2022-04-01 10:56:26 +03:00
|
|
|
from typing import Iterable, Tuple, Union, Optional, TYPE_CHECKING
|
2021-01-14 09:30:41 +03:00
|
|
|
import weakref
|
|
|
|
import struct
|
2022-04-01 10:56:26 +03:00
|
|
|
from copy import deepcopy
|
2021-01-14 09:30:41 +03:00
|
|
|
import srsly
|
2021-08-20 12:06:19 +03:00
|
|
|
|
|
|
|
from spacy.errors import Errors
|
2021-01-14 09:30:41 +03:00
|
|
|
from .span cimport Span
|
|
|
|
|
|
|
|
|
|
|
|
cdef class SpanGroup:
|
|
|
|
"""A group of spans that all belong to the same Doc object. The group
|
|
|
|
can be named, and you can attach additional attributes to it. Span groups
|
|
|
|
are generally accessed via the `doc.spans` attribute. The `doc.spans`
|
|
|
|
attribute will convert lists of spans into a `SpanGroup` object for you
|
|
|
|
automatically on assignment.
|
|
|
|
|
|
|
|
Example:
|
|
|
|
Construction 1
|
|
|
|
>>> doc = nlp("Their goi ng home")
|
|
|
|
>>> doc.spans["errors"] = SpanGroup(
|
|
|
|
doc,
|
|
|
|
name="errors",
|
2022-04-14 10:59:48 +03:00
|
|
|
spans=[doc[0:1], doc[1:3]],
|
2021-01-14 09:30:41 +03:00
|
|
|
attrs={"annotator": "matt"}
|
|
|
|
)
|
|
|
|
|
|
|
|
Construction 2
|
|
|
|
>>> doc = nlp("Their goi ng home")
|
2022-04-14 10:59:48 +03:00
|
|
|
>>> doc.spans["errors"] = [doc[0:1], doc[1:3]]
|
2021-01-14 09:30:41 +03:00
|
|
|
>>> assert isinstance(doc.spans["errors"], SpanGroup)
|
|
|
|
|
2021-01-30 12:09:38 +03:00
|
|
|
DOCS: https://spacy.io/api/spangroup
|
2021-01-14 09:30:41 +03:00
|
|
|
"""
|
|
|
|
def __init__(self, doc, *, name="", attrs={}, spans=[]):
|
|
|
|
"""Create a SpanGroup.
|
|
|
|
|
|
|
|
doc (Doc): The reference Doc object.
|
|
|
|
name (str): The group name.
|
|
|
|
attrs (Dict[str, Any]): Optional JSON-serializable attributes to attach.
|
|
|
|
spans (Iterable[Span]): The spans to add to the group.
|
|
|
|
|
2021-01-30 12:09:38 +03:00
|
|
|
DOCS: https://spacy.io/api/spangroup#init
|
2021-01-14 09:30:41 +03:00
|
|
|
"""
|
|
|
|
# We need to make this a weak reference, so that the Doc object can
|
|
|
|
# own the SpanGroup without circular references. We do want to get
|
|
|
|
# the Doc though, because otherwise the API gets annoying.
|
|
|
|
self._doc_ref = weakref.ref(doc)
|
|
|
|
self.name = name
|
|
|
|
self.attrs = dict(attrs) if attrs is not None else {}
|
|
|
|
cdef Span span
|
2022-04-01 10:56:26 +03:00
|
|
|
if len(spans) :
|
|
|
|
self.c.reserve(len(spans))
|
2021-01-14 09:30:41 +03:00
|
|
|
for span in spans:
|
|
|
|
self.push_back(span.c)
|
|
|
|
|
|
|
|
def __repr__(self):
|
|
|
|
return str(list(self))
|
|
|
|
|
|
|
|
@property
|
|
|
|
def doc(self):
|
|
|
|
"""RETURNS (Doc): The reference document.
|
|
|
|
|
2021-01-30 12:09:38 +03:00
|
|
|
DOCS: https://spacy.io/api/spangroup#doc
|
2021-01-14 09:30:41 +03:00
|
|
|
"""
|
2021-08-20 12:06:19 +03:00
|
|
|
doc = self._doc_ref()
|
|
|
|
if doc is None:
|
|
|
|
# referent has been garbage collected
|
2021-09-27 10:10:45 +03:00
|
|
|
raise RuntimeError(Errors.E865)
|
2021-08-20 12:06:19 +03:00
|
|
|
return doc
|
2021-01-14 09:30:41 +03:00
|
|
|
|
|
|
|
@property
|
|
|
|
def has_overlap(self):
|
|
|
|
"""RETURNS (bool): Whether the group contains overlapping spans.
|
|
|
|
|
2021-01-30 12:09:38 +03:00
|
|
|
DOCS: https://spacy.io/api/spangroup#has_overlap
|
2021-01-14 09:30:41 +03:00
|
|
|
"""
|
|
|
|
if not len(self):
|
|
|
|
return False
|
|
|
|
sorted_spans = list(sorted(self))
|
|
|
|
last_end = sorted_spans[0].end
|
|
|
|
for span in sorted_spans[1:]:
|
|
|
|
if span.start < last_end:
|
|
|
|
return True
|
|
|
|
last_end = span.end
|
|
|
|
return False
|
|
|
|
|
|
|
|
def __len__(self):
|
|
|
|
"""RETURNS (int): The number of spans in the group.
|
|
|
|
|
2021-01-30 12:09:38 +03:00
|
|
|
DOCS: https://spacy.io/api/spangroup#len
|
2021-01-14 09:30:41 +03:00
|
|
|
"""
|
|
|
|
return self.c.size()
|
|
|
|
|
2022-04-01 10:56:26 +03:00
|
|
|
def __getitem__(self, int i) -> Span:
|
|
|
|
"""Get a span from the group. Note that a copy of the span is returned,
|
|
|
|
so if any changes are made to this span, they are not reflected in the
|
|
|
|
corresponding member of the span group.
|
|
|
|
|
|
|
|
i (int): The item index.
|
|
|
|
RETURNS (Span): The span at the given index.
|
|
|
|
|
|
|
|
DOCS: https://spacy.io/api/spangroup#getitem
|
|
|
|
"""
|
|
|
|
i = self._normalize_index(i)
|
|
|
|
return Span.cinit(self.doc, self.c[i])
|
|
|
|
|
|
|
|
def __delitem__(self, int i):
|
|
|
|
"""Delete a span from the span group at index i.
|
|
|
|
|
|
|
|
i (int): The item index.
|
|
|
|
|
|
|
|
DOCS: https://spacy.io/api/spangroup#delitem
|
|
|
|
"""
|
|
|
|
i = self._normalize_index(i)
|
|
|
|
self.c.erase(self.c.begin() + i - 1)
|
|
|
|
|
|
|
|
def __setitem__(self, int i, Span span):
|
|
|
|
"""Set a span in the span group.
|
|
|
|
|
|
|
|
i (int): The item index.
|
|
|
|
span (Span): The span.
|
|
|
|
|
|
|
|
DOCS: https://spacy.io/api/spangroup#setitem
|
|
|
|
"""
|
|
|
|
if span.doc is not self.doc:
|
|
|
|
raise ValueError(Errors.E855.format(obj="span"))
|
|
|
|
|
|
|
|
i = self._normalize_index(i)
|
|
|
|
self.c[i] = span.c
|
|
|
|
|
|
|
|
def __iadd__(self, other: Union[SpanGroup, Iterable["Span"]]) -> SpanGroup:
|
|
|
|
"""Operator +=. Append a span group or spans to this group and return
|
|
|
|
the current span group.
|
|
|
|
|
|
|
|
other (Union[SpanGroup, Iterable["Span"]]): The SpanGroup or spans to
|
|
|
|
add.
|
|
|
|
|
|
|
|
RETURNS (SpanGroup): The current span group.
|
|
|
|
|
|
|
|
DOCS: https://spacy.io/api/spangroup#iadd
|
|
|
|
"""
|
|
|
|
return self._concat(other, inplace=True)
|
|
|
|
|
|
|
|
def __add__(self, other: SpanGroup) -> SpanGroup:
|
|
|
|
"""Operator +. Concatenate a span group with this group and return a
|
|
|
|
new span group.
|
|
|
|
|
|
|
|
other (SpanGroup): The SpanGroup to add.
|
|
|
|
|
|
|
|
RETURNS (SpanGroup): The concatenated SpanGroup.
|
|
|
|
|
|
|
|
DOCS: https://spacy.io/api/spangroup#add
|
|
|
|
"""
|
|
|
|
# For Cython 0.x and __add__, you cannot rely on `self` as being `self`
|
|
|
|
# or being the right type, so both types need to be checked explicitly.
|
|
|
|
if isinstance(self, SpanGroup) and isinstance(other, SpanGroup):
|
|
|
|
return self._concat(other)
|
|
|
|
return NotImplemented
|
|
|
|
|
2021-01-14 09:30:41 +03:00
|
|
|
def append(self, Span span):
|
|
|
|
"""Add a span to the group. The span must refer to the same Doc
|
|
|
|
object as the span group.
|
|
|
|
|
|
|
|
span (Span): The span to append.
|
|
|
|
|
2021-01-30 12:09:38 +03:00
|
|
|
DOCS: https://spacy.io/api/spangroup#append
|
2021-01-14 09:30:41 +03:00
|
|
|
"""
|
|
|
|
if span.doc is not self.doc:
|
2022-04-01 10:56:26 +03:00
|
|
|
raise ValueError(Errors.E855.format(obj="span"))
|
2021-01-14 09:30:41 +03:00
|
|
|
self.push_back(span.c)
|
|
|
|
|
2022-04-01 10:56:26 +03:00
|
|
|
def extend(self, spans_or_span_group: Union[SpanGroup, Iterable["Span"]]):
|
|
|
|
"""Add multiple spans or contents of another SpanGroup to the group.
|
|
|
|
All spans must refer to the same Doc object as the span group.
|
2021-01-14 09:30:41 +03:00
|
|
|
|
2022-04-01 10:56:26 +03:00
|
|
|
spans (Union[SpanGroup, Iterable["Span"]]): The spans to add.
|
2021-01-14 09:30:41 +03:00
|
|
|
|
2021-01-30 12:09:38 +03:00
|
|
|
DOCS: https://spacy.io/api/spangroup#extend
|
2021-01-14 09:30:41 +03:00
|
|
|
"""
|
2022-04-01 10:56:26 +03:00
|
|
|
self._concat(spans_or_span_group, inplace=True)
|
2021-01-14 09:30:41 +03:00
|
|
|
|
|
|
|
def to_bytes(self):
|
|
|
|
"""Serialize the SpanGroup's contents to a byte string.
|
|
|
|
|
|
|
|
RETURNS (bytes): The serialized span group.
|
|
|
|
|
2021-01-30 12:09:38 +03:00
|
|
|
DOCS: https://spacy.io/api/spangroup#to_bytes
|
2021-01-14 09:30:41 +03:00
|
|
|
"""
|
|
|
|
output = {"name": self.name, "attrs": self.attrs, "spans": []}
|
2022-04-01 10:56:26 +03:00
|
|
|
cdef int i
|
2021-01-14 09:30:41 +03:00
|
|
|
for i in range(self.c.size()):
|
|
|
|
span = self.c[i]
|
|
|
|
# The struct.pack here is probably overkill, but it might help if
|
|
|
|
# you're saving tonnes of spans, and it doesn't really add any
|
|
|
|
# complexity. We do take care to specify little-endian byte order
|
|
|
|
# though, to ensure the message can be loaded back on a different
|
|
|
|
# arch.
|
|
|
|
# Q: uint64_t
|
|
|
|
# q: int64_t
|
|
|
|
# L: uint32_t
|
|
|
|
# l: int32_t
|
|
|
|
output["spans"].append(struct.pack(
|
|
|
|
">QQQllll",
|
|
|
|
span.id,
|
|
|
|
span.kb_id,
|
|
|
|
span.label,
|
|
|
|
span.start,
|
|
|
|
span.end,
|
|
|
|
span.start_char,
|
|
|
|
span.end_char
|
|
|
|
))
|
|
|
|
return srsly.msgpack_dumps(output)
|
|
|
|
|
|
|
|
def from_bytes(self, bytes_data):
|
|
|
|
"""Deserialize the SpanGroup's contents from a byte string.
|
|
|
|
|
|
|
|
bytes_data (bytes): The span group to load.
|
|
|
|
RETURNS (SpanGroup): The deserialized span group.
|
|
|
|
|
2021-01-30 12:09:38 +03:00
|
|
|
DOCS: https://spacy.io/api/spangroup#from_bytes
|
2021-01-14 09:30:41 +03:00
|
|
|
"""
|
|
|
|
msg = srsly.msgpack_loads(bytes_data)
|
|
|
|
self.name = msg["name"]
|
|
|
|
self.attrs = dict(msg["attrs"])
|
|
|
|
self.c.clear()
|
|
|
|
self.c.reserve(len(msg["spans"]))
|
|
|
|
cdef SpanC span
|
|
|
|
for span_data in msg["spans"]:
|
|
|
|
items = struct.unpack(">QQQllll", span_data)
|
|
|
|
span.id = items[0]
|
|
|
|
span.kb_id = items[1]
|
|
|
|
span.label = items[2]
|
|
|
|
span.start = items[3]
|
|
|
|
span.end = items[4]
|
|
|
|
span.start_char = items[5]
|
|
|
|
span.end_char = items[6]
|
|
|
|
self.c.push_back(span)
|
|
|
|
return self
|
|
|
|
|
|
|
|
cdef void push_back(self, SpanC span) nogil:
|
|
|
|
self.c.push_back(span)
|
2022-04-01 10:56:26 +03:00
|
|
|
|
|
|
|
def copy(self) -> SpanGroup:
|
|
|
|
"""Clones the span group.
|
|
|
|
|
|
|
|
RETURNS (SpanGroup): A copy of the span group.
|
|
|
|
|
|
|
|
DOCS: https://spacy.io/api/spangroup#copy
|
|
|
|
"""
|
|
|
|
return SpanGroup(
|
|
|
|
self.doc,
|
|
|
|
name=self.name,
|
|
|
|
attrs=deepcopy(self.attrs),
|
|
|
|
spans=list(self),
|
|
|
|
)
|
|
|
|
|
|
|
|
def _concat(
|
|
|
|
self,
|
|
|
|
other: Union[SpanGroup, Iterable["Span"]],
|
|
|
|
*,
|
|
|
|
inplace: bool = False,
|
|
|
|
) -> SpanGroup:
|
|
|
|
"""Concatenates the current span group with the provided span group or
|
|
|
|
spans, either in place or creating a copy. Preserves the name of self,
|
|
|
|
updates attrs only with values that are not in self.
|
|
|
|
|
|
|
|
other (Union[SpanGroup, Iterable[Span]]): The spans to append.
|
|
|
|
inplace (bool): Indicates whether the operation should be performed in
|
|
|
|
place on the current span group.
|
|
|
|
|
|
|
|
RETURNS (SpanGroup): Either a new SpanGroup or the current SpanGroup
|
|
|
|
depending on the value of inplace.
|
|
|
|
"""
|
|
|
|
cdef SpanGroup span_group = self if inplace else self.copy()
|
|
|
|
cdef SpanGroup other_group
|
|
|
|
cdef Span span
|
|
|
|
|
|
|
|
if isinstance(other, SpanGroup):
|
|
|
|
other_group = other
|
|
|
|
if other_group.doc is not self.doc:
|
|
|
|
raise ValueError(Errors.E855.format(obj="span group"))
|
|
|
|
|
|
|
|
other_attrs = deepcopy(other_group.attrs)
|
|
|
|
span_group.attrs.update({
|
|
|
|
key: value for key, value in other_attrs.items() \
|
|
|
|
if key not in span_group.attrs
|
|
|
|
})
|
|
|
|
if len(other_group):
|
|
|
|
span_group.c.reserve(span_group.c.size() + other_group.c.size())
|
|
|
|
span_group.c.insert(span_group.c.end(), other_group.c.begin(), other_group.c.end())
|
|
|
|
else:
|
|
|
|
if len(other):
|
|
|
|
span_group.c.reserve(self.c.size() + len(other))
|
|
|
|
for span in other:
|
|
|
|
if span.doc is not self.doc:
|
|
|
|
raise ValueError(Errors.E855.format(obj="span"))
|
|
|
|
span_group.c.push_back(span.c)
|
|
|
|
|
|
|
|
return span_group
|
|
|
|
|
|
|
|
def _normalize_index(self, int i) -> int:
|
|
|
|
"""Checks list index boundaries and adjusts the index if negative.
|
|
|
|
|
|
|
|
i (int): The index.
|
|
|
|
RETURNS (int): The adjusted index.
|
|
|
|
"""
|
|
|
|
cdef int length = self.c.size()
|
|
|
|
if i < -length or i >= length:
|
|
|
|
raise IndexError(Errors.E856.format(i=i, length=length))
|
|
|
|
if i < 0:
|
|
|
|
i += length
|
|
|
|
return i
|