[Numpy-discussion] strange dimension-dependent behaviour of einsum

Mark Wiebe mwwiebe at gmail.com
Wed May 18 10:57:35 EDT 2011


On Tue, May 17, 2011 at 8:53 PM, Charles R Harris <charlesr.harris at gmail.com
> wrote:

> On Tue, May 17, 2011 at 7:46 PM, Wieland Brendel <wielandbrendel at gmx.net>wrote:
>
>>  > The equality being that the expression should be ~0?
>>
>> Exactly.
>>
>> > I see the problem when the last index is in the range 235 - 390.
>>
>> Good to see I am not the only one - I was getting crazy. Same range for me by the way.**> Out of curiosity, which machine/OS are you using? I'm on 64 bit fedora 14,
>> > AMD 940.
>>
>> I am using 64 bit Windows 7 / Intel i5 M 540 / Python(x,y) distribution / Numpy 1.6 (latest, no RC)
>>
>> Any idea where that might come from? Looks like a very odd problem...
>>
>> Wieland
>>
>> PS: Its my first time on a mailing list - hope this mail actually replies to the thread. If not: How do I reply?
>>
>>
>>
> You're doing fine. You have found a bug, not unexpected with einsum not
> having been tested as much as the rest of the release since it is new and
> hasn't been used before. Please open a ticket.
>
> Chuck
>

I've committed a fix to this, and a crash that I encountered when
experimenting with it, to both master and the 1.6.1 branch. It turned out to
be an incorrectly handled interaction in the iterator when the inner loop
for the reduction case doesn't fit all the way in the buffer.

Cheers,
Mark
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20110518/6a791691/attachment.html>


More information about the NumPy-Discussion mailing list