2 :mod:`collections` --- High-performance container datatypes
3 ===========================================================
5 .. module:: collections
6 :synopsis: High-performance datatypes
7 .. moduleauthor:: Raymond Hettinger <python@rcn.com>
8 .. sectionauthor:: Raymond Hettinger <python@rcn.com>
14 from collections import *
16 __name__ = '<doctest>'
18 This module implements high-performance container datatypes. Currently,
19 there are two datatypes, :class:`deque` and :class:`defaultdict`, and
20 one datatype factory function, :func:`namedtuple`.
22 .. versionchanged:: 2.5
23 Added :class:`defaultdict`.
25 .. versionchanged:: 2.6
26 Added :func:`namedtuple`.
28 The specialized containers provided in this module provide alternatives
29 to Python's general purpose built-in containers, :class:`dict`,
30 :class:`list`, :class:`set`, and :class:`tuple`.
32 Besides the containers provided here, the optional :mod:`bsddb`
33 module offers the ability to create in-memory or file based ordered
34 dictionaries with string keys using the :meth:`bsddb.btopen` method.
36 In addition to containers, the collections module provides some ABCs
37 (abstract base classes) that can be used to test whether a class
38 provides a particular interface, for example, is it hashable or
41 .. versionchanged:: 2.6
42 Added abstract base classes.
44 ABCs - abstract base classes
45 ----------------------------
47 The collections module offers the following ABCs:
49 ========================= ===================== ====================== ====================================================
50 ABC Inherits Abstract Methods Mixin Methods
51 ========================= ===================== ====================== ====================================================
52 :class:`Container` ``__contains__``
53 :class:`Hashable` ``__hash__``
54 :class:`Iterable` ``__iter__``
55 :class:`Iterator` :class:`Iterable` ``__next__`` ``__iter__``
56 :class:`Sized` ``__len__``
57 :class:`Callable` ``__call__``
59 :class:`Sequence` :class:`Sized`, ``__getitem__`` ``__contains__``. ``__iter__``, ``__reversed__``.
60 :class:`Iterable`, and ``__len__`` ``index``, and ``count``
63 :class:`MutableSequence` :class:`Sequence` ``__getitem__`` Inherited Sequence methods and
64 ``__delitem__``, ``append``, ``reverse``, ``extend``, ``pop``,
65 ``insert``, ``remove``, and ``__iadd__``
68 :class:`Set` :class:`Sized`, ``__len__``, ``__le__``, ``__lt__``, ``__eq__``, ``__ne__``,
69 :class:`Iterable`, ``__iter__``, and ``__gt__``, ``__ge__``, ``__and__``, ``__or__``
70 :class:`Container` ``__contains__`` ``__sub__``, ``__xor__``, and ``isdisjoint``
72 :class:`MutableSet` :class:`Set` ``add`` and Inherited Set methods and
73 ``discard`` ``clear``, ``pop``, ``remove``, ``__ior__``,
74 ``__iand__``, ``__ixor__``, and ``__isub__``
76 :class:`Mapping` :class:`Sized`, ``__getitem__``, ``__contains__``, ``keys``, ``items``, ``values``,
77 :class:`Iterable`, ``__len__``. and ``get``, ``__eq__``, and ``__ne__``
78 :class:`Container` ``__iter__``
80 :class:`MutableMapping` :class:`Mapping` ``__getitem__`` Inherited Mapping methods and
81 ``__setitem__``, ``pop``, ``popitem``, ``clear``, ``update``,
82 ``__delitem__``, and ``setdefault``
86 :class:`MappingView` :class:`Sized` ``__len__``
87 :class:`KeysView` :class:`MappingView`, ``__contains__``,
88 :class:`Set` ``__iter__``
89 :class:`ItemsView` :class:`MappingView`, ``__contains__``,
90 :class:`Set` ``__iter__``
91 :class:`ValuesView` :class:`MappingView` ``__contains__``, ``__iter__``
92 ========================= ===================== ====================== ====================================================
94 These ABCs allow us to ask classes or instances if they provide
95 particular functionality, for example::
98 if isinstance(myvar, collections.Sized):
101 Several of the ABCs are also useful as mixins that make it easier to develop
102 classes supporting container APIs. For example, to write a class supporting
103 the full :class:`Set` API, it only necessary to supply the three underlying
104 abstract methods: :meth:`__contains__`, :meth:`__iter__`, and :meth:`__len__`.
105 The ABC supplies the remaining methods such as :meth:`__and__` and
106 :meth:`isdisjoint` ::
108 class ListBasedSet(collections.Set):
109 ''' Alternate set implementation favoring space over speed
110 and not requiring the set elements to be hashable. '''
111 def __init__(self, iterable):
112 self.elements = lst = []
113 for value in iterable:
117 return iter(self.elements)
118 def __contains__(self, value):
119 return value in self.elements
121 return len(self.elements)
123 s1 = ListBasedSet('abcdef')
124 s2 = ListBasedSet('defghi')
125 overlap = s1 & s2 # The __and__() method is supported automatically
127 Notes on using :class:`Set` and :class:`MutableSet` as a mixin:
130 Since some set operations create new sets, the default mixin methods need
131 a way to create new instances from an iterable. The class constructor is
132 assumed to have a signature in the form ``ClassName(iterable)``.
133 That assumption is factored-out to an internal classmethod called
134 :meth:`_from_iterable` which calls ``cls(iterable)`` to produce a new set.
135 If the :class:`Set` mixin is being used in a class with a different
136 constructor signature, you will need to override :meth:`from_iterable`
137 with a classmethod that can construct new instances from
138 an iterable argument.
141 To override the comparisons (presumably for speed, as the
142 semantics are fixed), redefine :meth:`__le__` and
143 then the other operations will automatically follow suit.
146 The :class:`Set` mixin provides a :meth:`_hash` method to compute a hash value
147 for the set; however, :meth:`__hash__` is not defined because not all sets
148 are hashable or immutable. To add set hashabilty using mixins,
149 inherit from both :meth:`Set` and :meth:`Hashable`, then define
150 ``__hash__ = Set._hash``.
152 (For more about ABCs, see the :mod:`abc` module and :pep:`3119`.)
158 :class:`deque` objects
159 ----------------------
162 .. class:: deque([iterable[, maxlen]])
164 Returns a new deque object initialized left-to-right (using :meth:`append`) with
165 data from *iterable*. If *iterable* is not specified, the new deque is empty.
167 Deques are a generalization of stacks and queues (the name is pronounced "deck"
168 and is short for "double-ended queue"). Deques support thread-safe, memory
169 efficient appends and pops from either side of the deque with approximately the
170 same O(1) performance in either direction.
172 Though :class:`list` objects support similar operations, they are optimized for
173 fast fixed-length operations and incur O(n) memory movement costs for
174 ``pop(0)`` and ``insert(0, v)`` operations which change both the size and
175 position of the underlying data representation.
177 .. versionadded:: 2.4
179 If *maxlen* is not specified or is *None*, deques may grow to an
180 arbitrary length. Otherwise, the deque is bounded to the specified maximum
181 length. Once a bounded length deque is full, when new items are added, a
182 corresponding number of items are discarded from the opposite end. Bounded
183 length deques provide functionality similar to the ``tail`` filter in
184 Unix. They are also useful for tracking transactions and other pools of data
185 where only the most recent activity is of interest.
187 .. versionchanged:: 2.6
188 Added *maxlen* parameter.
190 Deque objects support the following methods:
193 .. method:: append(x)
195 Add *x* to the right side of the deque.
198 .. method:: appendleft(x)
200 Add *x* to the left side of the deque.
205 Remove all elements from the deque leaving it with length 0.
208 .. method:: extend(iterable)
210 Extend the right side of the deque by appending elements from the iterable
214 .. method:: extendleft(iterable)
216 Extend the left side of the deque by appending elements from *iterable*.
217 Note, the series of left appends results in reversing the order of
218 elements in the iterable argument.
223 Remove and return an element from the right side of the deque. If no
224 elements are present, raises an :exc:`IndexError`.
227 .. method:: popleft()
229 Remove and return an element from the left side of the deque. If no
230 elements are present, raises an :exc:`IndexError`.
233 .. method:: remove(value)
235 Removed the first occurrence of *value*. If not found, raises a
238 .. versionadded:: 2.5
241 .. method:: rotate(n)
243 Rotate the deque *n* steps to the right. If *n* is negative, rotate to
244 the left. Rotating one step to the right is equivalent to:
245 ``d.appendleft(d.pop())``.
248 In addition to the above, deques support iteration, pickling, ``len(d)``,
249 ``reversed(d)``, ``copy.copy(d)``, ``copy.deepcopy(d)``, membership testing with
250 the :keyword:`in` operator, and subscript references such as ``d[-1]``. Indexed
251 access is O(1) at both ends but slows to O(n) in the middle. For fast random
252 access, use lists instead.
258 >>> from collections import deque
259 >>> d = deque('ghi') # make a new deque with three items
260 >>> for elem in d: # iterate over the deque's elements
261 ... print elem.upper()
266 >>> d.append('j') # add a new entry to the right side
267 >>> d.appendleft('f') # add a new entry to the left side
268 >>> d # show the representation of the deque
269 deque(['f', 'g', 'h', 'i', 'j'])
271 >>> d.pop() # return and remove the rightmost item
273 >>> d.popleft() # return and remove the leftmost item
275 >>> list(d) # list the contents of the deque
277 >>> d[0] # peek at leftmost item
279 >>> d[-1] # peek at rightmost item
282 >>> list(reversed(d)) # list the contents of a deque in reverse
284 >>> 'h' in d # search the deque
286 >>> d.extend('jkl') # add multiple elements at once
288 deque(['g', 'h', 'i', 'j', 'k', 'l'])
289 >>> d.rotate(1) # right rotation
291 deque(['l', 'g', 'h', 'i', 'j', 'k'])
292 >>> d.rotate(-1) # left rotation
294 deque(['g', 'h', 'i', 'j', 'k', 'l'])
296 >>> deque(reversed(d)) # make a new deque in reverse order
297 deque(['l', 'k', 'j', 'i', 'h', 'g'])
298 >>> d.clear() # empty the deque
299 >>> d.pop() # cannot pop from an empty deque
300 Traceback (most recent call last):
301 File "<pyshell#6>", line 1, in -toplevel-
303 IndexError: pop from an empty deque
305 >>> d.extendleft('abc') # extendleft() reverses the input order
307 deque(['c', 'b', 'a'])
312 :class:`deque` Recipes
313 ^^^^^^^^^^^^^^^^^^^^^^
315 This section shows various approaches to working with deques.
317 The :meth:`rotate` method provides a way to implement :class:`deque` slicing and
318 deletion. For example, a pure python implementation of ``del d[n]`` relies on
319 the :meth:`rotate` method to position elements to be popped::
321 def delete_nth(d, n):
326 To implement :class:`deque` slicing, use a similar approach applying
327 :meth:`rotate` to bring a target element to the left side of the deque. Remove
328 old entries with :meth:`popleft`, add new entries with :meth:`extend`, and then
329 reverse the rotation.
330 With minor variations on that approach, it is easy to implement Forth style
331 stack manipulations such as ``dup``, ``drop``, ``swap``, ``over``, ``pick``,
332 ``rot``, and ``roll``.
334 Multi-pass data reduction algorithms can be succinctly expressed and efficiently
335 coded by extracting elements with multiple calls to :meth:`popleft`, applying
336 a reduction function, and calling :meth:`append` to add the result back to the
339 For example, building a balanced binary tree of nested lists entails reducing
340 two adjacent nodes into one by grouping them in a list:
342 >>> def maketree(iterable):
343 ... d = deque(iterable)
344 ... while len(d) > 1:
345 ... pair = [d.popleft(), d.popleft()]
349 >>> print maketree('abcdefgh')
350 [[[['a', 'b'], ['c', 'd']], [['e', 'f'], ['g', 'h']]]]
352 Bounded length deques provide functionality similar to the ``tail`` filter
355 def tail(filename, n=10):
356 'Return the last n lines of a file'
357 return deque(open(filename), n)
359 .. _defaultdict-objects:
361 :class:`defaultdict` objects
362 ----------------------------
365 .. class:: defaultdict([default_factory[, ...]])
367 Returns a new dictionary-like object. :class:`defaultdict` is a subclass of the
368 builtin :class:`dict` class. It overrides one method and adds one writable
369 instance variable. The remaining functionality is the same as for the
370 :class:`dict` class and is not documented here.
372 The first argument provides the initial value for the :attr:`default_factory`
373 attribute; it defaults to ``None``. All remaining arguments are treated the same
374 as if they were passed to the :class:`dict` constructor, including keyword
377 .. versionadded:: 2.5
379 :class:`defaultdict` objects support the following method in addition to the
380 standard :class:`dict` operations:
383 .. method:: defaultdict.__missing__(key)
385 If the :attr:`default_factory` attribute is ``None``, this raises a
386 :exc:`KeyError` exception with the *key* as argument.
388 If :attr:`default_factory` is not ``None``, it is called without arguments
389 to provide a default value for the given *key*, this value is inserted in
390 the dictionary for the *key*, and returned.
392 If calling :attr:`default_factory` raises an exception this exception is
393 propagated unchanged.
395 This method is called by the :meth:`__getitem__` method of the
396 :class:`dict` class when the requested key is not found; whatever it
397 returns or raises is then returned or raised by :meth:`__getitem__`.
400 :class:`defaultdict` objects support the following instance variable:
403 .. attribute:: defaultdict.default_factory
405 This attribute is used by the :meth:`__missing__` method; it is
406 initialized from the first argument to the constructor, if present, or to
410 .. _defaultdict-examples:
412 :class:`defaultdict` Examples
413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
415 Using :class:`list` as the :attr:`default_factory`, it is easy to group a
416 sequence of key-value pairs into a dictionary of lists:
418 >>> s = [('yellow', 1), ('blue', 2), ('yellow', 3), ('blue', 4), ('red', 1)]
419 >>> d = defaultdict(list)
424 [('blue', [2, 4]), ('red', [1]), ('yellow', [1, 3])]
426 When each key is encountered for the first time, it is not already in the
427 mapping; so an entry is automatically created using the :attr:`default_factory`
428 function which returns an empty :class:`list`. The :meth:`list.append`
429 operation then attaches the value to the new list. When keys are encountered
430 again, the look-up proceeds normally (returning the list for that key) and the
431 :meth:`list.append` operation adds another value to the list. This technique is
432 simpler and faster than an equivalent technique using :meth:`dict.setdefault`:
436 ... d.setdefault(k, []).append(v)
439 [('blue', [2, 4]), ('red', [1]), ('yellow', [1, 3])]
441 Setting the :attr:`default_factory` to :class:`int` makes the
442 :class:`defaultdict` useful for counting (like a bag or multiset in other
445 >>> s = 'mississippi'
446 >>> d = defaultdict(int)
451 [('i', 4), ('p', 2), ('s', 4), ('m', 1)]
453 When a letter is first encountered, it is missing from the mapping, so the
454 :attr:`default_factory` function calls :func:`int` to supply a default count of
455 zero. The increment operation then builds up the count for each letter.
457 The function :func:`int` which always returns zero is just a special case of
458 constant functions. A faster and more flexible way to create constant functions
459 is to use :func:`itertools.repeat` which can supply any constant value (not just
462 >>> def constant_factory(value):
463 ... return itertools.repeat(value).next
464 >>> d = defaultdict(constant_factory('<missing>'))
465 >>> d.update(name='John', action='ran')
466 >>> '%(name)s %(action)s to %(object)s' % d
467 'John ran to <missing>'
469 Setting the :attr:`default_factory` to :class:`set` makes the
470 :class:`defaultdict` useful for building a dictionary of sets:
472 >>> s = [('red', 1), ('blue', 2), ('red', 3), ('blue', 4), ('red', 1), ('blue', 4)]
473 >>> d = defaultdict(set)
478 [('blue', set([2, 4])), ('red', set([1, 3]))]
481 .. _named-tuple-factory:
483 :func:`namedtuple` Factory Function for Tuples with Named Fields
484 ----------------------------------------------------------------
486 Named tuples assign meaning to each position in a tuple and allow for more readable,
487 self-documenting code. They can be used wherever regular tuples are used, and
488 they add the ability to access fields by name instead of position index.
490 .. function:: namedtuple(typename, field_names, [verbose])
492 Returns a new tuple subclass named *typename*. The new subclass is used to
493 create tuple-like objects that have fields accessible by attribute lookup as
494 well as being indexable and iterable. Instances of the subclass also have a
495 helpful docstring (with typename and field_names) and a helpful :meth:`__repr__`
496 method which lists the tuple contents in a ``name=value`` format.
498 The *field_names* are a single string with each fieldname separated by whitespace
499 and/or commas, for example ``'x y'`` or ``'x, y'``. Alternatively, *field_names*
500 can be a sequence of strings such as ``['x', 'y']``.
502 Any valid Python identifier may be used for a fieldname except for names
503 starting with an underscore. Valid identifiers consist of letters, digits,
504 and underscores but do not start with a digit or underscore and cannot be
505 a :mod:`keyword` such as *class*, *for*, *return*, *global*, *pass*, *print*,
508 If *verbose* is true, the class definition is printed just before being built.
510 Named tuple instances do not have per-instance dictionaries, so they are
511 lightweight and require no more memory than regular tuples.
513 .. versionadded:: 2.6
518 :options: +NORMALIZE_WHITESPACE
520 >>> Point = namedtuple('Point', 'x y', verbose=True)
528 def __new__(cls, x, y):
529 return tuple.__new__(cls, (x, y))
532 def _make(cls, iterable, new=tuple.__new__, len=len):
533 'Make a new Point object from a sequence or iterable'
534 result = new(cls, iterable)
536 raise TypeError('Expected 2 arguments, got %d' % len(result))
540 return 'Point(x=%r, y=%r)' % self
543 'Return a new dict which maps field names to their values'
544 return {'x': t[0], 'y': t[1]}
546 def _replace(self, **kwds):
547 'Return a new Point object replacing specified fields with new values'
548 result = self._make(map(kwds.pop, ('x', 'y'), self))
550 raise ValueError('Got unexpected field names: %r' % kwds.keys())
553 def __getnewargs__(self):
556 x = property(itemgetter(0))
557 y = property(itemgetter(1))
559 >>> p = Point(11, y=22) # instantiate with positional or keyword arguments
560 >>> p[0] + p[1] # indexable like the plain tuple (11, 22)
562 >>> x, y = p # unpack like a regular tuple
565 >>> p.x + p.y # fields also accessible by name
567 >>> p # readable __repr__ with a name=value style
570 Named tuples are especially useful for assigning field names to result tuples returned
571 by the :mod:`csv` or :mod:`sqlite3` modules::
573 EmployeeRecord = namedtuple('EmployeeRecord', 'name, age, title, department, paygrade')
576 for emp in map(EmployeeRecord._make, csv.reader(open("employees.csv", "rb"))):
577 print emp.name, emp.title
580 conn = sqlite3.connect('/companydata')
581 cursor = conn.cursor()
582 cursor.execute('SELECT name, age, title, department, paygrade FROM employees')
583 for emp in map(EmployeeRecord._make, cursor.fetchall()):
584 print emp.name, emp.title
586 In addition to the methods inherited from tuples, named tuples support
587 three additional methods and one attribute. To prevent conflicts with
588 field names, the method and attribute names start with an underscore.
590 .. method:: somenamedtuple._make(iterable)
592 Class method that makes a new instance from an existing sequence or iterable.
600 .. method:: somenamedtuple._asdict()
602 Return a new dict which maps field names to their corresponding values::
607 .. method:: somenamedtuple._replace(kwargs)
609 Return a new instance of the named tuple replacing specified fields with new
614 >>> p = Point(x=11, y=22)
618 >>> for partnum, record in inventory.items():
619 ... inventory[partnum] = record._replace(price=newprices[partnum], timestamp=time.now())
621 .. attribute:: somenamedtuple._fields
623 Tuple of strings listing the field names. Useful for introspection
624 and for creating new named tuple types from existing named tuples.
628 >>> p._fields # view the field names
631 >>> Color = namedtuple('Color', 'red green blue')
632 >>> Pixel = namedtuple('Pixel', Point._fields + Color._fields)
633 >>> Pixel(11, 22, 128, 255, 0)
634 Pixel(x=11, y=22, red=128, green=255, blue=0)
636 To retrieve a field whose name is stored in a string, use the :func:`getattr`
642 To convert a dictionary to a named tuple, use the double-star-operator [#]_:
644 >>> d = {'x': 11, 'y': 22}
648 Since a named tuple is a regular Python class, it is easy to add or change
649 functionality with a subclass. Here is how to add a calculated field and
650 a fixed-width print format:
652 >>> class Point(namedtuple('Point', 'x y')):
656 ... return (self.x ** 2 + self.y ** 2) ** 0.5
657 ... def __str__(self):
658 ... return 'Point: x=%6.3f y=%6.3f hypot=%6.3f' % (self.x, self.y, self.hypot)
660 >>> for p in Point(3, 4), Point(14, 5/7.):
662 Point: x= 3.000 y= 4.000 hypot= 5.000
663 Point: x=14.000 y= 0.714 hypot=14.018
665 The subclass shown above sets ``__slots__`` to an empty tuple. This keeps
666 keep memory requirements low by preventing the creation of instance dictionaries.
668 Subclassing is not useful for adding new, stored fields. Instead, simply
669 create a new named tuple type from the :attr:`_fields` attribute:
671 >>> Point3D = namedtuple('Point3D', Point._fields + ('z',))
673 Default values can be implemented by using :meth:`_replace` to
674 customize a prototype instance:
676 >>> Account = namedtuple('Account', 'owner balance transaction_count')
677 >>> default_account = Account('<owner name>', 0.0, 0)
678 >>> johns_account = default_account._replace(owner='John')
680 Enumerated constants can be implemented with named tuples, but it is simpler
681 and more efficient to use a simple class declaration:
683 >>> Status = namedtuple('Status', 'open pending closed')._make(range(3))
684 >>> Status.open, Status.pending, Status.closed
687 ... open, pending, closed = range(3)
689 .. rubric:: Footnotes
691 .. [#] For information on the double-star-operator see
692 :ref:`tut-unpacking-arguments` and :ref:`calls`.