[SciPy-User] scipy.linalg.solve()'s overwrite option does not work

josef.pktd@gmai... josef.pktd@gmai...
Sat Nov 6 17:24:34 CDT 2010


On Sat, Nov 6, 2010 at 6:18 PM, braingateway <braingateway@gmail.com> wrote:
> josef.pktd@gmail.com :
>> On Sat, Nov 6, 2010 at 5:46 PM, braingateway <braingateway@gmail.com> wrote:
>>
>>> Joe Kington :
>>>
>>>> On Sat, Nov 6, 2010 at 12:13 PM, braingateway <braingateway@gmail.com
>>>> <mailto:braingateway@gmail.com>> wrote:
>>>>
>>>>     David Warde-Farley:
>>>>     > On 2010-11-05, at 9:21 PM, braingateway wrote:
>>>>     >
>>>>     >
>>>>     >> Hi everyone,
>>>>     >> I believe the overwrite option is used for reduce memory usage.
>>>>     But I
>>>>     >> did following test, and find out it does not work at all. Maybe I
>>>>     >> misunderstood the purpose of overwrite option. If anybody could
>>>>     explain
>>>>     >> this, I shall highly appreciate your help.
>>>>     >>
>>>>     >
>>>>     > First of all, this is a SciPy issue, so please don't crosspost
>>>>     to NumPy-discussion.
>>>>     >
>>>>     >
>>>>     >>>>> a=npy.random.randn(20,20)
>>>>     >>>>> x=npy.random.randn(20,4)
>>>>     >>>>> a=npy.matrix(a)
>>>>     >>>>> x=npy.matrix(x)
>>>>     >>>>> b=a*x
>>>>     >>>>> import scipy.linalg as sla
>>>>     >>>>> a0=npy.matrix(a)
>>>>     >>>>> a is a0
>>>>     >>>>>
>>>>     >> False
>>>>     >>
>>>>     >>>>> b0=npy.matrix(b)
>>>>     >>>>> b is b0
>>>>     >>>>>
>>>>     >> False
>>>>     >>
>>>>     >
>>>>     > You shouldn't use 'is' to compare arrays unless you mean to
>>>>     compare them by object identity. Use all(b == b0) to compare by value.
>>>>     >
>>>>     > David
>>>>     >
>>>>     >
>>>>     Thanks for reply, but I have to say u did not understand my post
>>>>     at all.
>>>>     I did this 'is' comparison on purpose, because I wanna know if the
>>>>     overwrite flag is work or not.
>>>>     See following example:
>>>>      >>> a=numpy.matrix([0,0,1])
>>>>      >>> a
>>>>     matrix([[0, 0, 1]])
>>>>      >>> a0=a
>>>>      >>> a0 is a
>>>>     True
>>>>
>>>>
>>>> Just because two ndarray objects aren't the same doesn't mean that
>>>> they don't share the same memory...
>>>>
>>>> Consider this:
>>>> import numpy as np
>>>> x = np.arange(10)
>>>> y = x.T
>>>> x is y # --> Yields False
>>>> Nonetheless, x and y share the same data, and storing y doesn't double
>>>> the amount of memory used, as it's effectively just a pointer to the
>>>> same memory as x
>>>>
>>>> Instead of using "is", you should use "numpy.may_share_memory(x, y)"
>>>>
>>> Thanks a lot for pointing this out! I were struggling to figure out
>>> whether the different objects share memory or not. And good to know
>>> a0=numpy.matrix(a) actually did not share the memory.
>>>  >>> print 'a0 shares memory with a?', npy.may_share_memory(a,a0)
>>> a0 shares memory with a? False
>>>  >>> print 'b0 shares memory with b?', npy.may_share_memory(b,b0)
>>> b0 shares memory with b? False
>>> I also heard that even may_share_memory is 'True', does not necessarily
>>> mean they share any element. Maybe, is 'a0.base is a' usually more
>>> suitable for this purpose?
>>>
>>> Back to the original question: is there anyone actually saw the
>>> overwrite_a or overwrite_b really showed its effect?
>>> If you could show me a repeatable example, not only for
>>> scipy.linalg.solve(), it can also be other functions, who provide this
>>> option, such as eig(). If it does not show any advantage in memory
>>> usage, I might still using numpy.linalg.
>>>
>>
>>
>> import numpy as np
>>
>> a=np.random.randn(20,20)
>> abak = a.copy()
>> x=np.random.randn(20,4)
>> xbak = x.copy()
>> a=np.matrix(a)
>> x=np.matrix(x)
>> b=a*x
>> b = np.array(b)
>> bbak = b.copy()
>>
>> import scipy.linalg as sla
>> a0=np.matrix(a)
>> print a is a0
>> #False
>> b0=np.matrix(b)
>> print b is b0
>> #False
>> X=sla.solve(a,b,overwrite_a=True,debug=True)
>> print X is b
>> #False
>> print (X==b).all()
>> #False
>> print 'a:', (a0==a).all(), (abak==a).all()
>> print 'b:', (b0==b).all(), (bbak==b).all()
>> #
>> Y = sla.solve(a,b,overwrite_a=True,overwrite_b=True,debug=True)
>> print 'a:', (a0==a).all(), (abak==a).all()
>> print 'b:', (b0==b).all(), (bbak==b).all()
>>
>> print (X==Y).all()
>>
>> printout
>> -----------
>> False
>> False
>> solve:overwrite_a= True
>> solve:overwrite_b= False
>> False
>> False
>> a: False False
>> b: True True
>> solve:overwrite_a= True
>> solve:overwrite_b= True
>> a: False False
>> b: True True
>> False
>>
>>
> Thanks a lot! So right now I see it might be some bugs in my Scipy
> version!, After running your code, I got following result:
>  >>> sla.__version__
> '0.4.9'

mine:
>>> sla.__version__
'0.4.9'

I have no idea if the overwrite option depends on which Lapack/Blas
implementation is used. I have a generic oldish ATLAS.

Josef


>  >>>
> False
> False
> solve:overwrite_a= True
> solve:overwrite_b= False
> False
> False
> a: True True
> b: True True
> solve:overwrite_a= True
> solve:overwrite_b= True
> a: True True
> b: True True
> True
>> The first solve overwrites a, the second solve solves a different
>> problem and the solutions X and Y are not the same.
>> (if the first solve allows overwriting of be instead, then X==Y )
>>
>> I never got a case with overwritten b, and as you said there was no
>> sharing memory with the original array, the copy has always the same
>> result as your original.
>>
>> This was for quick playing with your example, no guarantee on no mistakes.
>>
>> Josef
>>
>>
>>
>>>>     This means a0 and a is actually point to a same object. Then a0 act
>>>>     similar to the C pointer of a.
>>>>     I compared a0/b0 and a/b by 'is' first to show I did create a new
>>>>     object
>>>>     from the original matrix, so the following (a0==a).all()
>>>>     comparison can
>>>>     actually prove the values inside the a and b were not overwritten.
>>>>
>>>>     Sincerely,
>>>>     LittleBigBrain
>>>>     > _______________________________________________
>>>>     > SciPy-User mailing list
>>>>     > SciPy-User@scipy.org <mailto:SciPy-User@scipy.org>
>>>>     > http://mail.scipy.org/mailman/listinfo/scipy-user
>>>>     >
>>>>
>>>>     _______________________________________________
>>>>     SciPy-User mailing list
>>>>     SciPy-User@scipy.org <mailto:SciPy-User@scipy.org>
>>>>     http://mail.scipy.org/mailman/listinfo/scipy-user
>>>>
>>>>
>>>> ------------------------------------------------------------------------
>>>>
>>>> _______________________________________________
>>>> SciPy-User mailing list
>>>> SciPy-User@scipy.org
>>>> http://mail.scipy.org/mailman/listinfo/scipy-user
>>>>
>>>>
>>> _______________________________________________
>>> SciPy-User mailing list
>>> SciPy-User@scipy.org
>>> http://mail.scipy.org/mailman/listinfo/scipy-user
>>>
>>>
>> _______________________________________________
>> SciPy-User mailing list
>> SciPy-User@scipy.org
>> http://mail.scipy.org/mailman/listinfo/scipy-user
>>
>
> _______________________________________________
> SciPy-User mailing list
> SciPy-User@scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>


More information about the SciPy-User mailing list