Set operations on object attributes question

[TheSeeker]

Hi,

I have run into something I would like to do, but am not sure how to
code it up. I would like to perform 'set-like' operations (union,
intersection, etc) on a set of objects, but have the set operations
based on an attribute of the object, rather than the whole object.

For instance, say I have (pseudo-code):

LoTuples1 = [(1,1,0),(1,2,1) ,(1,3,3)]
Set1=set(LoTupl es1)
LoTuples2 = [(2,1,3),(2,2,4) ,(2,3,2)]
Set2=set(LoTupl es2)

What I would like to be able to do is:

Set3 = Set1union(Set2)
Set3.intersecti on(Set2, <use object[2]>)

to return:
set([(2,1,3), (1,3,3)])

How can one do this operation?

Thanks,
Duane

[A.T.Hofkamp]

Re: Set operations on object attributes question

On 2007-10-23, TheSeeker <duane.kaufman@ gmail.comwrote:

Hi,
>
I have run into something I would like to do, but am not sure how to
code it up. I would like to perform 'set-like' operations (union,
intersection, etc) on a set of objects, but have the set operations
based on an attribute of the object, rather than the whole object.
>
For instance, say I have (pseudo-code):
>
LoTuples1 = [(1,1,0),(1,2,1) ,(1,3,3)]
Set1=set(LoTupl es1)
LoTuples2 = [(2,1,3),(2,2,4) ,(2,3,2)]
Set2=set(LoTupl es2)
>
What I would like to be able to do is:
>
Set3 = Set1union(Set2)
Set3.intersecti on(Set2, <use object[2]>)
>
to return:
set([(2,1,3), (1,3,3)])
>
How can one do this operation?

Put your data in a class, and implement __hash__ and __eq__
Finally, put your objects in sets.

Albert

[Raymond Hettinger]

Re: Set operations on object attributes question

[Duane]

LoTuples1 = [(1,1,0),(1,2,1) ,(1,3,3)]
Set1=set(LoTupl es1)
LoTuples2 = [(2,1,3),(2,2,4) ,(2,3,2)]
Set2=set(LoTupl es2)
>
What I would like to be able to do is:
>
Set3 = Set1union(Set2)
Set3.intersecti on(Set2, <use object[2]>)
>
to return:
set([(2,1,3), (1,3,3)])
>
How can one do this operation?

Conceptually, there is more than one operation going on. First,
finding the attributes shared in both sets:
ca = set(t[2] for t in LoTuples1) & set(t[2] for t in LoTuples2)
which gives:
set([3])

Second, find any tuple which has that attribute (including multiple
results for the same attribute):
set(t for t in (LoTuples1 + LoTuples2) if t[2] in ca)
which returns:
set([(2, 1, 3), (1, 3, 3)])

Wanting multiple results for the same attribute value (i.e. both
(2,1,3) and (1,3,3) have 3 in the second position) is why multiple
steps are needed; otherwise, the behavior of intersection() is to
return a single representative of the equivalence class.


Raymond

[TheSeeker]

Re: Set operations on object attributes question

Hi,

Thanks for the response! (See below for more discussion)

On Oct 23, 10:39 am, Raymond Hettinger <pyt...@rcn.com wrote:

[Duane]
>

LoTuples1 = [(1,1,0),(1,2,1) ,(1,3,3)]
Set1=set(LoTupl es1)
LoTuples2 = [(2,1,3),(2,2,4) ,(2,3,2)]
Set2=set(LoTupl es2)

>

What I would like to be able to do is:

>

Set3 = Set1union(Set2)
Set3.intersecti on(Set2, <use object[2]>)

>

to return:
set([(2,1,3), (1,3,3)])

>

How can one do this operation?

>
Conceptually, there is more than one operation going on. First,
finding the attributes shared in both sets:
ca = set(t[2] for t in LoTuples1) & set(t[2] for t in LoTuples2)
which gives:
set([3])

In my use case, I already know object[2] is the key I wish to use, so
I could do this without the first step.
I am thinking of object[n] as the 'key' I wish to operate on, and the
other items in the tuple are useful attributes
I'll need in the end.

>
Second, find any tuple which has that attribute (including multiple
results for the same attribute):
set(t for t in (LoTuples1 + LoTuples2) if t[2] in ca)
which returns:
set([(2, 1, 3), (1, 3, 3)])
>
Wanting multiple results for the same attribute value (i.e. both
(2,1,3) and (1,3,3) have 3 in the second position) is why multiple
steps are needed; otherwise, the behavior of intersection() is to
return a single representative of the equivalence class.
>

This second operation is really much like what I cam up with before I
started looking into exploiting the power of sets
(this same operation can be done strictly with lists, right?)

Since what I _really_ wanted from this was the intersection of the
objects (based on attribute 2), I was looking for a set-based
solution,
kinda like a decorate - <set operation- undecorate pattern. Perhaps
the problem does not fall into that category.

Thanks for your help,
Duane

[Raymond Hettinger]

Re: Set operations on object attributes question

Since what I _really_ wanted from this was the intersection of the

objects (based on attribute 2), I was looking for a set-based
solution,
kinda like a decorate - <set operation- undecorate pattern. Perhaps
the problem does not fall into that category.

The "kinda" part is where the idea falls down. If you've decorated
the inputs with a key function (like the key= argument to sorted()),
then the intersection step will return only a single element result,
not both matches as you specified in your original request. IOW,
you cannot get set([(2, 1, 3), (1, 3, 3)]) as a result if both
set members are to be treated as equal.

It would be rather like writing set([1]).intersection( set([1.0]))
and expecting to get set([1, 1.0]) as a result. Does your
intuition support having an intersection return a set larger
than either of the two inputs?


Raymond

[Raymond Hettinger]

Re: Set operations on object attributes question

[Duane]

Since what I _really_ wanted from this was the intersection of the
objects (based on attribute 2), I was looking for a set-based
solution,
kinda like a decorate - <set operation- undecorate pattern. Perhaps
the problem does not fall into that category.

Decorating and undecorating do not apply to your example. If all the
input were decorated with new equality and hash methods, then the
intersection step would return just a single element, not the two that
you specified.

Compare this with ints and floats which already have cross-type
equality and hash functions. Running, set([1]).intersection([1.0]),
would you expect set([1.0]) or set([1, 1.0])? Your example specified
the latter behavior which has the unexpected result where the
set intersection returns more elements than exist in either input.


Raymond