spaCy/spacy/ml/models/span_predictor.py

from typing import List, Tuple
import torch

from thinc.api import Model, chain, tuplify
from thinc.api import PyTorchWrapper, ArgsKwargs
from thinc.types import Floats2d, Ints1d
from thinc.util import xp2torch, torch2xp

from ...tokens import Doc
from ...util import registry
from .coref_util import get_sentence_ids


@registry.architectures("spacy.SpanPredictor.v1")
def build_span_predictor(
    tok2vec: Model[List[Doc], List[Floats2d]],
    hidden_size: int = 1024,
    dist_emb_size: int = 64,
):
    # TODO fix this
    try:
        dim = tok2vec.get_dim("nO")
    except ValueError:
        # happens with transformer listener
        dim = 768

    with Model.define_operators({">>": chain, "&": tuplify}):
        span_predictor = PyTorchWrapper(
            SpanPredictor(dim, hidden_size, dist_emb_size),
            convert_inputs=convert_span_predictor_inputs,
        )
        # TODO use proper parameter for prefix
        head_info = build_get_head_metadata("coref_head_clusters")
        model = (tok2vec & head_info) >> span_predictor

    return model


def convert_span_predictor_inputs(
    model: Model, X: Tuple[Ints1d, Floats2d, Ints1d], is_train: bool
):
    tok2vec, (sent_ids, head_ids) = X
    # Normally we should use the input is_train, but for these two it's not relevant

    def backprop(args: ArgsKwargs) -> List[Floats2d]:
        gradients = torch2xp(args.args[1])
        return [[gradients], None]

    word_features = xp2torch(tok2vec[0], requires_grad=is_train)
    sent_ids = xp2torch(sent_ids[0], requires_grad=False)
    if not head_ids[0].size:
        head_ids = torch.empty(size=(0,))
    else:
        head_ids = xp2torch(head_ids[0], requires_grad=False)

    argskwargs = ArgsKwargs(args=(sent_ids, word_features, head_ids), kwargs={})
    return argskwargs, backprop


# TODO This probably belongs in the component, not the model.
def predict_span_clusters(
    span_predictor: Model, sent_ids: Ints1d, words: Floats2d, clusters: List[Ints1d]
):
    """
    Predicts span clusters based on the word clusters.

    span_predictor: a SpanPredictor instance
    sent_ids: For each word indicates, which sentence it appears in.
    words: Features for words.
    clusters: Clusters inferred by the CorefScorer.

    Returns:
        List[List[Tuple[int, int]]: span clusters
    """
    if not clusters:
        return []

    xp = span_predictor.ops.xp
    heads_ids = xp.asarray(sorted(i for cluster in clusters for i in cluster))
    scores = span_predictor.predict((sent_ids, words, heads_ids))
    starts = scores[:, :, 0].argmax(axis=1).tolist()
    ends = (scores[:, :, 1].argmax(axis=1) + 1).tolist()

    head2span = {
        head: (start, end) for head, start, end in zip(heads_ids.tolist(), starts, ends)
    }

    return [[head2span[head] for head in cluster] for cluster in clusters]


def build_get_head_metadata(prefix):
    # TODO this name is awful, fix it
    model = Model(
        "HeadDataProvider", attrs={"prefix": prefix}, forward=head_data_forward
    )
    return model


def head_data_forward(model, docs, is_train):
    """A layer to generate the extra data needed for the span predictor."""
    sent_ids = []
    head_ids = []
    prefix = model.attrs["prefix"]
    for doc in docs:
        sids = model.ops.asarray2i(get_sentence_ids(doc))
        sent_ids.append(sids)
        heads = []
        for key, sg in doc.spans.items():
            if not key.startswith(prefix):
                continue
            for span in sg:
                # TODO warn if spans are more than one token
                heads.append(span[0].i)
        heads = model.ops.asarray2i(heads)
        head_ids.append(heads)
    # each of these is a list with one entry per doc
    # backprop is just a placeholder
    # TODO it would probably be better to have a list of tuples than two lists of arrays
    return (sent_ids, head_ids), lambda x: []


# TODO this should maybe have a different name from the component
class SpanPredictor(torch.nn.Module):
    def __init__(self, input_size: int, hidden_size: int, dist_emb_size: int):
        super().__init__()
        # input size = single token size
        # 64 = probably distance emb size
        # TODO check that dist_emb_size use is correct
        self.ffnn = torch.nn.Sequential(
            torch.nn.Linear(input_size * 2 + dist_emb_size, hidden_size),
            torch.nn.ReLU(),
            torch.nn.Dropout(0.3),
            # TODO seems weird the 256 isn't a parameter???
            torch.nn.Linear(hidden_size, 256),
            torch.nn.ReLU(),
            torch.nn.Dropout(0.3),
            # this use of dist_emb_size looks wrong but it was 64...?
            torch.nn.Linear(256, dist_emb_size),
        )
        # TODO make the Convs also parametrizeable
        self.conv = torch.nn.Sequential(
            torch.nn.Conv1d(64, 4, 3, 1, 1), torch.nn.Conv1d(4, 2, 3, 1, 1)
        )
        # TODO make embeddings size a parameter
        self.emb = torch.nn.Embedding(128, dist_emb_size)  # [-63, 63] + too_far

    def forward(
        self,
        sent_id,
        words: torch.Tensor,
        heads_ids: torch.Tensor,
    ) -> torch.Tensor:
        """
        Calculates span start/end scores of words for each span
        for each head.

        sent_id: Sentence id of each word.
        words: features for each word in the document.
        heads_ids: word indices of span heads

        Returns:
            torch.Tensor: span start/end scores, (n_heads x n_words x 2)
        """
        # If we don't receive heads, return empty
        if heads_ids.nelement() == 0:
            return torch.empty(size=(0,))
        # Obtain distance embedding indices, [n_heads, n_words]
        relative_positions = heads_ids.unsqueeze(1) - torch.arange(
            words.shape[0]
        ).unsqueeze(0)
        # make all valid distances positive
        emb_ids = relative_positions + 63
        # "too_far"
        emb_ids[(emb_ids < 0) + (emb_ids > 126)] = 127
        # Obtain "same sentence" boolean mask: (n_heads x n_words)
        heads_ids = heads_ids.long()
        same_sent = sent_id[heads_ids].unsqueeze(1) == sent_id.unsqueeze(0)
        # To save memory, only pass candidates from one sentence for each head
        # pair_matrix contains concatenated span_head_emb + candidate_emb + distance_emb
        # for each candidate among the words in the same sentence as span_head
        # (n_heads x input_size * 2 x distance_emb_size)
        rows, cols = same_sent.nonzero(as_tuple=True)
        pair_matrix = torch.cat(
            (
                words[heads_ids[rows]],
                words[cols],
                self.emb(emb_ids[rows, cols]),
            ),
            dim=1,
        )
        lengths = same_sent.sum(dim=1)
        padding_mask = torch.arange(0, lengths.max().item()).unsqueeze(0)
        # (n_heads x max_sent_len)
        padding_mask = padding_mask < lengths.unsqueeze(1)
        # (n_heads x max_sent_len x input_size * 2 + distance_emb_size)
        # This is necessary to allow the convolution layer to look at several
        # word scores
        padded_pairs = torch.zeros(*padding_mask.shape, pair_matrix.shape[-1])
        padded_pairs[padding_mask] = pair_matrix
        res = self.ffnn(padded_pairs)  # (n_heads x n_candidates x last_layer_output)
        res = self.conv(res.permute(0, 2, 1)).permute(
            0, 2, 1
        )  # (n_heads x n_candidates, 2)

        scores = torch.full((heads_ids.shape[0], words.shape[0], 2), float("-inf"))
        scores[rows, cols] = res[padding_mask]
        # Make sure that start <= head <= end during inference
        if not self.training:
            valid_starts = torch.log((relative_positions >= 0).to(torch.float))
            valid_ends = torch.log((relative_positions <= 0).to(torch.float))
            valid_positions = torch.stack((valid_starts, valid_ends), dim=2)
            return scores + valid_positions
        return scores