spaCy/spacy/tokens.pyx

288 lines
9.3 KiB
Cython

# cython: embedsignature=True
from cython.view cimport array as cvarray
from preshed.maps cimport PreshMap
from preshed.counter cimport PreshCounter
from .vocab cimport EMPTY_LEXEME
from .typedefs cimport attr_id_t, attr_t
from .typedefs cimport LEMMA
from .typedefs cimport ID, SIC, DENSE, SHAPE, PREFIX, SUFFIX, LENGTH, CLUSTER, POS_TYPE
from .typedefs cimport POS, LEMMA
cimport cython
DEF PADDING = 5
cdef int bounds_check(int i, int length, int padding) except -1:
if (i + padding) < 0:
raise IndexError
if (i - padding) >= length:
raise IndexError
cdef attr_t get_token_attr(const TokenC* token, attr_id_t feat_name) nogil:
if feat_name == LEMMA:
return token.lemma
elif feat_name == POS:
return token.pos
else:
return get_lex_attr(token.lex, feat_name)
cdef attr_t get_lex_attr(const Lexeme* lex, attr_id_t feat_name) nogil:
if feat_name < (sizeof(flags_t) * 8):
return check_flag(lex, feat_name)
elif feat_name == ID:
return lex.id
elif feat_name == SIC:
return lex.sic
elif feat_name == DENSE:
return lex.dense
elif feat_name == SHAPE:
return lex.shape
elif feat_name == PREFIX:
return lex.prefix
elif feat_name == SUFFIX:
return lex.suffix
elif feat_name == LENGTH:
return lex.length
elif feat_name == CLUSTER:
return lex.cluster
elif feat_name == POS_TYPE:
return lex.pos_type
else:
return 0
cdef class Tokens:
"""Access and set annotations onto some text.
"""
def __init__(self, Vocab vocab, string_length=0):
self.vocab = vocab
if string_length >= 3:
size = int(string_length / 3.0)
else:
size = 5
self.mem = Pool()
# Guarantee self.lex[i-x], for any i >= 0 and x < padding is in bounds
# However, we need to remember the true starting places, so that we can
# realloc.
data_start = <TokenC*>self.mem.alloc(size + (PADDING*2), sizeof(TokenC))
cdef int i
for i in range(size + (PADDING*2)):
data_start[i].lex = &EMPTY_LEXEME
self.data = data_start + PADDING
self.max_length = size
self.length = 0
def __getitem__(self, i):
"""Retrieve a token.
Returns:
token (Token):
"""
bounds_check(i, self.length, PADDING)
return Token(self, i)
def __iter__(self):
"""Iterate over the tokens.
Yields:
token (Token):
"""
for i in range(self.length):
yield self[i]
def __len__(self):
return self.length
cdef int push_back(self, int idx, LexemeOrToken lex_or_tok) except -1:
if self.length == self.max_length:
self._realloc(self.length * 2)
cdef TokenC* t = &self.data[self.length]
if LexemeOrToken is TokenC_ptr:
t[0] = lex_or_tok[0]
else:
t.lex = lex_or_tok
t.idx = idx
self.length += 1
return idx + t.lex.length
@cython.boundscheck(False)
cpdef long[:,:] to_array(self, object attr_ids):
"""Given a list of M attribute IDs, export the tokens to a numpy ndarray
of shape N*M, where N is the length of the sentence.
Arguments:
attr_ids (list[int]): A list of attribute ID ints.
Returns:
feat_array (numpy.ndarray[long, ndim=2]): A feature matrix, with one
row per word, and one column per attribute indicated in the input
attr_ids.
"""
cdef int i, j
cdef attr_id_t feature
cdef long[:,:] output = cvarray(shape=(self.length, len(attr_ids)),
itemsize=sizeof(long), format="l")
for i in range(self.length):
for j, feature in enumerate(attr_ids):
output[i, j] = get_token_attr(&self.data[i], feature)
return output
def count_by(self, attr_id_t attr_id):
"""Produce a dict of {attribute (int): count (ints)} frequencies, keyed
by the values of the given attribute ID.
>>> from spacy.en import English, attrs
>>> nlp = English()
>>> tokens = nlp(u'apple apple orange banana')
>>> tokens.count_by(attrs.SIC)
{12800L: 1, 11880L: 2, 7561L: 1}
>>> tokens.to_array([attrs.SIC])
array([[11880],
[11880],
[ 7561],
[12800]])
"""
cdef int i
cdef attr_t attr
cdef size_t count
cdef PreshCounter counts = PreshCounter(2 ** 8)
for i in range(self.length):
attr = get_token_attr(&self.data[i], attr_id)
counts.inc(attr, 1)
return dict(counts)
def _realloc(self, new_size):
self.max_length = new_size
n = new_size + (PADDING * 2)
# What we're storing is a "padded" array. We've jumped forward PADDING
# places, and are storing the pointer to that. This way, we can access
# words out-of-bounds, and get out-of-bounds markers.
# Now that we want to realloc, we need the address of the true start,
# so we jump the pointer back PADDING places.
cdef TokenC* data_start = self.data - PADDING
data_start = <TokenC*>self.mem.realloc(data_start, n * sizeof(TokenC))
self.data = data_start + PADDING
cdef int i
for i in range(self.length, self.max_length + PADDING):
self.data[i].lex = &EMPTY_LEXEME
@cython.freelist(64)
cdef class Token:
"""An individual token.
Internally, the Token is a tuple (i, tokens) --- it delegates to the Tokens
object.
"""
def __init__(self, Tokens tokens, int i):
self._seq = tokens
self.i = i
def __unicode__(self):
cdef const TokenC* t = &self._seq.data[self.i]
cdef int end_idx = t.idx + t.lex.length
if self.i + 1 == self._seq.length:
return self.string
if end_idx == t[1].idx:
return self.string
else:
return self.string + ' '
def __len__(self):
"""The number of unicode code-points in the original string.
Returns:
length (int):
"""
return self._seq.data[self.i].lex.length
property idx:
"""The index into the original string at which the token starts.
The following is supposed to always be true:
>>> original_string[token.idx:token.idx len(token) == token.string
"""
def __get__(self):
return self._seq.data[self.i].idx
property cluster:
"""The Brown cluster ID of the word: en.wikipedia.org/wiki/Brown_clustering
Similar words have better-than-chance likelihood of having similar cluster
IDs, although the clustering is quite noisy. Cluster IDs make good features,
and help to make models slightly more robust to domain variation.
A common trick is to use only the first N bits of a cluster ID in a feature,
as the more general part of the hierarchical clustering is often more accurate
than the lower categories.
To assist in this, I encode the cluster IDs little-endian, to allow a simple
bit-mask:
>>> six_bits = cluster & (2**6 - 1)
"""
def __get__(self):
return self._seq.data[self.i].lex.cluster
property string:
"""The unicode string of the word, with no whitespace padding."""
def __get__(self):
cdef const TokenC* t = &self._seq.data[self.i]
if t.lex.sic == 0:
return ''
cdef bytes utf8string = self._seq.vocab.strings[t.lex.sic]
return utf8string.decode('utf8')
property lemma:
"""The unicode string of the word's lemma. If no part-of-speech tag is
assigned, the most common part-of-speech tag of the word is used.
"""
def __get__(self):
cdef const TokenC* t = &self._seq.data[self.i]
if t.lemma == 0:
return self.string
cdef bytes utf8string = self._seq.vocab.strings[t.lemma]
return utf8string.decode('utf8')
property dep_tag:
"""The ID integer of the word's dependency label. If no parse has been
assigned, defaults to 0.
"""
def __get__(self):
return self._seq.data[self.i].dep_tag
property pos:
"""The ID integer of the word's part-of-speech tag, from the 13-tag
Google Universal Tag Set. Constants for this tag set are available in
spacy.typedefs.
"""
def __get__(self):
return self._seq.data[self.i].pos
property fine_pos:
"""The ID integer of the word's fine-grained part-of-speech tag, as assigned
by the tagger model. Fine-grained tags include morphological information,
and other distinctions, and allow a more accurate tagger to be trained.
"""
def __get__(self):
return self._seq.data[self.i].fine_pos
property sic:
def __get__(self):
return self._seq.data[self.i].lex.sic
property head:
"""The token predicted by the parser to be the head of the current token."""
def __get__(self):
cdef const TokenC* t = &self._seq.data[self.i]
return Token(self._seq, self.i + t.head)