The portion
library (formerly distributed as python-intervals
) provides data structure and operations for intervals in Python 3.6+.
Latest release:
portion
: 2.1.6 on 2021-04-17 (documentation, changes).
python-intervals
: 1.10.0 on 2019-09-26 (documentation, changes).
Note that python-intervals
will no longer receive updates since it has been replaced by portion
.
You can use pip
to install it, as usual: pip install portion
. This will install the latest available version from PyPI.
Pre-releases are available from the master branch on GitHub
and can be installed with pip install git+https://github.com/AlexandreDecan/portion
.
Note that portion
is also available on conda-forge.
You can install portion
and its development environment using pip install -e .[test]
at the root of this repository. This automatically installs pytest (for the test suites) and black (for code formatting).
Assuming this library is imported using import portion as P
, intervals can be easily
created using one of the following helpers:
>>> P.open(1, 2)
(1,2)
>>> P.closed(1, 2)
[1,2]
>>> P.openclosed(1, 2)
(1,2]
>>> P.closedopen(1, 2)
[1,2)
>>> P.singleton(1)
[1]
>>> P.empty()
()
The bounds of an interval can be any arbitrary values, as long as they are comparable:
>>> P.closed(1.2, 2.4)
[1.2,2.4]
>>> P.closed('a', 'z')
['a','z']
>>> import datetime
>>> P.closed(datetime.date(2011, 3, 15), datetime.date(2013, 10, 10))
[datetime.date(2011, 3, 15),datetime.date(2013, 10, 10)]
Infinite and semi-infinite intervals are supported using P.inf
and -P.inf
as upper or lower bounds.
These two objects support comparison with any other object.
When infinities are used as a lower or upper bound, the corresponding boundary is automatically converted to an open one.
>>> P.inf > 'a', P.inf > 0, P.inf > True
(True, True, True)
>>> P.openclosed(-P.inf, 0)
(-inf,0]
>>> P.closed(-P.inf, P.inf) # Automatically converted to an open interval
(-inf,+inf)
Empty intervals always resolve to (P.inf, -P.inf)
, regardless of the provided bounds:
>>> P.empty() == P.open(P.inf, -P.inf)
True
>>> P.closed(4, 3) == P.open(P.inf, -P.inf)
True
>>> P.openclosed('a', 'a') == P.open(P.inf, -P.inf)
True
Intervals created with this library are Interval
instances.
An Interval
instance is a disjunction of atomic intervals each representing a single interval (e.g. [1,2]
).
Intervals can be iterated to access the underlying atomic intervals, sorted by their lower and upper bounds.
>>> list(P.open(10, 11) | P.closed(0, 1) | P.closed(20, 21))
[[0,1], (10,11), [20,21]]
Nested intervals can also be retrieved with a position or a slice:
>>> (P.open(10, 11) | P.closed(0, 1) | P.closed(20, 21))[0]
[0,1]
>>> (P.open(10, 11) | P.closed(0, 1) | P.closed(20, 21))[-2]
(10,11)
>>> (P.open(10, 11) | P.closed(0, 1) | P.closed(20, 21))[:2]
[0,1] | (10,11)
For convenience, intervals are automatically simplified:
>>> P.closed(0, 2) | P.closed(2, 4)
[0,4]
>>> P.closed(1, 2) | P.closed(3, 4) | P.closed(2, 3)
[1,4]
>>> P.empty() | P.closed(0, 1)
[0,1]
>>> P.closed(1, 2) | P.closed(2, 3) | P.closed(4, 5)
[1,3] | [4,5]
Note that simplification of discrete intervals is not supported by portion
(but it can be simulated though, see #24).
For example, combining [0,1]
with [2,3]
will not result in [0,3]
even if there is
no integer between 1
and 2
.
An Interval
defines the following properties:
i.empty
is True
if and only if the interval is empty.
>>> P.closed(0, 1).empty
False
>>> P.closed(0, 0).empty
False
>>> P.openclosed(0, 0).empty
True
>>> P.empty().empty
True
i.atomic
is True
if and only if the interval is a disjunction of a single (possibly empty) interval.
>>> P.closed(0, 2).atomic
True
>>> (P.closed(0, 1) | P.closed(1, 2)).atomic
True
>>> (P.closed(0, 1) | P.closed(2, 3)).atomic
False
i.enclosure
refers to the smallest atomic interval that includes the current one.
>>> (P.closed(0, 1) | P.open(2, 3)).enclosure
[0,3)
The left and right boundaries, and the lower and upper bounds of an interval can be respectively accessed
with its left
, right
, lower
and upper
attributes.
The left
and right
bounds are either P.CLOSED
or P.OPEN
.
By definition, P.CLOSED == ~P.OPEN
and vice-versa.
>> P.CLOSED, P.OPEN
CLOSED, OPEN
>>> x = P.closedopen(0, 1)
>>> x.left, x.lower, x.upper, x.right
(CLOSED, 0, 1, OPEN)
If the interval is not atomic, then left
and lower
refer to the lower bound of its enclosure,
while right
and upper
refer to the upper bound of its enclosure:
>>> x = P.open(0, 1) | P.closed(3, 4)
>>> x.left, x.lower, x.upper, x.right
(OPEN, 0, 4, CLOSED)
One can easily check for some interval properties based on the bounds of an interval:
>>> x = P.openclosed(-P.inf, 0)
>>> # Check that interval is left/right closed
>>> x.left == P.CLOSED, x.right == P.CLOSED
(False, True)
>>> # Check that interval is left/right bounded
>>> x.lower == -P.inf, x.upper == P.inf
(True, False)
>>> # Check for singleton
>>> x.lower == x.upper
False
Interval
instances support the following operations:
i.intersection(other)
and i & other
return the intersection of two intervals.
>>> P.closed(0, 2) & P.closed(1, 3)
[1,2]
>>> P.closed(0, 4) & P.open(2, 3)
(2,3)
>>> P.closed(0, 2) & P.closed(2, 3)
[2]
>>> P.closed(0, 2) & P.closed(3, 4)
()
i.union(other)
and i | other
return the union of two intervals.
>>> P.closed(0, 1) | P.closed(1, 2)
[0,2]
>>> P.closed(0, 1) | P.closed(2, 3)
[0,1] | [2,3]
i.complement(other)
and ~i
return the complement of the interval.
>>> ~P.closed(0, 1)
(-inf,0) | (1,+inf)
>>> ~(P.open(-P.inf, 0) | P.open(1, P.inf))
[0,1]
>>> ~P.open(-P.inf, P.inf)
()
i.difference(other)
and i - other
return the difference between i
and other
.
>>> P.closed(0,2) - P.closed(1,2)
[0,1)
>>> P.closed(0, 4) - P.closed(1, 2)
[0,1) | (2,4]
i.contains(other)
and other in i
hold if given item is contained in the interval.
It supports intervals and arbitrary comparable values.
>>> 2 in P.closed(0, 2)
True
>>> 2 in P.open(0, 2)
False
>>> P.open(0, 1) in P.closed(0, 2)
True
i.adjacent(other)
tests if the two intervals are adjacent, i.e., if they do not overlap and their union form a single atomic interval.
While this definition corresponds to the usual notion of adjacency for atomic
intervals, it has stronger requirements for non-atomic ones since it requires
all underlying atomic intervals to be adjacent (i.e. that one
interval fills the gaps between the atomic intervals of the other one).
>>> P.closed(0, 1).adjacent(P.openclosed(1, 2))
True
>>> P.closed(0, 1).adjacent(P.closed(1, 2))
False
>>> (P.closed(0, 1) | P.closed(2, 3)).adjacent(P.open(1, 2) | P.open(3, 4))
True
>>> (P.closed(0, 1) | P.closed(2, 3)).adjacent(P.open(3, 4))
False
>>> P.closed(0, 1).adjacent(P.open(1, 2) | P.open(3, 4))
False
i.overlaps(other)
tests if there is an overlap between two intervals.
>>> P.closed(1, 2).overlaps(P.closed(2, 3))
True
>>> P.closed(1, 2).overlaps(P.open(2, 3))
False
Finally, intervals are hashable as long as their bounds are hashable (and we have defined a hash value for P.inf
and -P.inf
).
Equality between intervals can be checked with the classical ==
operator:
>>> P.closed(0, 2) == P.closed(0, 1) | P.closed(1, 2)
True
>>> P.closed(0, 2) == P.open(0, 2)
False
Moreover, intervals are comparable using >
, >=
, <
or <=
.
These comparison operators have a different behaviour than the usual ones.
For instance, a < b
holds if a
is entirely on the left of the lower bound of b
and a > b
holds if a
is entirely
on the right of the upper bound of b
.
>>> P.closed(0, 1) < P.closed(2, 3)
True
>>> P.closed(0, 1) < P.closed(1, 2)
False
Similarly, a <= b
holds if a
is entirely on the left of the upper bound of b
, and a >= b
holds if a
is entirely on the right of the lower bound of b
.
>>> P.closed(0, 1) <= P.closed(2, 3)
True
>>> P.closed(0, 2) <= P.closed(1, 3)
True
>>> P.closed(0, 3) <= P.closed(1, 2)
False
Intervals can also be compared with single values. If i
is an interval and x
a value, then
x < i
holds if x
is on the left of the lower bound of i
and x <= i
holds if x
is on the
left of the upper bound of i
.
>>> 5 < P.closed(0, 10)
False
>>> 5 <= P.closed(0, 10)
True
>>> P.closed(0, 10) < 5
False
>>> P.closed(0, 10) <= 5
True
This behaviour is similar to the one that could be obtained by first converting x
to a
singleton interval (except for infinities since they resolve to empty intervals).
Note that all these semantics differ from classical comparison operators. As a consequence, some intervals are never comparable in the classical sense, as illustrated hereafter:
>>> P.closed(0, 4) <= P.closed(1, 2) or P.closed(0, 4) >= P.closed(1, 2)
False
>>> P.closed(0, 4) < P.closed(1, 2) or P.closed(0, 4) > P.closed(1, 2)
False
>>> P.empty() < P.empty()
True
>>> P.empty() < P.closed(0, 1) and P.empty() > P.closed(0, 1)
True
Intervals are immutable but provide a replace
method to create a new interval based on the
current one. This method accepts four optional parameters left
, lower
, upper
, and right
:
>>> i = P.closed(0, 2)
>>> i.replace(P.OPEN, -1, 3, P.CLOSED)
(-1,3]
>>> i.replace(lower=1, right=P.OPEN)
[1,2)
Functions can be passed instead of values. If a function is passed, it is called with the current corresponding value.
>>> P.closed(0, 2).replace(upper=lambda x: 2 * x)
[0,4]
The provided function won't be called on infinities, unless ignore_inf
is set to False
.
>>> i = P.closedopen(0, P.inf)
>>> i.replace(upper=lambda x: 10) # No change, infinity is ignored
[0,+inf)
>>> i.replace(upper=lambda x: 10, ignore_inf=False) # Infinity is not ignored
[0,10)
When replace
is applied on an interval that is not atomic, it is extended and/or restricted such that
its enclosure satisfies the new bounds.
>>> i = P.openclosed(0, 1) | P.closed(5, 10)
>>> i.replace(P.CLOSED, -1, 8, P.OPEN)
[-1,1] | [5,8)
>>> i.replace(lower=4)
(4,10]
To apply arbitrary transformations on the underlying atomic intervals, intervals expose an apply
method that acts like map
.
This method accepts a function that will be applied on each of the underlying atomic intervals to perform the desired transformation.
The provided function is expected to return either an Interval
, or a 4-uple (left, lower, upper, right)
.
>>> i = P.closed(2, 3) | P.open(4, 5)
>>> # Increment bound values
>>> i.apply(lambda x: (x.left, x.lower + 1, x.upper + 1, x.right))
[3,4] | (5,6)
>>> # Invert bounds
>>> i.apply(lambda x: (~x.left, x.lower, x.upper, ~x.right))
(2,3) | [4,5]
The apply
method is very powerful when used in combination with replace
.
Because the latter allows functions to be passed as parameters and ignores infinities by default, it can be
conveniently used to transform (disjunction of) intervals in presence of infinities.
>>> i = P.openclosed(-P.inf, 0) | P.closed(3, 4) | P.closedopen(8, P.inf)
>>> # Increment bound values
>>> i.apply(lambda x: x.replace(upper=lambda v: v + 1))
(-inf,1] | [3,5] | [8,+inf)
>>> # Intervals are still automatically simplified
>>> i.apply(lambda x: x.replace(lower=lambda v: v * 2))
(-inf,0] | [16,+inf)
>>> # Invert bounds
>>> i.apply(lambda x: x.replace(left=lambda v: ~v, right=lambda v: ~v))
(-inf,0) | (3,4) | (8,+inf)
>>> # Replace infinities with -10 and 10
>>> conv = lambda v: -10 if v == -P.inf else (10 if v == P.inf else v)
>>> i.apply(lambda x: x.replace(lower=conv, upper=conv, ignore_inf=False))
(-10,0] | [3,4] | [8,10)
The iterate
function takes an interval, and returns a generator to iterate over
the values of an interval. Obviously, as intervals are continuous, it is required to specify the
step
between consecutive values. The iteration then starts from the lower bound and ends on the upper one. Only values contained by the interval are returned this way.
>>> list(P.iterate(P.closed(0, 3), step=1))
[0, 1, 2, 3]
>>> list(P.iterate(P.closed(0, 3), step=2))
[0, 2]
>>> list(P.iterate(P.open(0, 3), step=2))
[2]
When an interval is not atomic, iterate
consecutively iterates on all underlying atomic
intervals, starting from each lower bound and ending on each upper one:
>>> list(P.iterate(P.singleton(0) | P.singleton(3) | P.singleton(5), step=2)) # Won't be [0]
[0, 3, 5]
>>> list(P.iterate(P.closed(0, 2) | P.closed(4, 6), step=3)) # Won't be [0, 6]
[0, 4]
By default, the iteration always starts on the lower bound of each underlying atomic interval.
The base
parameter can be used to change this behaviour, by specifying how the initial value to start
the iteration from must be computed. This parameter accepts a callable that is called with the lower
bound of each underlying atomic interval, and that returns the initial value to start the iteration from.
It can be helpful to deal with (semi-)infinite intervals, or to align the generated values of the iterator:
>>> # Align on integers
>>> list(P.iterate(P.closed(0.3, 4.9), step=1, base=int))
[1, 2, 3, 4]
>>> # Restrict values of a (semi-)infinite interval
>>> list(P.iterate(P.openclosed(-P.inf, 2), step=1, base=lambda x: max(0, x)))
[0, 1, 2]
The base
parameter can be used to change how iterate
applies on unions of atomic interval, by
specifying a function that returns a single value, as illustrated next:
>>> base = lambda x: 0
>>> list(P.iterate(P.singleton(0) | P.singleton(3) | P.singleton(5), step=2, base=base))
[0]
>>> list(P.iterate(P.closed(0, 2) | P.closed(4, 6), step=3, base=base))
[0, 6]
Notice that defining base
such that it returns a single value can be extremely inefficient in
terms of performance when the intervals are "far apart" each other (i.e., when the gaps between
atomic intervals are large).
Finally, iteration can be performed in reverse order by specifying reverse=True
.
>>> list(P.iterate(P.closed(0, 3), step=-1, reverse=True)) # Mind step=-1
[3, 2, 1, 0]
>>> list(P.iterate(P.closed(0, 3), step=-2, reverse=True)) # Mind step=-2
[3, 1]
Again, this library does not make any assumption about the objects being used in an interval, as long as they
are comparable. However, it is not always possible to provide a meaningful value for step
(e.g., what would
be the step between two consecutive characters?). In these cases, a callable can be passed instead of a value.
This callable will be called with the current value, and is expected to return the next possible value.
>>> list(P.iterate(P.closed('a', 'd'), step=lambda d: chr(ord(d) + 1)))
['a', 'b', 'c', 'd']
>>> # Since we reversed the order, we changed plus to minus in step.
>>> list(P.iterate(P.closed('a', 'd'), step=lambda d: chr(ord(d) - 1), reverse=True))
['d', 'c', 'b', 'a']
The library provides an IntervalDict
class, a dict
-like data structure to store and query data
along with intervals. Any value can be stored in such data structure as long as it supports
equality.
>>> d = P.IntervalDict()
>>> d[P.closed(0, 3)] = 'banana'
>>> d[4] = 'apple'
>>> d
{[0,3]: 'banana', [4]: 'apple'}
When a value is defined for an interval that overlaps an existing one, it is automatically updated to take the new value into account:
>>> d[P.closed(2, 4)] = 'orange'
>>> d
{[0,2): 'banana', [2,4]: 'orange'}
An IntervalDict
can be queried using single values or intervals. If a single value is used as a
key, its behaviour corresponds to the one of a classical dict
:
>>> d[2]
'orange'
>>> d[5] # Key does not exist
Traceback (most recent call last):
...
KeyError: 5
>>> d.get(5, default=0)
0
When the key is an interval, a new IntervalDict
containing the values
for the specified key is returned:
>>> d[~P.empty()] # Get all values, similar to d.copy()
{[0,2): 'banana', [2,4]: 'orange'}
>>> d[P.closed(1, 3)]
{[1,2): 'banana', [2,3]: 'orange'}
>>> d[P.closed(-2, 1)]
{[0,1]: 'banana'}
>>> d[P.closed(-2, -1)]
{}
By using .get
, a default value (defaulting to None
) can be specified.
This value is used to "fill the gaps" if the queried interval is not completely
covered by the IntervalDict
:
>>> d.get(P.closed(-2, 1), default='peach')
{[-2,0): 'peach', [0,1]: 'banana'}
>>> d.get(P.closed(-2, -1), default='peach')
{[-2,-1]: 'peach'}
>>> d.get(P.singleton(1), default='peach') # Key is covered, default is not used
{[1]: 'banana'}
For convenience, an IntervalDict
provides a way to look for specific data values.
The .find
method always returns a (possibly empty) Interval
instance for which given
value is defined:
>>> d.find('banana')
[0,2)
>>> d.find('orange')
[2,4]
>>> d.find('carrot')
()
The active domain of an IntervalDict
can be retrieved with its .domain
method.
This method always returns a single Interval
instance, where .keys
returns a sorted view of disjoint intervals.
>>> d.domain()
[0,4]
>>> list(d.keys())
[[0,2), [2,4]]
>>> list(d.values())
['banana', 'orange']
>>> list(d.items())
[([0,2), 'banana'), ([2,4], 'orange')]
The .keys
, .values
and .items
methods return exactly one element for each stored value (i.e., if two intervals share a value, they are merged into a disjunction), as illustrated next.
See #44 to know how to obtained a sorted list of atomic intervals instead.
>>> d = P.IntervalDict()
>>> d[P.closed(0, 1)] = d[P.closed(2, 3)] = 'peach'
>>> list(d.items())
[([0,1] | [2,3], 'peach')]
Two IntervalDict
instances can be combined together using the .combine
method.
This method returns a new IntervalDict
whose keys and values are taken from the two
source IntervalDict
. Values corresponding to non-intersecting keys are simply copied,
while values corresponding to intersecting keys are combined together using the provided
function, as illustrated hereafter:
>>> d1 = P.IntervalDict({P.closed(0, 2): 'banana'})
>>> d2 = P.IntervalDict({P.closed(1, 3): 'orange'})
>>> concat = lambda x, y: x + '/' + y
>>> d1.combine(d2, how=concat)
{[0,1): 'banana', [1,2]: 'banana/orange', (2,3]: 'orange'}
Resulting keys always correspond to an outer join. Other joins can be easily simulated
by querying the resulting IntervalDict
as follows:
>>> d = d1.combine(d2, how=concat)
>>> d[d1.domain()] # Left join
{[0,1): 'banana', [1,2]: 'banana/orange'}
>>> d[d2.domain()] # Right join
{[1,2]: 'banana/orange', (2,3]: 'orange'}
>>> d[d1.domain() & d2.domain()] # Inner join
{[1,2]: 'banana/orange'}
Finally, similarly to a dict
, an IntervalDict
also supports len
, in
and del
, and defines
.clear
, .copy
, .update
, .pop
, .popitem
, and .setdefault
.
For convenience, one can export the content of an IntervalDict
to a classical Python dict
using
the as_dict
method.
Intervals can be exported to string, either using repr
(as illustrated above) or with the to_string
function.
>>> P.to_string(P.closedopen(0, 1))
'[0,1)'
The way string representations are built can be easily parametrized using the various parameters supported by
to_string
:
>>> params = {
... 'disj': ' or ',
... 'sep': ' - ',
... 'left_closed': '<',
... 'right_closed': '>',
... 'left_open': '..',
... 'right_open': '..',
... 'pinf': '+oo',
... 'ninf': '-oo',
... 'conv': lambda v: '"{}"'.format(v),
... }
>>> x = P.openclosed(0, 1) | P.closed(2, P.inf)
>>> P.to_string(x, **params)
'.."0" - "1"> or <"2" - +oo..'
Similarly, intervals can be created from a string using the from_string
function.
A conversion function (conv
parameter) has to be provided to convert a bound (as string) to a value.
>>> P.from_string('[0, 1]', conv=int) == P.closed(0, 1)
True
>>> P.from_string('[1.2]', conv=float) == P.singleton(1.2)
True
>>> converter = lambda s: datetime.datetime.strptime(s, '%Y/%m/%d')
>>> P.from_string('[2011/03/15, 2013/10/10]', conv=converter)
[datetime.datetime(2011, 3, 15, 0, 0),datetime.datetime(2013, 10, 10, 0, 0)]
Similarly to to_string
, function from_string
can be parametrized to deal with more elaborated inputs.
Notice that as from_string
expects regular expression patterns, we need to escape some characters.
>>> s = '.."0" - "1"> or <"2" - +oo..'
>>> params = {
... 'disj': ' or ',
... 'sep': ' - ',
... 'left_closed': '<',
... 'right_closed': '>',
... 'left_open': r'\.\.', # from_string expects regular expression patterns
... 'right_open': r'\.\.', # from_string expects regular expression patterns
... 'pinf': r'\+oo', # from_string expects regular expression patterns
... 'ninf': '-oo',
... 'conv': lambda v: int(v[1:-1]),
... }
>>> P.from_string(s, **params)
(0,1] | [2,+inf)
When a bound contains a comma or has a representation that cannot be automatically parsed with from_string
,
the bound
parameter can be used to specify the regular expression that should be used to match its representation.
>>> s = '[(0, 1), (2, 3)]' # Bounds are expected to be tuples
>>> P.from_string(s, conv=eval, bound=r'\(.+?\)')
[(0, 1),(2, 3)]
Intervals can also be exported to a list of 4-uples with to_data
, e.g., to support JSON serialization.
P.CLOSED
and P.OPEN
are represented by Boolean values True
(inclusive) and False
(exclusive).
>>> P.to_data(P.openclosed(0, 2))
[(False, 0, 2, True)]
The values used to represent positive and negative infinities can be specified with
pinf
and ninf
. They default to float('inf')
and float('-inf')
respectively.
>>> x = P.openclosed(0, 1) | P.closedopen(2, P.inf)
>>> P.to_data(x)
[(False, 0, 1, True), (True, 2, inf, False)]
The function to convert bounds can be specified with the conv
parameter.
>>> x = P.closedopen(datetime.date(2011, 3, 15), datetime.date(2013, 10, 10))
>>> P.to_data(x, conv=lambda v: (v.year, v.month, v.day))
[(True, (2011, 3, 15), (2013, 10, 10), False)]
Intervals can be imported from such a list of 4-tuples with from_data
.
The same set of parameters can be used to specify how bounds and infinities are converted.
>>> x = [(True, (2011, 3, 15), (2013, 10, 10), False)]
>>> P.from_data(x, conv=lambda v: datetime.date(*v))
[datetime.date(2011, 3, 15),datetime.date(2013, 10, 10))
This library adheres to a semantic versioning scheme. See CHANGELOG.md for the list of changes.
Contributions are very welcome! Feel free to report bugs or suggest new features using GitHub issues and/or pull requests.
Distributed under LGPLv3 - GNU Lesser General Public License, version 3.
You can refer to this library using:
@software{portion,
author = {Decan, Alexandre},
title = {portion: Python data structure and operations for intervals},
url = {https://github.com/AlexandreDecan/portion},
}