patricia-trie

A pure Python 2.7+ implementation of a PATRICIA trie for effcient matching of string collections on text.

Note that you probably first want to have a look at the Python wrapper marisa-trie or its PyPi package before using particia-trie; according to simple timeit comparisons, these wrappers for the C-based MARISA library are about twice as fast as this pure Python implementation.

patricia-trie does have its merits, however - it is small, clear, and has a very clean interface that imitates the dict API and works with Py3k.

Installation

pip install patricia-trie

Usage

>>> T = trie('root', key='value', king='kong') # a root value and two pairs
>>> T['four'] = None # setting new values as in a dict
>>> '' in T # check if the value exits (note: the [empty] root is '')
True
>>> 'kong' in T # existence checks as in a dict
False
>>> T['king'] # get the value for an exact key ... as in a dict
'kong'
>>> T['kong'] # error from non-existing keys (as in a dict)
Traceback (most recent call last):
    ...
KeyError: 'kong'
>>> len(T) # count keys ("terminals") in the tree
4
>>> sorted(T) # plus "traditional stuff": .keys(), .values(), and .items()
['', 'four', 'key', 'king']
>>> # scanning a string S with key(S), value(S), and item(S):
>>> S = 'keys and kewl stuff'
>>> T.key(S) # report the (longest) key that is a prefix of S
'key'
>>> T.value(S, 9) # using offsets; NB: a root value always matches!
'root'
>>> del T[''] # interlude: deleting keys (here, the root)
>>> T.item(S, 9) # raise error if no key is a prefix of S
Traceback (most recent call last):
    ...
KeyError: 'k'
>>> # info: the error string above contains the matched path so far
>>> T.item(S, 9, default=None) # avoid the error by specifying a default
(None, None)
>>> # iterate all matching content with keys(S), values(S), and items(S):
>>> list(T.items(S))
[('key', 'value')]
>>> T.isPrefix('k') # reverse lookup: check if S is a prefix of any key
True
>>> T.isPrefix('kong')
False
>>> sorted(T.iter('k')) # and get all keys that have S as prefix
['key', 'king']

Deleting entries is a "half-supported" operation only. The key appears "removed", but the trie is not actually changed, only the node state is changed from terminal to non-terminal. I.e., if you frequently delete keys, the compaction will become fragmented and less efficient. To mitigate this effect, make a copy of the trie (using a copy constructor idiom):

T = trie(**T)

If you are only interested in scanning for the presence of keys, but do not care about mapping a value to each key, using None as the value of your keys and scanning with key(S, None, start=i) at every offset i in the string S is perfectly fine (because the return value will be the key string iff a full match was made and None otherwise):

>>> T = trie(present=None)
>>> T.key('is absent here', None, start=3) # start scanning at offset 3
>>> T.key('is present here', None, start=3) # start scanning at offset 3
'present'

API

trie(*value, **branch)

Create a new tree node.
Any arguments will be used as the value of this node.
If keyword arguments are given, they initialize a whole branch.
Note that None is a valid value for a node.

trie.isPrefix(prefix)

Return True if any key starts with prefix.

trie.item(string, start=0, end=None, default=NULL)

Return the key, value pair of the longest key that is a prefix of string (beginning at start and ending at end).
If no key matches, raise a KeyError or return the None, default pair if any default value was set.

trie.items([string [, start [, end ]]])

Return all key, value pairs (for keys that are a prefix of string (beginning at start (and terminating before end))).

trie.iter(prefix)

Return an iterator over all keys that start with prefix.

trie.key(string, start=0, end=None, default=NULL)

Return the longest key that is a prefix of string (beginning at start and ending at end).
If no key matches, raise a KeyError or return the default value if it was set.

trie.keys([string [, start [, end ]]])

Return all keys (that are a prefix of string (beginning at start (and terminating before end))).

trie.value(string, start=0, end=None, default=NULL)

Return the value of the longest key that is a prefix of string (beginning at start and ending at end).
If no key matches, raise a KeyError or return the default value if it was set.

trie.values([string [, start [, end ]]])

Return all values (for keys that are a prefix of string (beginning at start (and terminating before end))).

History

1. Initial release.
2. Update: Full documentation and corrections.

3. Feature: optional keyword parameters to indicate an offset start when scanning a string with the methods key(), keys(), item(), items(), value(), and values(), so it is not necessary to slice strings for each scan:

   >>> # Old usage to scan 'string' in 'the string' was:
   >>> T.keys('the string'[4:])
   >>> # With the new optional keyword parameter:
   >>> T.keys('the string', start=4)

4. Important API change: item() now returns key, value pairs even when a default value is given, using None as the "key":

   >>> # Old behaviour was:
   >>> T.item('string', default=False)
   False
   >>> # While now, the same call produces:
   >>> T.item('string', default=False)
   None, False

   Improvement: Switched from using dictionaries to two-tuple lists internally (thanks to Pedro Gaio for the suggestion!) to improve the overall performance a bit (about 20% faster on simple tests).

5. Bugfix: When splitting edges while adding a new key that is shorter than the current edge, a index error would have occurred.

6. Feature: Added optional keyword parameter end to the methods key(), keys(), item(), items(), value(), and values(), so it is not necessary to scan within a window:

   T.key('string', start=2, end=3, default=None)
   T.keys('string', start=2, end=3)

7. Improvement: Switched back to a very efficient internal dictionary implementation; Runs about two- to three times as fast as the two-tuple list from update 4 against the simple (and newly added) time_patricia.py "benchmark".
8. Bugfix: Correct behavior when using a negative start index. Added a comparison to marisa-trie - by now, it seems, patricia-trie is roughly only a factor two slower than the marisa-trie PyPI version wrapping a C library. Also makes it nice to compare the two usages.
9. Bugfix (15/09/2014): Correct behaviour when using an exactly matching prefix as query (issue described in #1 by @zachrahan). Also fixes code-smells (PEP8, code complexity) and a failing test case code.
10. Bugfix (14/12/2014): Added the missing README to PyPI package. (MANIFEST.in)

Copyright

Copyright 2013, Florian Leitner. All rights reserved.

License

Apache License v2