order independent hash?

[Dan Stromberg]

On 12/5/11, 88888 Dihedral <dihedral88888 at googlemail.com> wrote:
> On Monday, December 5, 2011 1:50:08 PM UTC+8, Dan Stromberg wrote:
>> Two methods:
>> 1) If you need your hash only once in an infrequent while, then save
>> the elements in a list, appending as needed, and sort prior to
>> hashing, as needed
>>
>> 2) If you need your hash more often, you could keep your elements in a
>> treap or red-black tree; these will maintain sortedness throughout the
>> life of the datastructure.
>>
>> 3) If A bunch of log(n) or n or nlog(n) operations doesn't sound
>> appealing, then you might try this one: Create some sort of mapping
>> from your elements to the integers.  Then just use a sum.  This won't
>> scatter things nearly as well as a cryptographic hash, but it's very
>> fast, because you don't need to reevaluate some of your members as you
>> go.
>>
>> HTH
>>
> A sorted list can behave like a hash table.  This is of O(log(n)) in
> accesses
> of n items in theory.
>
> I agree with you a hash key computation method too slow than a list of n
> items  in accesses for a range of n items  should be reloadable.
>
> But this is not supported in Python yet.
>
> For tedious trivial jobs of non-heavy computing , I'll opt for easy use.

A sorted list is O(log(n)) for lookups, but O(n) for insertions.  If
you have a process doing both, the table operations are O(n).

A hash table that isn't overfilled is O(1) for lookups, O(1) for
insertions.  But this is not ordered.

Here's a straightforward treap implementation for python, with pure
python and cython versions:
http://pypi.python.org/pypi/treap/0.995

There's also at least one red-black tree implementation available.