marisa-trie 0.3.6

marisa_trie.Trie

Create a new trie:

>>> import marisa_trie
>>> trie = marisa_trie.Trie([u'key1', u'key2', u'key12'])

Check if key is in trie:

>>> u'key1' in trie
True
>>> u'key20' in trie
False

Each key is assigned an unique ID from 0 to (n - 1), where n is the number of keys; you can use this ID to store a value in a separate structure (e.g. python list):

>>> trie.key_id(u'key2')
1

Key can be reconstructed from the ID:

>>> trie.restore_key(1)
u'key2'

Find all prefixes of a given key:

>>> trie.prefixes(u'key12')
[u'key1', u'key12']

There is also a generator version of .prefixes method called .iter_prefixes.

Find all keys from this trie that starts with a given prefix:

>> trie.keys(u'key1')
[u'key1', u'key12']

(iterator version .iterkeys(prefix) is also available).

marisa_trie.RecordTrie

Create a new trie:

>>> keys = [u'foo', u'bar', u'foobar', u'foo']
>>> values = [(1, 2), (2, 1), (3, 3), (2, 1)]
>>> fmt = "<HH"   # a tuple with 2 short integers
>>> trie = marisa_trie.RecordTrie(fmt, zip(keys, values))

Trie initial data must be an iterable of tuples (unicode_key, data_tuple). Data tuples will be converted to bytes with struct.pack(fmt, *data_tuple).

Take a look at http://docs.python.org/library/struct.html#format-strings for the format string specification.

Duplicate keys are allowed.

Check if key is in trie:

>>> u'foo' in trie
True
>>> u'spam' in trie
False

Get a values list:

>>> trie[u'bar']
[(2, 1)]
>>> trie[u'foo']
[(1, 2), (2, 1)]
>>> trie.get(u'bar', 123)
[(2, 1)]
>>> trie.get(u'BAAR', 123) # default value
123

Find all prefixes of a given key:

>>> trie.prefixes(u'foobarz')
[u'foo', u'foobar']

Find all keys from this trie that starts with a given prefix:

>> trie.keys(u'fo')
[u'foo', u'foo', u'foobar']

Find all items from this trie that starts with a given prefix:

>> trie.items(u'fo')
[(u'foo', (1, 2)), (u'foo', (2, 1), (u'foobar', (3, 3))]

marisa_trie.BytesTrie

BytesTrie is similar to RecordTrie, but the values are raw bytes, not tuples:

>>> keys = [u'foo', u'bar', u'foobar', u'foo']
>>> values = [b'foo-value', b'bar-value', b'foobar-value', b'foo-value2']
>>> trie = marisa_trie.BytesTrie(zip(keys, values))
>>> trie[u'bar']
[b'bar-value']

Persistence

Trie objects supports saving/loading, pickling/unpickling and memory mapped I/O.

Write trie to a stream:

>>> with open('my_trie.marisa', 'w') as f:
...     trie.write(f)

Save trie to a file:

>>> trie.save('my_trie_copy.marisa')

Read trie from stream:

>>> trie2 = marisa_trie.Trie()
>>> with open('my_trie.marisa', 'r') as f:
...     trie.read(f)

Load trie from file:

>>> trie2.load('my_trie.marisa')

Trie objects are picklable:

>>> import pickle
>>> data = pickle.dumps(trie)
>>> trie3 = pickle.loads(data)

You may also build a trie using marisa-build command-line utility (provided by underlying C++ library; it should be downloaded and compiled separately) and then load the trie from the resulting file using .load() method.

Memory mapped I/O

It is possible to use memory mapped file as data source:

>>> trie = marisa_trie.RecordTrie(fmt).mmap('my_record_trie.marisa')

This way the whole dictionary won’t be loaded to memory; memory mapped I/O is an easy way to share dictionary data among processes.

Trie storage options

marisa-trie C++ library provides some configuration options for trie storage; check http://marisa-trie.googlecode.com/svn/trunk/docs/readme.en.html page (scroll down to “Enumeration Constants” section) to get an idea.

These options are exposed as order, num_tries, cache_size and binary keyword arguments for trie constructors.

For example, set order to marisa_trie.LABEL_ORDER in order to make trie functions return results in alphabetical oder:

>>> trie = marisa_trie.RecordTrie(fmt, data, order=marisa_trie.LABEL_ORDER)

How is source code organized

There are 4 folders in repository:

• bench - benchmarks & benchmark data;

• lib - original unmodified marisa-trie C++ library which is bundled for easier distribution; if something is have to be fixed in this library consider fixing it in the original repo ;

• src - wrapper code; src/marisa_trie.pyx is a wrapper implementation; src/*.pxd files are Cython headers for corresponding C++ headers; src/*.cpp files are the pre-built extension code and shouldn’t be modified directly (they should be updated via update_cpp.sh script).

• tests - the test suite.

Running tests and benchmarks

Make sure tox is installed and run

$ tox

from the source checkout. Tests should pass under python 2.6, 2.7, 3.2.

In order to run benchmarks, type

$ tox -c bench.ini

Authors & Contributors

• Mikhail Korobov <kmike84@gmail.com>

This module is based on marisa-trie C++ library by Susumu Yata & contributors.

0.3.6 (2012-09-05)

• much faster (3x-7x) .items() and .keys() methods for all tries; faster (up to 3x) .prefixes() method for Trie.

0.3.5 (2012-08-30)

• Pickling of RecordTrie is fixed (thanks lazarou for the report);

• error messages should become more useful.

0.3.4 (2012-08-29)

• Issues with mingw32 should be resolved (thanks Susumu Yata).

0.3.3 (2012-08-27)

• .get(key, default=None) method for BytesTrie and RecordTrie;

• small README improvements.

0.3.2 (2012-08-26)

• Small code cleanup;

• load, read and mmap methods returns ‘self’;

• I can’t run tests (via tox) under Python 3.3 so it is removed from supported versions for now.

0.3.1 (2012-08-23)

• .prefixes() support for RecordTrie and BytesTrie.

0.3 (2012-08-23)

• RecordTrie and BytesTrie are introduced;

• IntTrie class is removed (probably temporary?);

• dumps/loads methods are renamed to tobytes/frombytes;

• benchmark & tests improvements;

• support for MARISA-trie config options is added.

0.2 (2012-08-19)

• Pickling/unpickling support;

• dumps/loads methods;

• python 3.3 workaround;

• improved tests;

• benchmarks.

0.1 (2012-08-17)

Initial release.