Schwartzian transform for tuple in list

David Di Biase dave.dibiase at gmail.com
Thu Sep 25 00:45:35 CEST 2008


When you say slightly, is it enough to make a difference? Why would  
getitems be faster even - not sure I can think why...

Sent from my iPhone

On 24-Sep-08, at 5:46 PM, Matt Nordhoff <mnordhoff at mattnordhoff.com>  
wrote:

> Chris Rebert wrote:
>> On Wed, Sep 24, 2008 at 2:02 PM, David Di Biase <dave.dibiase at gmail.com 
>> > wrote:
>>> Hi,
>>>
>>> I have a rather large list structure with tuples contained in them  
>>> (it's
>>> part of a specification I received) looks like so:
>>> [(x1,y1,r1,d1),(x2,y2,r2,d2)...]
>>>
>>> The list can range from about 800-1500 tuples in size and I'm  
>>> currently
>>> sorting it with this:
>>>
>>> a_list.sort(lambda a, b: cmp(b[3], a[3]))
>>
>> You'd probably be better off using the 'key' keyword argument to
>> .sort(), which is made for the Schwartzian Transform:
>>
>> a_list.sort(key=lambda x: x[3])
>>
>> This sorts the list items by the value produced by the key function
>> for each item. It's also (IIRC) faster than using a comparison
>> function like you're currently doing.
>>
>> Regards,
>> Chris
>
> Yes, using 'key' is faster than 'cmp'.
>
> If you have Python 2.4 or newer, it would also be slightly faster to  
> use
> operator.itemgetter() instead of a lambda:
>
>>>> import operator
>>>> a_list.sort(key=operator.itemgetter(3))
>
>>> I'm actually sorting it by the last value in the tuple (d2). I  
>>> have been
>>> researching more efficient sorting algorithms and came across  
>>> Schwartzian
>>> transform via these links:
>>>
>>> http://www.biais.org/blog/index.php/2007/01/28/23-python-sorting-efficiency
>>> http://dev.fyicenter.com/Interview-Questions/Python/Can_you_do_a_Schwartzian_Transform_in_Python_.html
>>>
>>> I get what's happening (sorta...errr...lol)  but I'm not sure if  
>>> it is more
>>> efficient in my scenario, if it is then I have no idea how to  
>>> implement it
>>> properly :-/
>>>
>>> Would be great if a true expert would offer a suggestion for me...
>>>
>>> Thanks!
>>>
>>> David
> -- 



More information about the Python-list mailing list