merge & de-duplicate lists

[Alan Little]

I need to merge and de-duplicate some lists, and I have some code
which works but doesn't seem particularly elegant. I was wondering if
somebody could point me at a cleaner way to do it.

Here's my function:

+++++++++++++++ ++++

from sets import Set

def mergeLists (intialList, sourceOfAdditio nalLists,
nameOfMethodToC all) :
workingSet = Set(initialList )
for s in sourceOfAdditio nalLists :
getList = s.__getAttribut e__(nameOfMetho dToCall)
workingSet = workingSet.unio n(Set \
(callable(getLi st) and getList() or getList))
return workingSet

++++++++++++++

Two questions - passing the *name* of the method to call, and then
looking it up for each object in the list of extra sources (all of
which need to be new-style objects - not a problem in my application)
seems inelegant. My "sourcesOfAddit ionalLists" are normally all of the
same class - is there something I can bind at class level that
automagically refers to instance-level attributes when invoked?

Second (hopefully clearer & less obscure) question : is
sets.Set.union( ) an efficient way to do list de-duplication? Seems
like the obvious tool for the job.

[Alex Martelli]

Re: merge & de-duplicate lists

Alan Little wrote:
[color=blue]
> I need to merge and de-duplicate some lists, and I have some code
> which works but doesn't seem particularly elegant. I was wondering if
> somebody could point me at a cleaner way to do it.
>
> Here's my function:
>
> +++++++++++++++ ++++
>
> from sets import Set
>
> def mergeLists (intialList, sourceOfAdditio nalLists,
> nameOfMethodToC all) :
> workingSet = Set(initialList )
> for s in sourceOfAdditio nalLists :
> getList = s.__getAttribut e__(nameOfMetho dToCall)[/color]

Normal expression of this line would rather be:
getList = getattr(s, nameOfMethodToC all)
[color=blue]
> workingSet = workingSet.unio n(Set \
> (callable(getLi st) and getList() or getList))
> return workingSet
>
> ++++++++++++++
>
> Two questions - passing the *name* of the method to call, and then
> looking it up for each object in the list of extra sources (all of
> which need to be new-style objects - not a problem in my application)
> seems inelegant. My "sourcesOfAddit ionalLists" are normally all of the
> same class - is there something I can bind at class level that
> automagically refers to instance-level attributes when invoked?[/color]

I'm not quite sure what you mean here. You can of course play
around with descriptors, e.g. properties or your own custom ones.
But that 'normally' is worrisome -- what happens in the not-so-
normal cases where one or two items are of a different class...?

[color=blue]
> Second (hopefully clearer & less obscure) question : is
> sets.Set.union( ) an efficient way to do list de-duplication? Seems
> like the obvious tool for the job.[/color]

Well, .union must make a new set each time and then you throw
away the old one; this is inefficient in much the same way in
which concatenating a list of lists would be if you coded that:
result = baselist
for otherlist in otherlists:
result = result + otherlist
here, too, the + would each time make a new list and you would
throw away the old one with the assignment. This inefficiency
is removed by in-place updates, e.g. result.extend(o therlist)
for the case of list concatenation, and
workingSet.upda te(otherlist)
for your case (don't bother explicitly making a Set out of
the otherlist -- update can take any iterable).

Overall, I would code your function (including some
renamings for clarity, and the removal of a very error
prone, obscure and useless and/or -- just imagine what
would happen if getList() returned an EMPTY list...) as
follows:

def mergeLists(inti alList, sourceOfAdditio nalLists,
nameOfAttribute ):
workingSet = Set(initialList )
for s in sourceOfAdditio nalLists :
getList = getattr(s, nameOfAttribute )
if callable(getLis t): getList=getList ()
workingSet.upda te(getList)
return workingSet


Alex

[anton muhin]

Re: merge & de-duplicate lists

Alan Little wrote:[color=blue]
> I need to merge and de-duplicate some lists, and I have some code
> which works but doesn't seem particularly elegant. I was wondering if
> somebody could point me at a cleaner way to do it.
>
> Here's my function:
>
> +++++++++++++++ ++++
>
> from sets import Set
>
> def mergeLists (intialList, sourceOfAdditio nalLists,
> nameOfMethodToC all) :
> workingSet = Set(initialList )
> for s in sourceOfAdditio nalLists :
> getList = s.__getAttribut e__(nameOfMetho dToCall)
> workingSet = workingSet.unio n(Set \
> (callable(getLi st) and getList() or getList))
> return workingSet
>
> ++++++++++++++
>
> Two questions - passing the *name* of the method to call, and then
> looking it up for each object in the list of extra sources (all of
> which need to be new-style objects - not a problem in my application)
> seems inelegant. My "sourcesOfAddit ionalLists" are normally all of the
> same class - is there something I can bind at class level that
> automagically refers to instance-level attributes when invoked?[/color]
If they all are of the same class, you may just introduce method to call
that returns your lists.

BTW, your design might be not perfect. I personally whould rather split
this function into a couple: one to merge lists and the second that will
produce lists to merege (this approach might help with the problem above).

regards,
anton.

[Alan Little]

Re: merge & de-duplicate lists

Thanks for the advice guys. I've only been playing with python in my
spare time for a few weeks, and am hugely impressed with how clean and
fun and powerful it all is (compared to digging in the java and oracle
mines, which is what I do in the day job). Getting this sort of
informed critique of my ideas is great.

[Alan Little]

Re: merge & de-duplicate lists

Alex wrote:
[color=blue][color=green]
> > Two questions - passing the *name* of the method to call, and then
> > looking it up for each object in the list of extra sources (all of
> > which need to be new-style objects - not a problem in my application)
> > seems inelegant. My "sourcesOfAddit ionalLists" are normally all of the
> > same class - is there something I can bind at class level that
> > automagically refers to instance-level attributes when invoked?[/color]
>
> I'm not quite sure what you mean here. You can of course play
> around with descriptors, e.g. properties or your own custom ones.
> But that 'normally' is worrisome -- what happens in the not-so-
> normal cases where one or two items are of a different class...?[/color]

what I was getting at was that I was thnínking I would like to be able
to call my function something like this (pseudocode):

def f (listOfObjects, method) :
for o in listOfObjects :
o.method # passed method gets invoked on the instances

l = [a,b,c] # collection of objects of known class C
result = f(l, C.method) # call with method of the class

Stashing the method name in a string is a level of indirection that
somehow felt to me like it *ought* to be unnecessary, but I think I
was just thinking in non-pythonic function pointer terms. Which
doesn't work in a world where (a) we don't know if two objects of the
same class have the same methods, and (b) we have the flexibility to
write functions that will work with anything that has a method of the
requisite name, regardless of its type.

Just learning out loud here (with the help of your Nutshell book, I
might add).

[anton muhin]

Re: merge & de-duplicate lists

Alan Little wrote:[color=blue]
> def f (listOfObjects, method) :
> for o in listOfObjects :
> o.method # passed method gets invoked on the instances
>
> l = [a,b,c] # collection of objects of known class C
> result = f(l, C.method) # call with method of the class
>[/color]

Maybe the following might interest you:

1.

class A(object):
def fooA(self):
return "fooA"

class B(object):
def fooB(self):
return "fooB"

a, b = A(), B()

invoke_fooA = lambda obj: obj.fooA()
invoke_fooB = lambda obj: obj.fooB()

print invoke_fooA(a)
ptin invoke_fooB(b)

2. If the class is the same you can use:

class A(object):
def foo(self):
return "A.foo"

method = A.foo

a = A()

print method(a)

HTH,
anton.