bags in collections - PEP 320

[Peter Dobcsanyi]

Dear All,

I have just read PEP-320 and noticed the following comments regarding
the "collections" package:

...
- ? bag (only if use cases established)
...

I would like to argue for such a use case. We at designtheory.org are
working on a (maths / stats / comp. sci.) project in the area of Design
Theory. Design theory intersects with many fields, to mention a few:
combinatorics, graph theory, finite geometry, coding theory, and the
design of statistical experiments (from which then name of the field
comes). Most of our software development will be in Python, although
the released python software so far is very modest.

To show why bags are important for us and, in general, for discrete
mathematicians, here is the definition of the most important type of
designs. A binary block design is a multiset (that is a 'bag') of
subsets of a 'base' set; the elements of the base set are called
'points' and its subsets are called 'blocks'. (Personally, I prefer the
name bag but most mathematicians use multiset.) A non-binary block
design is one whose blocks can also be multisets.

The computations we are facing are very much combinatorial in nature so
we are happy to see that Sets became a builtin and, of course, would
like to see a C implementation for bags too. Our pydesign package (will)
heavily use C extensions. So far we use numarray to compute statistical
properties of designs, but many more combinatorial functionalities will
be implemented in C.

In particular, we are planning to implement a basic permutation group
package whose core will eventually be a C extension. We would like to
deal with automorphism groups and isomorphisms of designs on at least a
basic level within the pydesign package without having to resort to
specialized mathematical packages like GAP. These functionalities can be
useful not only for design theorist, but for anybody using Python to
deal with combinatorial structures, like graphs for example.

In all of these areas the use of multisets (bags) is pervasive.
Please, implement it -- so we don't have to

For more details on our project, please visit http://designtheory.org/

-- ,
Peter Dobcsanyi

 

[Aahz]

- ? bag (only if use cases established)

To show why bags are important for us and, in general, for discrete
mathematicians, here is the definition of the most important type of
designs. A binary block design is a multiset (that is a 'bag') of
subsets of a 'base' set; the elements of the base set are called
'points' and its subsets are called 'blocks'. (Personally, I prefer the
name bag but most mathematicians use multiset.) A non-binary block
design is one whose blocks can also be multisets.

"Use case" in this context refers more to the question of whether a
given feature should be included as part of the standard package rather
than expecting people to implement their own. One important question
that needs to be answered is whether "reasonable" people would be likely
to come up with different answers for interface and implementation
(that's what killed much of the TZ support in the datetime module).

Given the lack of response to your post, I'd suggest that you send your
comments directly to the PEP author.

 

[Peter Dobcsanyi]

Given the lack of response to your post, I'd suggest that you send your
comments directly to the PEP author.

Thanks for your suggestion. I already sent my comments to Raymond
Hettinger and we are being engaged in a discussion.

Peter