[Numpy-discussion] Memory leak with matrices?

[josef.pktd at gmail.com]

On Tue, Mar 9, 2010 at 1:28 PM, Robert Kern <robert.kern at gmail.com> wrote:
> On Tue, Mar 9, 2010 at 11:31, David Paul Reichert
> <D.P.Reichert at sms.ed.ac.uk> wrote:
>> Hi,
>>
>> I've got two issues:
>>
>> First, the following seems to cause a memory leak,
>> using numpy 1.3.0:
>>
>> a = matrix(ones(1))
>>
>> while True:
>>    a += 0
>>
>>
>> This only seems to happen when a is a matrix rather
>> than an array, and when the short hand '+=' is used.
>
> Yes, I can verify that there is a leak in the current SVN, too. It
> appears that we are creating dictionaries with the value {'_getitem':
> False} and never decrefing them. The matrix object itself owns the
> reference.

With numpy 1.4.0 memory usage also keeps growing until I kill the process.

Josef

>
>> Second, I'm not sure whether that's a bug or whether
>> I just don't understand what's going on, but when a is a column
>> array, (e.g. a = ones((10, 1))), then
>>
>> a -= a[0,:]
>>
>> only subtracts from a[0, 0], whereas not using the short hand
>> or using something else than a on the righthand side seems
>> to subtract from all rows as expected.
>
> a[0,:] creates a view onto the same memory of the original array.
> Since you modify the values in-place, a[0,0] gets set to
> a[0,0]-a[0,0]==0, then a[1,0] gets set to a[1,0] - a[0,0] == a[1,0] -
> 0 == a[1,0], etc. Try this instead:
>
> a -= a[0,:].copy()
>
> --
> Robert Kern
>
> "I have come to believe that the whole world is an enigma, a harmless
> enigma that is made terrible by our own mad attempt to interpret it as
> though it had an underlying truth."
>  -- Umberto Eco
> _______________________________________________
> NumPy-Discussion mailing list
> NumPy-Discussion at scipy.org
> http://mail.scipy.org/mailman/listinfo/numpy-discussion
>