Why are there no ordered dictionaries?

[bonono]

Ok, so if not in the standard library, what is the problem? Can't find what
you want with google and PyPI etc.? Or haven't really settled on what your
_requirements_ are? That seems to be the primary problem people who complain
with "why no sprollificator mode?" questions. They don't know what they really
mean when it comes down to a DYFR (Define Your Felicitous Requirements) challenge.
So DYFR ;-)

Beat me. I am not the one asking the question.

So you'd like the mechanics to be automated and hidden? Then you need to
DYFR for using the black box you want. Methods, semantics.

Lose you. don't know what you want to say.

 

[bonono]

What I don't understand is why legitimate questions such as "why are
there no ordered dictionaries" are immediately interpreted as
*complaints* and not just as questions. If I ask such a question, I am
not complaining but trying to simply figure out *why* there is no such
thing. Probably there are reasons and all I want to know is find these
reasons and learn a little bit more about Python in doing so.

Why can't such questions be discussed in a factual, calm and friendly way?

Using "why can't" is already too much. Even you turn it into "is there
a thing for ordered dict", you would get similar treatment

Get used to it

I don't think that this was true in this case, and even if this is the
outcome, those who asked the question will have learned something.

I think a discussion group is not there for only presenting mature,
sophisticated thoughts and concepts, but also for "thinking loud"
together with other about these issues. We all know that clarifying our
thoughts works often best if you discuss them with others. And I think
that's one purpose of discussion lists. Asking questions should not be
immediately be discouraged, even silly questions. If it is really a FAQ,
you can simply point to the FAQ or add the answer in the FAQ list if it
is missing there.

Well, different groups has different "personality", just don't be
scared.

 

[bonono]

You're right in pointing out that the advantage of ordered dictionaries
(unless you use an omptimized C implementation) is not a performance gain.

But please see my other reply: If the dictionary has more than 3 items
(say 10 or 20), and an effective ordered dict is used, it's not really
"a lot" slower. At least if we are talking about a situation were "on
demand" is "always". So, on the other side there isn't such a big
performance loss when using ordered dictionaries as well.

The advantage of using an ordered dictionary is that you can set up your
ordered dictionary (say, describing your database columns) once, and
then can access it in any way you like in the following: Iterate over it
in a guaranteed order or access item, always refering to the same
object, without needing to care about building and caching auxiliary
objects with different names depending on what you are doing.

Well, I don't think performance is the concern(at least not for me),
but how best to blend with the rest of the code which I have no
interest to explain as I am not here to convincing anyone for such a
thing. I just present a use case, if they see it, fine. If not, that is
fine too.

But I did learn something that creating a dict on a list cost me
nothing, I would be less worry about the performance hit in the future.

 

[A.M. Kuchling]

Does anyone actually use this site? While the Vaults offered a nice
place and a nice interface the Cheese Shop has the appeal of a code
slum.

Looking at the Cheese Shop's home page at
http://cheeseshop.python.org/pypi, which lists recent package updates,
three packages were updated on 11/22, and three on 11/21. Two on
11/20, six on 11/19 (someone was busy!).

Looking at the Vaults's 'latest' page at
http://py.vaults.ca/apyllo.py?a=l, two packages were updated on 08/23,
and five on 08/06.

What would improve the Cheese Shop's interface for you?

--amk

 

[Kay Schluehr]

Looking at the Cheese Shop's home page at
http://cheeseshop.python.org/pypi, which lists recent package updates,
three packages were updated on 11/22, and three on 11/21. Two on
11/20, six on 11/19 (someone was busy!).

Looking at the Vaults's 'latest' page at
http://py.vaults.ca/apyllo.py?a=l, two packages were updated on 08/23,
and five on 08/06.

What would improve the Cheese Shop's interface for you?

--amk

From the treasures of the vaults to cheese shops - an upside down

evolution Personally I find the usability of the Vaults quite well
and it's design o.k. I guess it is a technical detail that causes the
Vaults to be phased out.

If I'd try something more innovative by myself I'd think about
accessing and filtering packages through the Python console itself.
Probably an enhanced standard console enabling SQL commands. Updates of
a local database ( e.g. SQLite ) should be cheap and may be performed
by only one request. Searching, filtering, loading, installing would be
done in a Python+SQL environment.

Kay

 

[Magnus Lycka]

I still believe that the concept of an "ordered dictionary" ("behave
like dict, only keep the order of the keys") is intuitive and doesn't
give you so much scope for ambiguity.

Sure. Others think so too. The problem is that if you and these other
people actually write down exactly how this unambigous ordered dict
should behave, you'll end up with a dozen different sets of semantics,
which can be divided in at least three main groups.

People use different idioms, and often gather collections of classes
and functions that fit their style of coding and their typical problem
domains. There is nothing wrong with that, and they might well be made
publically available if they are more generally useful.

When adding things to the standard library, there are some more things
to consider, particularly for something general such as an odict class:
Is is general enough, and does it add enough value to outweigh the
increased cognitive burden of more classes/functions/types in the
library?

For a beginner, it's easy to believe that things would be easier if
the tool you use had already solved the problem you have at hand for
you, but it turns out that the world has so many problems, that you
would end up with a impenetrable forest of standard library modules
if we actually tried to achieve this.

 

[Fuzzyman]

Of course ours is ordered *and* orderable ! You can explicitly alter
the sequence attribute to change the ordering.

I think we're looking at improving performance based on some of the
suggestions here - as well as possibly widening it to include some of
the alternative use cases. (Or at least Nicola Larosa will do it and
I'll criticise it).

All for another day though...

Fuzzyman
http://www.voidspace.org.uk/python/index.shtml

 

[Alex Martelli]

Maybe I shouldn't have talked about Perl when I'm an ignoramus about
that language... You're right, Perl has unordered arrays. That was new
to me since I associate the term "array" always with "ordered" and I
just remembered that PHP's assoc arrays are similar to Perl's but in
fact, and the examples I have read did not mention about that problem.

Perl's _arrays_ are a bit like Python _lists_, and ordered; it's the
_hashes_ that are a bit like Python _dicts_, and unordered. PHP's use
of "array" for both concepts may indeed be a bit confusing.

You can conclude that PHP's assoc arrays are ordered from the fact that
the language provides a ksort() function to order the keys. And I think
PHP's insertion order is the one I mentioned in my last post.

Anyway, it would be interesting to examine this in detail and how this
is implemented in other languages.

Yep. After just a bit of research I suspect you're right re PHP but
haven't found a specific reference-manual page URL about it.

But it would not satisfy the concept of "keys of a dictionary" which are
always unique.

Why not? Since keys are unique, any 'sequence' of keys is a permutation
of a set, a perfectly well-defined concept.

BTW, there are some boundary conditions that should be fulfilled for the
insertion order, most obviously:

If you define an ordered dict that way:

d = odict()
d['a'] = 1
d['b'] = 2
d['c'] = 3

The keys should then be orderes as ('a', 'b', 'c').

Yep, but both 'first-insertion' and 'latest-insertion' (and many other
rules) meet that constraint.

That would be also biased (in favour of Python) by the fact that
probably very little people would look for and use the package in the
cheese shop if they were looking for ordered dicts. I for example would
probably use ordered dicts if they were part of the standard lib, but
not if I have to install it as an obscure separate package. So I don't
think that will give us a real clue how many people would like ordered
dicts in the standard lib.

A package on cheese shop need not be "obscure" -- that depends on
announcing it well, etc. And the standard library is so huge that many
things inside IT are in fact obscure -- I find myself often pointing out
standard-library solutions to rather experienced Pythonistas who had
been rolling their own, unaware of the stdlib alternative (thus making
the stdlib even bigger has a cost...). So, I don't think this effect
invalidates the experiment; while not all who download an add-on would
like it in the stdlib, and vice versa, there is surely a correlation
between amount of interest/need for the add-on's functionality, and both
rate of downloads as well as interest in having it in the stdlib.

But anyway, if I find some time, I will research a little bit more about
the issue and create such a package, because it seems to me that the
existing packages and recipes are not really satisfying and you're right
it seems to be reasonably easy. It's on my todo list now...

It presumably should have a C implementation for speed and a Python one
for wide availability and ease of installation (the latter gives all the
advantages of prototyping in fixing the API &c, and is made especially
easy by the existence of UserDict.DictMixin in the stdlib). I'll be
glad to help out with the C part when it gets to that, just holler...


Alex

 

[Alex Martelli]

But math folks usually name things after the person(s) who came
up with the idea, not just some random implementer. The idea of

Wrong: you're forgetting Stigler's Law of Misonomy (which I imagine must
have NOT been discovered by Stigler...;-). The Poisson distribution was
in fact described earlier by Bernoulli, Gosset's z-test is universally
known as Student's t-test, etc, etc.

Salsburg's delightful "The Lady Tasting Tea" has a lot of fun with
Stigler's law in the footnotes...;-)


Alex

 

[Stuart McGraw]

What would improve the Cheese Shop's interface for you?

Getting rid of those damn top level links to old versions.
Seeing a long list of old versions, when 99% of visitors are
only interested in the current version, is just visual noise,
and really lame. Move the old version links onto the page
describing the software.

 

[Anton Vredegoor]

But of course, it will always be slower since it is constructed on top
of the built-in dict. In end effect, you always have to maintain a
sequence *plus* a dictionary, which will be always slower than a sheer
dictionary. The ordered dictionary class just hides this uglyness of
having to maintain a dictionary plus a sequence, so it's rather an issue
of convenience in writing and reading programs than a performance issue.

It may be different if the ordered dict would be implemented directly as
an ordered hash table in C.

The problem with hashing is that one is storing data from a possibly
wildly varying range in a set of values with a limited range. That's
where the ordering problems come from in the first place. If one wants
to solve it once and for all one has to give up the speed advantage
that makes hashing so popular. I wonder if optimized C code using
bisect would be very much slower for small ranges?

The current set implementation uses dicts to implement sets, while sets
are a more basic data type than dicts. At least dicts and sets should
be derived from the same type. Just speaking from an intuitive point of
view of course

Here's a sorted dict implementation without using hashes, which maybe
would be fast if implemented in C. The insertion order can be retrieved
using the keys and values lists from the object itself, items() gives a
sorted sequence.

Anton.

NB warning untested code.

When using mutables as keys which is possible by this implementation,
don't change the keys after they're used in the object.

from array import array

class VDict:

def __init__(self,sequence = []):
self.keys = []
self.values = []
self.unranks = array('L',[])
for key,value in sequence:
self[key] = value

def __setitem__(self,key,value):
keys = self.keys
values = self.values
unranks = self.unranks
n = len(keys)
i = self.bisect_left(key)
if i == n or keys[unranks] != key:
keys.append(key)
values.append(value)
unranks.insert(i,n)
else:
values = value

def __getitem__(self,key):
i = self.bisect_left(key)
return self.values[self.unranks]

def bisect_left(self,key, lo=0, hi=None):
keys = self.keys
unranks = self.unranks
if hi is None:
hi = len(keys)
while lo < hi:
mid = (lo+hi)//2
if keys[unranks[mid]] < key:
lo = mid+1
else:
hi = mid
return lo

def __contains__(self,key):
keys = self.keys
unranks = self.unranks
n = len(keys)
i = self.bisect_left(key)
return i < n and keys[unranks] == key

def __len__(self):
return len(self.keys)

def items(self):
keys = self.keys
values = self.values
unranks = self.unranks
return [(keys,values) for i in unranks]

def __iter__(self):
return iter(self.items())

def remove(self,key):
keys = self.keys
values = self.values
unranks = self.unranks
n = len(keys)
i = self.bisect_left(key)
x = unranks
if i < n and keys[x] == key:
del keys[x]
del values[x]
del unranks
for j,k in enumerate(unranks):
if k > x:
unranks[j] -= 1

def test():
V = VDict()
L = [1,2,3]
V[L] = 10
print V[L]
V[L] = 20
print V[L]
V.remove(L)
print V.items()
V = VDict(zip('edcba',range(5)))
print V.items()
print V['d']
V.remove('d')
print V.items()

if __name__=='__main__':
test()

 

[Ben Sizer]

pointers to the members of each pair, yes. but a pointer copy is a
cheap operation (for the given use case, we're only talking about a
few dozen pairs anyway, at the most).

I was really thinking more about the other work, such as the hashing
and whatever, but I guess that is very efficient anyway.

this is a common fallacy; Python programmers underestimate the
cost of adding extra layers to their code (e.g. by using an ordered
dict data structure that has to incrementally update both a list and
a dictionary), and overestimate the cost of letting the C layer do
bulk operations.

If it was me I would probably have just used a list and searched it
linearly: premature optimisation is the root of all evil, etc. But then
I've never found a need for an ordered dictionary anyway; I always felt
they were more an artifact of the language implementation than a
reflection of something inherently useful.

However, you have to forgive people for falling prey to the 'fallacy'
you describe - for years there's been an attempt to teach people to use
proper data structures and algorithms instead of relying on
micro-optimisations (ie. "it's too slow: redo it in assembly"). So
often, the first port of call for a good programmer will be to try and
find a structure that maps directly to the problem.

 

[Fredrik Lundh]

Stuart McGraw wrote

Getting rid of those damn top level links to old versions.
Seeing a long list of old versions, when 99% of visitors are
only interested in the current version, is just visual noise,
and really lame. Move the old version links onto the page
describing the software.

hmm? the pypi package automatically hides old versions when
you post new ones, and it's been that way for ages...

(which is bloody annoying if you're a package developers, since it
means that alphas for the next release hides the most recent stable
version)

looking at the full index, ZODB seems to be the only package that's
available in more than just one stable and one development version...

</F>

 

[Bengt Richter]

Hmmm.. it's *the* repository for Python code. I expect quite a few
people use it...

I hadn't realized how much stuff was there. I generally google for stuff,
but I will be looking directly now.

BTW, I made a mod[1] to your odict that I think/hope is going to be generally faster.
It requires 2.4 though. It passes the same doctest, so its probably close to the same.
It stores the ordering info differently, but is also a dict subclass.

Do you happen to have a timing test that exercises various aspects, so I can try it
before I embarrass myself? Otherwise I guess I'll write something.

Would the would-be users chime in with some idea of what kinds of operations are
most important timing-wise? Which would get the most use? How dynamic would ordering typically be?

[1] fairly pervasive little mods actually
[ 9:15] C:\pywk\clp>diff odict.py odictb.py |wc
146 458 4948

[ 9:15] C:\pywk\clp>wc odict*.py
467 1228 12483 odict.py
511 1500 14728 odictb.py
978 2728 27211 Totals

Regards,
Bengt Richter

 

[Stuart McGraw]

Stuart McGraw wrote


hmm? the pypi package automatically hides old versions when
you post new ones, and it's been that way for ages...

(which is bloody annoying if you're a package developers, since it
means that alphas for the next release hides the most recent stable
version)

looking at the full index, ZODB seems to be the only package that's
available in more than just one stable and one development version...

http://cheeseshop.python.org/pypi?:action=browse&asdf=405
- ClientForm-0.1.17
- ClientForm-0.2.1b
....
- EmPy_3.1
- EmPy_3.1.1
- EmPy_3.2
- EmPy_3.3
....
- FauxIdent-1.1
- FauxIdent-1.2
- FauxIdent-1.2.1
....
Well, it is better than I remember it being a while (year?) ago, my
recollection is that many packages had many, many old versions
listed but now I usualy see only a couple versions.

Hmm, so two versions means one is a development version,
and the other is a stable version? I did not know that, and did
not see it documented on the site. I would say documenting
that would be an interface improvement.

I still think it would be better to have just a package name
(with current version) listed in the index page(s), and have alternate
versions (old, alpha testing, etc) listed on the package's description
page.

 

[Bengt Richter]

What I don't understand is why legitimate questions such as "why are
there no ordered dictionaries" are immediately interpreted as
*complaints* and not just as questions. If I ask such a question, I am
not complaining but trying to simply figure out *why* there is no such
thing. Probably there are reasons and all I want to know is find these
reasons and learn a little bit more about Python in doing so.

Why can't such questions be discussed in a factual, calm and friendly way?

Sorry, I was tired and vented some impatience. I'm mostly friendly ;-)
I took the "why" in a different sense than it was meant, I guess.
Sort of like hearing "why haven't I been served yet" in a restaurant, and
having to say, "I don't work here, you'll have to ask the waiter."

I don't think that this was true in this case, and even if this is the
outcome, those who asked the question will have learned something. I agree again.

I think a discussion group is not there for only presenting mature,
sophisticated thoughts and concepts, but also for "thinking loud"
together with other about these issues. We all know that clarifying our
thoughts works often best if you discuss them with others. And I think
that's one purpose of discussion lists. Asking questions should not be
immediately be discouraged, even silly questions. If it is really a FAQ,
you can simply point to the FAQ or add the answer in the FAQ list if it
is missing there.

Agreed again. Thank you for your nice reply.

Regards,
Bengt Richter

 

[Fredrik Lundh]

Hmm, so two versions means one is a development version,
and the other is a stable version? I did not know that, and did
not see it documented on the site. I would say documenting
that would be an interface improvement.

well, that's up to the developer. when you upload a new version, all
older ones are automagically hidden. the only way to make old versions
appear again is to "unhide" them via a web form. for the few packages
I sampled, the older versions were stable, and the latest one was "less
stable", but I didn't check all of the...

I still think it would be better to have just a package name
(with current version) listed in the index page(s), and have alternate
versions (old, alpha testing, etc) listed on the package's description
page.

agreed.

a "nonstable"-property in the setup file would be nice too (so that stable
versions don't disappear when you upload an alpha or beta...)

</F>

 

[Bengt Richter]

Beat me. I am not the one asking the question.

Sorry, I thought you wanted an ordered dict too.

Lose you. don't know what you want to say.

I like solving problems. I just get frustrated when people don't focus on getting
the problem defined, which IME is 2/3 of the way to a solution. I don't mind,
in fact enjoy, rambling musings, but if someone seems actually to want a solution
for something, I like to try to realize it concretely.

After finally reading that the odict.py in PyPI by Larosa/Foord was what was desired,
but slower in use than what Fredrik posted, I decided to see if I could speed up odict.py.
I now have a version that I think may be generally faster.

I still don't know whether it will be of any user w.r.t. the requirements of anyone
on the bandwagon of asking for some kind of ordered dict, but we'll see what we'll see ;-)

Regards,
Bengt Richter

 

[Bengt Richter]

d = OrderedDict(); d[1]='one'; d[2]='two' =>> list(d) => [1, 2]
ok, now we do d[1]='ein' and what is the order? list(d) => [2, 1] ??
Or do replacements not count as "insertions"?

If you simply set a value for a key that already exists, the order
should not be changed. I think this is intuitive.

Or maybe you want to permit append and NOT prevent
[('a',1), ('a':2)] and maybe d['a'] => [1, 2] ???

You could ask the same question about dict. I think that is not an
option. Why should you want odict behave different than dict?

Well, it was beginning to remind of RDB with possible non-unique keys,
where a select can get you multiple records back.

I still believe that the concept of an "ordered dictionary" ("behave
like dict, only keep the order of the keys") is intuitive and doesn't
give you so much scope for ambiguity. But probably I need to work on an
implementation to become more clear about possible hidden subtleties.

Does that mean that the Larosa/Foord odict.py implementation in PyPI
does not do what you want?

Regards,
Bengt Richter

 

[Christoph Zwerschke]

> Perl's _arrays_ are a bit like Python _lists_, and ordered; it's the
> _hashes_ that are a bit like Python _dicts_, and unordered. PHP's use
> of "array" for both concepts may indeed be a bit confusing.

Perl's _hashes_ have been also called _associative arrays_ originally.

Ok, I just did a little research an compared support for ordered dicts
in some other languages:

* Perl has hashes (associative arrays) which are not ordered.
Here also people are asking for and implementing "ordered hashes",
e.g. http://perltraining.com.au/tips/2005-06-29.html
http://search.cpan.org/dist/Tie-IxHash/lib/Tie/IxHash.pm
http://search.cpan.org/dist/Tie-InsertOrderHash/InsertOrderHash.pm
http://www.yapc.org/America/previous-years/19100/schedule/author/pinyan.html

* Ruby hashes are not ordered.
Again people are asking for and implementing "ordered hashes".
e.g. http://raa.ruby-lang.org/project/orderedhash/
http://groups.google.com/group/comp.lang.ruby/browse_frm/thread/8ebe8d1c5830c801/6428a870925f23f4

* Smalltalk: Innately has unordered Dictionaries.
People are asking for and implementing "OrderedDictionaries" as well,
e.g. http://exept.eu.org:8080/ClassDoc/classDocOf:,OrderedDictionary

* Lisp has (ordered) "association lists".

* PHP has ordered associative arrays.

* Awk, TCL: Associative arrays are unordered.

* C++ has a Map template in the STL which is ordered (a "Sorted
Associative Container").

* Java: Has "LinkedHashMap" which is ordered.

So ordered dictionaries don't seem to be such an exotic idea.

All implementations I found were pretty unequivocally the same that I
had in mind, using insertion order, appending the latest inserted keys
at the end, not changing the order if an existing key is re-inserted,
and deleting keys together with entries.

I also found a discussion thread like the current where similar
arguments were mentioned for and against ordered dictionaries:

In http://mail.python.org/pipermail/python-dev/2005-March/052041.html,
Nick Coghlan raises the following rhetorical question:

Would the default semantics below really be that suprising?

"An ordered dictionary remembers the order in which keys are first seen
and, when iterated over, returns entries based on that order. This
applies to direct iteration, iteration over values and (key, value)
pairs, to the list-producing methods (i.e. keys(), values() and items())
and to any other operations that involve implicit iteration (e.g.
converting to a string representation). Overwriting an entry replaces
its value, but does not affect its position in the key order. Removing
an entry (using 'del') _does_ remove it from the key order. Accordingly,
if the entry is later recreated, it will then occur last in the key
order. This behaviour is analagous to that of a list constructed using
only list.append() to add items (indeed, the key order can be thought of
as a list constructed in this fashion, with keys appended to the list
when they are first encountered)."

This was also the semantics I immediately had in mind when I thought
about ordered dictionaries, though I hadn't read anything about ordered
dictionaries before and it is the same semantics that I found others
have implemented in other languages.

I can't help but I still find it unambiguous and intuitive enough to
consider it "the one" standard implementation for ordered dictionaries.

Also, in the use cases mentioned (describing database columns, html form
fields, configuration parameters etc.), the dictionary is usually only
created once and then not changed, so different handling of re-insertion
or deletion of keys would not even be relevant in these cases.

-- Christoph