span_field.py

from typing import Dict, Tuple

from overrides import overrides
import torch

from allennlp.data.fields.field import Field
from allennlp.data.fields.sequence_field import SequenceField


class SpanField(Field[torch.Tensor]):
    """
    A `SpanField` is a pair of inclusive, zero-indexed (start, end) indices into a
    :class:`~allennlp.data.fields.sequence_field.SequenceField`, used to represent a span of text.
    Because it's a pair of indices into a :class:`SequenceField`, we take one of those as input
    to make the span's dependence explicit and to validate that the span is well defined.

    # Parameters

    span_start : `int`, required.
        The index of the start of the span in the :class:`SequenceField`.
    span_end : `int`, required.
        The inclusive index of the end of the span in the :class:`SequenceField`.
    sequence_field : `SequenceField`, required.
        A field containing the sequence that this `SpanField` is a span inside.
    """

    __slots__ = ["span_start", "span_end", "sequence_field"]

    def __init__(self, span_start: int, span_end: int, sequence_field: SequenceField) -> None:
        self.span_start = span_start
        self.span_end = span_end
        self.sequence_field = sequence_field

        if not isinstance(span_start, int) or not isinstance(span_end, int):
            raise TypeError(
                f"SpanFields must be passed integer indices. Found span indices: "
                f"({span_start}, {span_end}) with types "
                f"({type(span_start)} {type(span_end)})"
            )
        if span_start > span_end:
            raise ValueError(
                f"span_start must be less than span_end, " f"but found ({span_start}, {span_end})."
            )

        if span_end > self.sequence_field.sequence_length() - 1:
            raise ValueError(
                f"span_end must be <= len(sequence_length) - 1, but found "
                f"{span_end} and {self.sequence_field.sequence_length() - 1} respectively."
            )

    @overrides
    def get_padding_lengths(self) -> Dict[str, int]:

        return {}

    @overrides
    def as_tensor(self, padding_lengths: Dict[str, int]) -> torch.Tensor:

        tensor = torch.LongTensor([self.span_start, self.span_end])
        return tensor

    @overrides
    def empty_field(self):
        return SpanField(-1, -1, self.sequence_field.empty_field())

    def __str__(self) -> str:
        return f"SpanField with spans: ({self.span_start}, {self.span_end})."

    def __eq__(self, other) -> bool:
        if isinstance(other, tuple) and len(other) == 2:
            return other == (self.span_start, self.span_end)
        return super().__eq__(other)

    def __len__(self):
        return 2

    @overrides
    def human_readable_repr(self) -> Tuple[int, int]:
        return self.span_start, self.span_end