
line 4, in <module>
from sympy.physics.quantum.gate import (XGate, YGate, ZGate, 
random_circuit,
 File "sympy/physics/quantum/gate.py", line 31, in <module>
  from sympy.physics.quantum.matrixcache import matrix_cache
File "sympy/physics/quantum/matrixcache.py", line 92, in <module> 
  matrix_cache.cache_matrix('Y', Matrix([[0, -I], [I, 0]]))
 File "sympy/physics/quantum/matrixcache.py", line 38, in
cache_matrix
  self._numpy_matrix(name, m)
 File "sympy/physics/quantum/matrixcache.py", line 72, in
_numpy_matrix
  m = to_numpy(m, dtype=self.dtype)
File "sympy/physics/quantum/matrixutils.py", line 99, in to_numpy 
  return sympy_to_numpy(m, dtype=dtype)
 File "sympy/physics/quantum/matrixutils.py", line 53, in
sympy_to_numpy
  return np.matrix(m.tolist(), dtype=dtype)
 File "numpy/matrixlib/defmatrix.py", line 257, in __new__
TypeError: a float is required

--
You received this message because you are subscribed to the Google 
Groups
"sympy" group.
To post to this group, send email to sympy@googlegroups.com.
To unsubscribe from this group, send email to
sympy+unsubscr...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/sympy?hl=en.














--
You received this message because you are subscribed to the Google Groups 
"sympy" group.
To post to this group, send email to sympy@googlegroups.com.
To unsubscribe from this group, send email to
sympy+unsubscr...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/sympy?hl=en.



--
You received this message because you are subscribed to the Google Groups 
"sympy" group.
To post to this group, send email to sympy@googlegroups.com.
To unsubscribe from this group, send email to
sympy+unsubscr...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/sympy?hl=en.






--
You received this message because you are subscribed to the Google Groups "sympy" group. 
To post to this group, send email to sympy@googlegroups.com.
To unsubscribe from this group, send email to sympy+unsubscr...@googlegroups.com . For more options, visit this group at http://groups.google.com/group/sympy?hl=en . 



--
You received this message because you are subscribed to the Google Groups 
"sympy" group.
To post to this group, send email to sympy@googlegroups.com.
To unsubscribe from this group, send email to 
sympy+unsubscr...@googlegroups.com.
For more options, visit this group at 
http://groups.google.com/group/sympy?hl=en.

Reply via email to