You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

268 lines
9.2 KiB
Python

# -*- coding: utf-8 -*-
# Natural Language Toolkit: Transformation-based learning
#
# Copyright (C) 2001-2020 NLTK Project
# Author: Marcus Uneson <marcus.uneson@gmail.com>
# based on previous (nltk2) version by
# Christopher Maloof, Edward Loper, Steven Bird
# URL: <http://nltk.org/>
# For license information, see LICENSE.TXT
from abc import ABCMeta, abstractmethod
class Feature(metaclass=ABCMeta):
"""
An abstract base class for Features. A Feature is a combination of
a specific property-computing method and a list of relative positions
to apply that method to.
The property-computing method, M{extract_property(tokens, index)},
must be implemented by every subclass. It extracts or computes a specific
property for the token at the current index. Typical extract_property()
methods return features such as the token text or tag; but more involved
methods may consider the entire sequence M{tokens} and
for instance compute the length of the sentence the token belongs to.
In addition, the subclass may have a PROPERTY_NAME, which is how
it will be printed (in Rules and Templates, etc). If not given, defaults
to the classname.
"""
json_tag = "nltk.tbl.Feature"
PROPERTY_NAME = None
def __init__(self, positions, end=None):
"""
Construct a Feature which may apply at C{positions}.
#For instance, importing some concrete subclasses (Feature is abstract)
>>> from nltk.tag.brill import Word, Pos
#Feature Word, applying at one of [-2, -1]
>>> Word([-2,-1])
Word([-2, -1])
#Positions need not be contiguous
>>> Word([-2,-1, 1])
Word([-2, -1, 1])
#Contiguous ranges can alternatively be specified giving the
#two endpoints (inclusive)
>>> Pos(-3, -1)
Pos([-3, -2, -1])
#In two-arg form, start <= end is enforced
>>> Pos(2, 1)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "nltk/tbl/template.py", line 306, in __init__
raise TypeError
ValueError: illegal interval specification: (start=2, end=1)
:type positions: list of int
:param positions: the positions at which this features should apply
:raises ValueError: illegal position specifications
An alternative calling convention, for contiguous positions only,
is Feature(start, end):
:type start: int
:param start: start of range where this feature should apply
:type end: int
:param end: end of range (NOTE: inclusive!) where this feature should apply
"""
self.positions = None # to avoid warnings
if end is None:
self.positions = tuple(sorted(set(int(i) for i in positions)))
else: # positions was actually not a list, but only the start index
try:
if positions > end:
raise TypeError
self.positions = tuple(range(positions, end + 1))
except TypeError:
# let any kind of erroneous spec raise ValueError
raise ValueError(
"illegal interval specification: (start={0}, end={1})".format(
positions, end
)
)
# set property name given in subclass, or otherwise name of subclass
self.PROPERTY_NAME = self.__class__.PROPERTY_NAME or self.__class__.__name__
def encode_json_obj(self):
return self.positions
@classmethod
def decode_json_obj(cls, obj):
positions = obj
return cls(positions)
def __repr__(self):
return "%s(%r)" % (self.__class__.__name__, list(self.positions))
@classmethod
def expand(cls, starts, winlens, excludezero=False):
"""
Return a list of features, one for each start point in starts
and for each window length in winlen. If excludezero is True,
no Features containing 0 in its positions will be generated
(many tbl trainers have a special representation for the
target feature at [0])
For instance, importing a concrete subclass (Feature is abstract)
>>> from nltk.tag.brill import Word
First argument gives the possible start positions, second the
possible window lengths
>>> Word.expand([-3,-2,-1], [1])
[Word([-3]), Word([-2]), Word([-1])]
>>> Word.expand([-2,-1], [1])
[Word([-2]), Word([-1])]
>>> Word.expand([-3,-2,-1], [1,2])
[Word([-3]), Word([-2]), Word([-1]), Word([-3, -2]), Word([-2, -1])]
>>> Word.expand([-2,-1], [1])
[Word([-2]), Word([-1])]
a third optional argument excludes all Features whose positions contain zero
>>> Word.expand([-2,-1,0], [1,2], excludezero=False)
[Word([-2]), Word([-1]), Word([0]), Word([-2, -1]), Word([-1, 0])]
>>> Word.expand([-2,-1,0], [1,2], excludezero=True)
[Word([-2]), Word([-1]), Word([-2, -1])]
All window lengths must be positive
>>> Word.expand([-2,-1], [0])
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "nltk/tag/tbl/template.py", line 371, in expand
:param starts: where to start looking for Feature
ValueError: non-positive window length in [0]
:param starts: where to start looking for Feature
:type starts: list of ints
:param winlens: window lengths where to look for Feature
:type starts: list of ints
:param excludezero: do not output any Feature with 0 in any of its positions.
:type excludezero: bool
:returns: list of Features
:raises ValueError: for non-positive window lengths
"""
if not all(x > 0 for x in winlens):
raise ValueError("non-positive window length in {0}".format(winlens))
xs = (starts[i : i + w] for w in winlens for i in range(len(starts) - w + 1))
return [cls(x) for x in xs if not (excludezero and 0 in x)]
def issuperset(self, other):
"""
Return True if this Feature always returns True when other does
More precisely, return True if this feature refers to the same property as other;
and this Feature looks at all positions that other does (and possibly
other positions in addition).
#For instance, importing a concrete subclass (Feature is abstract)
>>> from nltk.tag.brill import Word, Pos
>>> Word([-3,-2,-1]).issuperset(Word([-3,-2]))
True
>>> Word([-3,-2,-1]).issuperset(Word([-3,-2, 0]))
False
#Feature subclasses must agree
>>> Word([-3,-2,-1]).issuperset(Pos([-3,-2]))
False
:param other: feature with which to compare
:type other: (subclass of) Feature
:return: True if this feature is superset, otherwise False
:rtype: bool
"""
return self.__class__ is other.__class__ and set(self.positions) >= set(
other.positions
)
def intersects(self, other):
"""
Return True if the positions of this Feature intersects with those of other
More precisely, return True if this feature refers to the same property as other;
and there is some overlap in the positions they look at.
#For instance, importing a concrete subclass (Feature is abstract)
>>> from nltk.tag.brill import Word, Pos
>>> Word([-3,-2,-1]).intersects(Word([-3,-2]))
True
>>> Word([-3,-2,-1]).intersects(Word([-3,-2, 0]))
True
>>> Word([-3,-2,-1]).intersects(Word([0]))
False
#Feature subclasses must agree
>>> Word([-3,-2,-1]).intersects(Pos([-3,-2]))
False
:param other: feature with which to compare
:type other: (subclass of) Feature
:return: True if feature classes agree and there is some overlap in the positions they look at
:rtype: bool
"""
return bool(
(
self.__class__ is other.__class__
and set(self.positions) & set(other.positions)
)
)
# Rich comparisons for Features. With @functools.total_ordering (Python 2.7+),
# it will be enough to define __lt__ and __eq__
def __eq__(self, other):
return self.__class__ is other.__class__ and self.positions == other.positions
def __lt__(self, other):
return (
self.__class__.__name__ < other.__class__.__name__
or
# self.positions is a sorted tuple of ints
self.positions < other.positions
)
def __ne__(self, other):
return not (self == other)
def __gt__(self, other):
return other < self
def __ge__(self, other):
return not self < other
def __le__(self, other):
return self < other or self == other
@staticmethod
@abstractmethod
def extract_property(tokens, index):
"""
Any subclass of Feature must define static method extract_property(tokens, index)
:param tokens: the sequence of tokens
:type tokens: list of tokens
:param index: the current index
:type index: int
:return: feature value
:rtype: any (but usually scalar)
"""