README.rst

intervaltree

A mutable, self-balancing interval tree for Python 2 and 3. Queries may be by point, by range overlap, or by range envelopment.

This library was designed to allow tagging text and time intervals, where the intervals include the lower bound but not the upper bound.

Installing

pip install intervaltree

Features

• Supports Python 2.6+ and Python 3.2+
• Initializing
  • blank tree = IntervalTree()
  • from an iterable of Interval objects (tree = IntervalTree(intervals))
  • from an iterable of tuples (tree = IntervalTree.from_tuples(interval_tuples))
• Insertions
  • tree[begin:end] = data
  • tree.add(interval)
  • tree.addi(begin, end, data)
• Deletions
  • tree.remove(interval) (raises ValueError if not present)
  • tree.discard(interval) (quiet if not present)
  • tree.removei(begin, end, data) (short for tree.remove(Interval(begin, end, data)))
  • tree.discardi(begin, end, data) (short for tree.discard(Interval(begin, end, data)))
  • tree.remove_overlap(point)
  • tree.remove_overlap(begin, end) (removes all overlapping the range)
  • tree.remove_envelop(begin, end) (removes all enveloped in the range)
• Overlap queries
  • tree[point]
  • tree[begin:end]
  • tree.search(point)
  • tree.search(begin, end)
• Envelop queries
  • tree.search(begin, end, strict=True)
• Membership queries
  • interval_obj in tree (this is fastest, O(1))
  • tree.containsi(begin, end, data)
  • tree.overlaps(point)
  • tree.overlaps(begin, end)
• Iterable
  • for interval_obj in tree:
  • tree.items()
• Sizing
  • len(tree)
  • tree.is_empty()
  • not tree
  • tree.begin() (the begin coordinate of the leftmost interval)
  • tree.end() (the end coordinate of the rightmost interval)
• Set-like operations
  • union
    • result_tree = tree.union(iterable)
    • result_tree = tree1 | tree2
    • tree.update(iterable)
    • tree |= other_tree
  • difference
    • result_tree = tree.difference(iterable)
    • result_tree = tree1 - tree2
    • tree.difference_update(iterable)
    • tree -= other_tree
  • intersection
    • result_tree = tree.intersection(iterable)
    • result_tree = tree1 & tree2
    • tree.intersection_update(iterable)
    • tree &= other_tree
  • symmetric difference
    • result_tree = tree.symmetric_difference(iterable)
    • result_tree = tree1 ^ tree2
    • tree.symmetric_difference_update(iterable)
    • tree ^= other_tree
  • comparison
    • tree1.issubset(tree2) or tree1 <= tree2
    • tree1 <= tree2
    • tree1.issuperset(tree2) or tree1 > tree2
    • tree1 >= tree2
    • tree1 == tree2
• Restructuring
  • chop(begin, end) (slice intervals and remove everything between begin and end)
  • slice(point) (slice intervals at point)
  • split_overlaps() (slice at all interval boundaries)
• Copying and typecasting
  • IntervalTree(tree) (Interval objects are same as those in tree)
  • tree.copy() (Interval objects are shallow copies of those in tree)
  • set(tree) (can later be fed into IntervalTree())
  • list(tree) (ditto)
• Pickle-friendly
• Automatic AVL balancing

Examples

• Getting started

  >>> from intervaltree import Interval, IntervalTree
  >>> t = IntervalTree()
  >>> t
  IntervalTree()

• Adding intervals - any object works!

  >>> t[1:2] = "1-2"
  >>> t[4:7] = (4, 7)
  >>> t[5:9] = {5: 9}

• Query by point

  The result of a query is a set object, so if ordering is important,

  you must sort it first.

  >>> sorted(t[6])
  [Interval(4, 7, (4, 7)), Interval(5, 9, {5: 9})]
  >>> sorted(t[6])[0]
  Interval(4, 7, (4, 7))

• Query by range

  Note that ranges are inclusive of the lower limit, but non-inclusive of the upper limit. So:

  >>> sorted(t[2:4])
  []

  But:

  >>> sorted(t[1:5])
  [Interval(1, 2, '1-2'), Interval(4, 7, (4, 7))]

• Accessing an Interval object

  >>> iv = Interval(4, 7, (4, 7))
  >>> iv.begin
  4
  >>> iv.end
  7
  >>> iv.data
  (4, 7)
  
  >>> begin, end, data = iv
  >>> begin
  4
  >>> end
  7
  >>> data
  (4, 7)

• Constructing from lists of intervals

  We could have made a similar tree this way:

  >>> ivs = [(1, 2), (4, 7), (5, 9)]
  >>> t = IntervalTree(
  ...    Interval(begin, end, "%d-%d" % (begin, end)) for begin, end in ivs
  ... )

  Or, if we don't need the data fields:

  >>> t2 = IntervalTree(Interval(*iv) for iv in ivs)

• Removing intervals

  >>> t.remove( Interval(1, 2, "1-2") )
  >>> sorted(t)
  [Interval(4, 7, '4-7'), Interval(5, 9, '5-9')]
  
  >>> t.remove( Interval(500, 1000, "Doesn't exist"))  # raises ValueError
  Traceback (most recent call last):
  ValueError
  
  >>> t.discard(Interval(500, 1000, "Doesn't exist"))  # quietly does nothing
  
  >>> del t[5]  # same as t.remove_overlap(5)
  >>> t
  IntervalTree()

  We could also empty a tree entirely:

  >>> t2.clear()
  >>> t2
  IntervalTree()

  Or remove intervals that overlap a range:

  >>> t = IntervalTree([
  ...     Interval(0, 10),
  ...     Interval(10, 20),
  ...     Interval(20, 30),
  ...     Interval(30, 40)])
  >>> t.remove_overlap(25, 35)
  >>> sorted(t)
  [Interval(0, 10), Interval(10, 20)]

  We can also remove only those intervals completely enveloped in a range:

  >>> t.remove_envelop(5, 20)
  >>> sorted(t)
  [Interval(0, 10)]

• Chopping

  We could also chop out parts of the tree:

  >>> t = IntervalTree([Interval(0, 10)])
  >>> t.chop(3, 7)
  >>> sorted(t)
  [Interval(0, 3), Interval(7, 10)]

  To modify the new intervals' data fields based on which side of the interval is being chopped:

  >>> def datafunc(iv, islower):
  ...     oldlimit = iv[islower]
  ...     return "oldlimit: {0}, islower: {1}".format(oldlimit, islower)
  >>> t = IntervalTree([Interval(0, 10)])
  >>> t.chop(3, 7, datafunc)
  >>> sorted(t)[0]
  Interval(0, 3, 'oldlimit: 10, islower: True')
  >>> sorted(t)[1]
  Interval(7, 10, 'oldlimit: 0, islower: False')

• Slicing

  You can also slice intervals in the tree without removing them:

  >>> t = IntervalTree([Interval(0, 10), Interval(5, 15)])
  >>> t.slice(3)
  >>> sorted(t)
  [Interval(0, 3), Interval(3, 10), Interval(5, 15)]

  You can also set the data fields, for example, re-using datafunc() from above:

  >>> t = IntervalTree([Interval(5, 15)])
  >>> t.slice(10, datafunc)
  >>> sorted(t)[0]
  Interval(5, 10, 'oldlimit: 15, islower: True')
  >>> sorted(t)[1]
  Interval(10, 15, 'oldlimit: 5, islower: False')

Future improvements

See the issue tracker on GitHub.

Based on

• Eternally Confuzzled's AVL tree
• Wikipedia's Interval Tree
• Heavily modified from Tyler Kahn's Interval Tree implementation in Python (GitHub project)
• Incorporates contributions from:
  • konstantint/Konstantin Tretyakov of the University of Tartu (Estonia)
  • siniG/Avi Gabay
  • lmcarril/Luis M. Carril of the Karlsruhe Institute for Technology (Germany)

Copyright

• Chaim-Leib Halbert, 2013-2015
• Modifications, Konstantin Tretyakov, 2014

Licensed under the Apache License, version 2.0.

The source code for this project is at https://github.com/chaimleib/intervaltree

Change log

Version 2.1.0

• Added:
  • merge_overlaps() method and tests
  • merge_equals() method and tests
  • range() method
  • span() method, for returning the difference between end() and begin()
• Fixes:
  • Development version numbering is changing to be compliant with PEP440. Version numbering now contains major, minor and micro release numbers, plus the number of builds following the stable release version, e.g. 2.0.4b34
  • Speed improvement: begin() and end() methods used iterative min() and max() builtins instead of the more efficient iloc member available to SortedDict
  • overlaps() method used to return True even if provided null test interval
• Maintainers:
  • Added coverage test (make coverage) with html report (htmlcov/index.html)
  • Tests run slightly faster

Version 2.0.4

• Fix: Issue #27: README incorrectly showed using a comma instead of a colon when querying the IntervalTree: it showed tree[begin, end] instead of tree[begin:end]

Version 2.0.3

• Fix: README showed using + operator for setlike union instead of the correct | operator
• Removed tests from release package to speed up installation; to get the tests, download from GitHub

Version 2.0.2

• Fix: Issue #20: performance enhancement for large trees. IntervalTree.search() made a copy of the entire boundary_table resulting in linear search time. The sortedcollections package is now the sole install dependency

Version 2.0.1

• Fix: Issue #26: failed to prune empty Node after a rotation promoted contents of s_center

Version 2.0.0

• IntervalTree now supports the full collections.MutableSet API
• Added:
  • __delitem__ to IntervalTree
  • Interval comparison methods lt(), gt(), le() and ge() to Interval, as an alternative to the comparison operators, which are designed for sorting
  • IntervalTree.from_tuples(iterable)
  • IntervalTree.clear()
  • IntervalTree.difference(iterable)
  • IntervalTree.difference_update(iterable)
  • IntervalTree.union(iterable)
  • IntervalTree.intersection(iterable)
  • IntervalTree.intersection_update(iterable)
  • IntervalTree.symmetric_difference(iterable)
  • IntervalTree.symmetric_difference_update(iterable)
  • IntervalTree.chop(a, b)
  • IntervalTree.slice(point)
• Deprecated IntervalTree.extend() -- use update() instead
• Internal improvements:
  • More verbose tests with progress bars
  • More tests for comparison and sorting behavior
  • Code in the README is included in the unit tests
• Fixes
  • BACKWARD INCOMPATIBLE: On ranged queries where begin >= end, the query operated on the overlaps of begin. This behavior was documented as expected in 1.x; it is now changed to be more consistent with the definition of Intervals, which are half-open.
  • Issue #25: pruning empty Nodes with staggered descendants could result in invalid trees
  • Sorting Intervals and numbers in the same list gathered all the numbers at the beginning and the Intervals at the end
  • IntervalTree.overlaps() and friends returned None instead of False
  • Maintainers: make install-testpypi failed because the pip was missing a --pre flag

Version 1.1.1

• Removed requirement for pyandoc in order to run functionality tests.

Version 1.1.0

• Added ability to use Interval.distance_to() with points, not just Intervals
• Added documentation on return types to IntervalTree and Interval
• Interval.__cmp__() works with points too
• Fix: IntervalTree.score() returned maximum score of 0.5 instead of 1.0. Now returns max of subscores instead of avg
• Internal improvements:
  • Development version numbering scheme, based on git describe the "building towards" release is appended after a hyphen, eg. 1.0.2-37-g2da2ef0-1.10. The previous tagged release is 1.0.2, and there have been 37 commits since then, current tag is g2da2ef0, and we are getting ready for a 1.1.0 release
  • Optimality tests added
  • Interval overlap tests for ranges, Intervals and points added

Version 1.0.2

-Bug fixes:

- Node.depth_score_helper() raised AttributeError

- README formatting

Version 1.0.1

• Fix: pip install failure because of failure to generate README.rst

Version 1.0.0

• Renamed from PyIntervalTree to intervaltree
• Speed improvements for adding and removing Intervals (~70% faster than 0.4)
• Bug fixes:
  • BACKWARD INCOMPATIBLE: len() of an Interval is always 3, reverting to default behavior for namedtuples. In Python 3, len returning a non-integer raises an exception. Instead, use Interval.length(), which returns 0 for null intervals and end - begin otherwise. Also, if the len() === 0, then not iv is True.
  • When inserting an Interval via __setitem__ and improper parameters given, all errors were transformed to IndexError
  • split_overlaps did not update the boundary_table counts
• Internal improvements:
  • More robust local testing tools
  • Long series of interdependent tests have been separated into sections

Version 0.4

• Faster balancing (~80% faster)
• Bug fixes:
  • Double rotations were performed in place of a single rotation when presented an unbalanced Node with a balanced child.
  • During single rotation, kept referencing an unrotated Node instead of the new, rotated one

Version 0.3.3

• Made IntervalTree crash if inited with a null Interval (end <= begin)
• IntervalTree raises ValueError instead of AssertionError when a null Interval is inserted

Version 0.3.2

• Support for Python 3.2+ and 2.6+
• Changed license from LGPL to more permissive Apache license
• Merged changes from https://github.com/konstantint/PyIntervalTree to https://github.com/chaimleib/PyIntervalTree
  • Interval now inherits from a namedtuple. Benefits: should be faster. Drawbacks: slight behavioural change (Intervals not mutable anymore).
  • Added float tests
  • Use setup.py for tests
  • Automatic testing via travis-ci
  • Removed dependency on six
• Interval improvements:
  • Intervals without data have a cleaner string representation
  • Intervals without data are pickled more compactly
  • Better hashing
  • Intervals are ordered by begin, then end, then by data. If data is not orderable, sorts by type(data)
• Bug fixes:
  • Fixed crash when querying empty tree
  • Fixed missing close parenthesis in examples
  • Made IntervalTree crash earlier if a null Interval is added
• Internals:
  • New test directory
  • Nicer display of data structures for debugging, using custom test/pprint.py (Python 2.6, 2.7)
  • More sensitive exception handling
  • Local script to test in all supported versions of Python
  • Added IntervalTree.score() to measure how optimally a tree is structured

Version 0.2.3

• Slight changes for inclusion in PyPI.
• Some documentation changes
• Added tests
• Bug fix: interval addition via [] was broken in Python 2.7 (see http://bugs.python.org/issue21785)
• Added intervaltree.bio subpackage, adding some utilities for use in bioinformatics

Version 0.2.2b

• Forked from https://github.com/MusashiAharon/PyIntervalTree